Introduction

Acute myeloid leukemia (AML) is a clinically and biologically heterogeneous disease characterized cytogenetically by recurrent abnormalities, which provide powerful prognostic information.1 Cytogenetically normal AML (CN-AML) is the largest subgroup, representing ~45% of adult patients with AML.2 In the last years, emerging data have indicated that somatic mutations are associated with treatment outcome and serve as a basis for molecularly guided risk assessment and treatment stratification.3, 4, 5, 6, 7

Mutations in the Wilms tumor 1 gene (WT1) occur in ~10% of adults with CN-AML8 and are potential markers in AML. Although these mutations are predicted to lead to loss of function of WT1 with several studies reporting a worse outcome,9, 10, 11, 12, 13, 14, 15 other authors found that these mutations have no prognostic impact.16, 17, 18

The WT1 gene, located on chromosome band 11p13, encodes a zinc-finger transcription factor, which has emerged as an important regulator of normal and malignant hematopoiesis. Originally, WT1 was identified as a tumor-suppressor gene isolated in Wilms tumor,19 a pediatric kidney malignancy. However, accumulating data revealed that WT1 appears to have an oncogenic rather than tumor-suppressor role.20 Although the WT1 role in hematopoiesis has not been clarified, expression was inversely associated with stem cell proliferation and differentiation.21

Almost all leukemia-associated WT1 mutations occur within the zinc-finger domains; mainly within a hotspot in exon 7. This exon is the location of the synonymous single-nucleotide polymorphism (SNP) rs16754 that results in adenine (A) or guanine (G) containing alleles. The frequencies of these alleles vary between ethnic groups, so that G is the minor allele in Caucasians and the major allele in Asian populations.22 WT1 polymorphism does not seem to increase susceptibility to AML, with similar frequency between healthy volunteers and AML patients.17 This synonymous SNP may affect to treatment outcome with different potential mechanisms, as alterations in RNA expression, stability, splicing and binding and changes in translational kinetics,23 or on the other hand it could be in linkage disequilibrium with a functional SNP. As for WT1 mutation, several publications have investigated the prognostic impact of WT1 rs16754 SNP in AML cohorts producing non-conclusive or contradictory results.17, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37

Considering that WT1 rs16754 SNP may have a promising albeit inconclusive role in AML treatment, we carried out a meta-analysis on all eligible observational studies to estimate the effect of WT1 polymorphism on AML patients and to quantify the potential between-study heterogeneity.

Materials and methods

Search strategy and selection of studies

This meta-analysis and systematic review was conducted and reported in accordance with the PRISMA guidelines (Preferred Reporting Items for Systematic Reviews and Meta-analyses)38 by two independent authors (JM and MH).

We searched the following databases without restrictions: MEDLINE, Cochrane Central Register, EMBASE, Web of Science and Database of Abstracts of Reviews of Effects (DARE), ProQuest Medical Library, EBSCOhost Online Research Databases, WanFang and Chinese National Knowledge Infraestructure (CNKI) and LILACS. There were no limitations in language, date or status publication restrictions. Additional studies were identified by manual search of the following journals: Leukemia, Cancer, British Journal of Haematology, Blood, Lancet, Lancet Oncology, Journal of Clinical Oncology, Pharmacogenomics, Pharmacogenomics Journal and Pharmacogenetics and Genomics. Congress abstracts of the American Society of Hematology (ASH), the European Hematology Association (EHA) and the Spanish Society of Hematology and Hemotherapy (SEHH) were reviewed. We also hand searched the reference lists of important studies and reviews. The literature last search was on 4 June 2015.

Similar keywords were used in different databases: WT1 (or wilms tumor gen or rs16754), AML and polymorphism (or single-nucleotide or SNP or genetic polymorphisms or pharmacogenetics).

Study selection was conducted by both authors independently. In case of disagreement a third reviewer (LR) was contacted. Studies that fulfilled the following criteria were included: (1) AML studies using standard induction (including cytarabine, anthracyclines and/or etoposide); (2) studies containing useful genotype frequencies of WT1 rs16754 polymorphism; (3) studies evaluating the association of the WT1 polymorphism and AML outcomes; (4) in vivo studies. Studies that included patients with promyelocytic leukemias (FAB subtype M3) were excluded since these leukemias were treated with different regimens.

Data extraction

Information was extracted independently by two reviewers. From included studies the following data was extracted (summarized in Table 1): characteristics of the study (language, publication status and methodological quality), patients baseline characteristics (sex, median age, ethnic origen, AML status, FAB subtype, diagnosis white blood cells (WBC), cytogenetics and main mutations), chemotherapy scheme and polymorphism frequencies of rs16754 SNP (A>G) and if they are in accordance with Hardy–Weinberg equilibrium, as well as genotyping method. We also collected effectiveness outcomes: mean overall survival (OS), complete remission (CR) and others.

Table 1 Characteristics of the studies included in the meta-analysis for WT1 rs16754 polymorphism
Full size table

Three authors were contacted to request missing information regarding outcomes, but the response was not received and we inferred the effectiveness data with the available information (like Kaplan–Meier graphs or number of events), when possible.

Methodological quality

Two reviewers (JM and MH) independently assessed the methodological quality of the included studies. Disagreements were recorded and resolved by a third reviewer (LR). Kappa statistics were used to evaluate reviewers agreement. Criteria used were showed in Table 2.

Table 2 Methodological quality of studies
Full size table

We define the risk of bias with the following criteria: low risk if all criteria were met, moderate risk if only one was not met, and high risk if the number of unmet criteria were two or more.

Analysis

The influence of the WT1 rs16754 genotypes and effectiveness variables was evaluated by pooled odds ratios (OR) and 95% confidence interval (95% CI) using the fixed effects (Mantel–Haenszel method). The statistical significance of pooled OR were determined with Z test (values of P<0.05 were considered statistically significant). RevMan software (version 5.2 The Cochrane Collaboration, The Nordic Cochrane Center, Copenhagen, Denmark) was used to conduct this meta-analysis. The association between AML outcomes and WT1 rs16754 genotypes was performed using dominant (AA vs AG/GG) and recessive models (AA/AG vs GG).

Statistical heterogeneity across studies was tested using the χ2-test (heterogeneity if P<0.1) and the I2 statistic (significant heterogeneity when I2>50%). If heterogeneity was present, the meta-analysis was repeated using the random effects model.

Subgroup analyses were performed for effectiveness variables (OS and CR) based on the ethnicity, age and chemotherapy scheme of the patients. Other predefined variables (included in Table 1) were not analyzed because the information provided by authors was not related with the genotype frequencies and effectiveness variables. We also analyzed OS and CR in patients with CN-AML. Differences between subgroups were evaluated with interaction test (chi-squared).

The possibility of publication bias was conducted evaluating the funnel plots symmetry and with Egger's test (statistical significance if P<0.1).

Results

Systematic search obtained 201 citations from databases and journals and 8 records identified through other sources (Figure 1). Of the 43 citations selected for full reading, only 14 fulfilled the inclusion criteria and were included (all in English). Reviewers showed an excellent agreement in study selection (kappa=0.91).

Figure 1
figure 1

Summary of evidence search and selection.

PowerPoint slide

Full size image

The OS at some of the analyzed times was inferred from the Kaplan–Meier plots in eight studies.24, 25, 29, 30, 31, 34, 37

Study and patient characteristics

Fourteen cohort studies were included (3618 patients).17, 24, 25, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37 The characteristics of the individual studies included are provided in Table 1. Patients’ mean age was 35.4 years (range of 1–88 years) and they were males in 53.5% of the cases. Three studies did not include age or gender data.24, 25, 28 The most abundant ethnic group was Caucasian (72%), followed by Asian (14.7%). Concerning AML, 95.5% were de novo AML and the predominant FAB subtypes represented were M2 (30.9%), M4 (24.7%) and M1 (19.6%). Normal cytogenetic risk was present in 70.4% of patients, whereas 18% had favorable risk and 11.5% had unfavorable one. Mutations were evaluated in nine studies17, 27, 30, 31, 32, 33, 34, 35, 36 and they are summarized in Table 1.

Genotype distributions were in accordance with the Hardy–Weinberg equilibrium in most studies, with three exceptions27, 30, 31 and there were two studies without genotype frequencies.24, 25 The method used for genotyping consisted of PCR and direct sequencing in eight studies,24, 28, 29, 30, 31, 33, 35, 36 real-time reverse transcriptase PCR in four studies,17, 25, 32 mass spectrometry and PCR-restriction fragment length polymorphism to validate in one study37 and two different methods in the last study.34 The baselines characteristics and evaluable AML outcomes for included studies concerning WT1 rs16754 polymorphism are listed in Table 1.

Risk of bias

Only three of the included studies met all 9 criteria and were classified as low risk of bias.17, 32, 37 Eight studies met 8 criteria (intermediate risk of bias).27, 28, 30, 31, 33, 34, 35, 36 The other three studies were categorized as high risk of bias, one met 7 criteria24 and two studies met 6 criteria25, 29 (Table 2). Reviewers showed a substancial agreement (kappa=0.66).

Overall survival

OS for rs16 754 polymorphism was analyzed in 14 studies (Table 1) with 3618 total patients,17, 24, 25, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37 10 of them included enough data to evaluate the recessive model (2875 patients).17, 24, 27, 30, 31, 32, 33, 35, 36, 37 We evaluated OS at 3 and 5 years and calculated OR and 95% CI (Table 3). Two studies only reported OS at 3 years30, 31 and we inferred these data as OS at 5 years. Percentage of OS at 5 years was estimated through Kaplan–Meier graphs in 8 studies that estimated OS at >5 years24, 25, 29, 30, 31, 34, 36, 37 and 8 studies for OS estimations at 3 years.17, 24, 25, 29, 33, 34, 36, 37

Table 3 Summary of ORs with 95% CIs for ordinary genetic contrasts of the association between the WT1 rs16754 polymorphism and effectiveness variables with fixed effects
Full size table

We found evidence indicating that the variant allele G is associated with higher OS (Table 3). Statistical significance was obtained with the dominant model (OR: 1.24, 95% CI: 1.06–1.45, P=0.007, I2: 67%; Figure 2). Significant heterogeneity was detected using the fixed-effect model and the meta-analyses were repeated using the random-effect model (OR: 1.19, 95% CI: 0.86–1.63, I2: 67%, P: 0.30) in which statistical significance was lost. This association was not replicated with recessive model as well as the analysis of OS at 3 years (Table 3).

Figure 2
figure 2

Overall survival at 5 years the dominant model (AA vs AG/GG).

PowerPoint slide

Full size image

The analysis for publication bias showed a little asymmetry in funnel plot of dominant model, as the studies of Damm et al.17 and Becker et al.35 were slightly out the funnel in the dominant model, but this was not significant in the Egger’s test (P=0.65 for OS at 5 years and P=0.76 at 3 years).

Complete remission

Ten studies evaluated CR for WT1 polymorphism17, 27, 30, 31, 32, 33, 34, 35, 36, 37 with 2793 total patients. The pooled effect estimate did not show association between CR and WT1 genotypes in dominant (OR: 1.10, 95% CI: 0.89–1.37, P=0.38, I2: 0%; Table 3) or recessive model (OR: 0.80, 95% CI: 0.58–1.11, P=0.18, I2: 7%; Table 3).

The publication bias analysis yielded nonsignificant asymmetry in funnel plot (though Renneville et al.33 was a bit out the funnel in recessive model) as well as in the Egger’s test (P=0.74).

Other effectiveness variables

The included studies analyzed other effectiveness variables (Table 3), as relapse-free survival (RFS) (5 studies, 703 patients),17, 31, 34, 36, 37 disease-free survival (DFS; 4 studies, 1303 patients)27, 29, 34, 35 and rate of relapse (RR; 5 studies, 1387 patients).27, 30, 31, 33, 36 Two studies evaluated event-free survival32, 36 and other two studies analyzed treatment-related mortality,27, 28 but the limited data published did not allow to measure these variables.

We evaluated separately these variables at 5 years with dominant and recessive models (Table 3). We estimated RFS at 5 years in 4 studies31, 34, 36, 37 and DFS at 5 years in 2 studies,30, 34 using the Kaplan–Meier graphs.

We did not find any difference in RFS and RR for this SNP with both models. The pooled effect estimate showed a higher DFS in G allele carriers, statistically significant in the recessive model (OR: 1.77, 95% CI: 1.09–2.86, P=0.02, I2: 5%).

Subgroup analysis

Ethnics: An interaction between OS and ethnic origen for WT1 was found with the dominant model (Table 4). Caucasian participants carrying G allele showed statistically significant higher OS (OR: 1.21, 95% CI: 1.01–1.45, P=0.04, I2: 75%; Figure 3), but with significant heterogeneity. We recalculated with random effects (OR: 1.27, 95% CI: 0.86–1.87, P=0.24, I2: 75%). Regarding CR, no significant associations were found.

Table 4 Summary of ORs with 95% CIs for ordinary genetic contrasts of the association between the WT1 rs16754 polymorphism and subgroup analyses with fixed effects
Full size table
Figure 3
figure 3

Overall survival at 5 years AA vs AG/GG in the different ethnic subgroups.

PowerPoint slide

Full size image

Age: Pediatric patients carrying G allele showed statistically significant higher OS with the recessive model (OR: 1.42, 95% CI: 1.14–1.77, P=0.002, I2: 60%; Figure 4). It was reanalyzed with random effects model (OR: 1.24, 95% CI: 0.80–1.92, I2: 60%, P: 0.33). We found no significant correlations regarding CR (Table 4). Evaluation of publication bias was nonsignificant in all comparisions.

Figure 4
figure 4

Overall survival at 5 years AA vs AG/GG in the different age subgroups.

PowerPoint slide

Full size image

CN-AML: Our results did not demonstrate an effect of WT1 genotypes in OS and CR with dominant or recessive model in AML patients with normal cytogenetics (Table 4).

Chemotherapy scheme: All the studies included in this systematic review employed cytarabine plus anthracyclines with or without etoposide (Table 1). Patients with G allele treated with idarubicin (OR: 1.38, 95% CI: 1.16–1.65, P=0.0003, I2: 71%; Figure 5) showed higher OS with the dominant model, but this effect was not reproducible with daunorubicin schemes and was lost with random effects model (OR: 1.38, 95% CI: 0.94–2.02, I2: 71%, P: 0.10). Regarding etoposide schemes, the variant allele was associated with higher OS in patients treated with etoposide (OR: 1.74, 95% CI: 1.40–2.15, P<0.00001, I2: 63%; Figure 5) and the opposite effect in treatments without etoposide (OR: 0.79, 95% CI: 0.61–1.02, P=0.07, I2: 0%; Figure 5). In this association a significant heterogeneity was detected and random effects analysis was performed (OR: 1.78, 95% CI: 1.17–2.72, I2: 63%, P: 0.007; Table 4). No significant correlations were found with CR (Table 4).

Figure 5
figure 5

Overall survival at 5 years AA vs AG/GG in the different chemotherapy scheme subgroups.

PowerPoint slide

Full size image

Sensitivity analyses

After excluding outlier studies17, 35 significant differences were not found in OS analysis (OR: 1.23, 95% CI: 1.04–1.47, P=0.02, I2: 52%), denoting that our results are robust and reliable. The pooled effect estimated with fixed or random-effect models did not show significant discrepancies (Table 3).

The studies of Chen et al.31 and Luo et al.30 are the only two studies that evaluated OS at 3 years. No differences were obtained after their exclusion in OS with both models.

Discussion

The findings of this meta-analysis suggest that the WT1 rs16754 polymorphism in AML could influence the standard chemotherapy effectiveness, specifically the variant allele (G) could be associated with increased OS. Our results suggest that the variant genotype of WT1 SNP increased OS, with statistical significance using the dominant model, although the slight increase in OR reveals that the clinical impact of the variant genotype could be limited. In Caucasian and pediatric patients this effect was more manifest. We found significant heterogeneity in OS analysis with the dominant model, and we recalculated this analysis with random effects and statistical significance was lost. This heterogeneity was introduced by two studies,17, 35 which were performed only in CN-AML patients, thus providing a reason for this heterogeneity. The exclusion of these studies in the sensitivity analysis produced a decrease in heterogeneity, whereas significant results in OS were maintained. Similar results with variant G allele were obtained with DFS.

We did not observe any significant effect for CR, possibly as a consequence of the small number of studies that evaluated it. These results were consistent with the CR obtained in the individual studies, all of them without significant differences in this outcome. Other variables related with CR, such as RFS and RR, showed similar values after meta-analysis.

The reason for the increase in survival rates associated with the variant allele of rs16754 remains undetermined. Several action mechanisms have been proposed to explain how this synonymous SNP may alter protein amount, structure and/or activity of WT1. This polymorphism consists of the replacement of CGA by CGG codon, and the latter is used two times more often than the former to encode arginine.39 Therefore, the presence of G allele is predicted to increase the rate of translation, which could potentially affect protein folding. A second possibility is that SNP rs16754 is in linkage disequilibrium with another genetic variant that affects drug metabolism, although only few studies40, 41, 42 reported tag SNPs of WT1 gene (any of them in AML patients), and only one of them40 found other WT1 polymorphisms in linkage disequilibrium with rs16754. A third possibility is that this SNP can affect the sensitivity or timing of co-translational folding and binding of non-coding RNAs. In other synonymous polymorphism, such as C3435T SNP of the multidrug resistance 1 (ABCB1) gene, the variant has been postulated to change P-glycoprotein substrate and inhibitor interaction sites.43 Other explanation could be that GG genotype was more sensitive to cytarabine, and therefore this genotype conferred a favorable outcome in treatment with high-dose cytarabine.17

Although some studies associated the presence of G allele with an increase in WT1 expression,27, 30, 37 this relation is not completely understood. One of these studies30 also demonstrated that GG genotype did not predict improved outcome in stratified groups according to the median expression of WT1. This fact and the contradictory data exposed in other studies17, 32 suggest that the influence on WT1 expression of WT1 SNP rs16754 is not as significant as that of other genetic or nongenetic factors. The subgroup analysis by ethnic origin showed similar results in different races in OS and CR. The opposite genotype frequencies in Asian and Caucasian populations are well known. Specifically, G is a minor allele in Western populations, whereas it is a major allele in Asian populations. Nevertheless, the effect of this SNP in AML outcomes seems to be equivalent in both ethnic groups. Caucasian patients were the unique ethnic origin that obtained statistical significant increase of OS related with variant allele. The impact of this polymorphism in other races remains poorly studied.

The age subgroup analysis found a significant relation between variant allele and higher OS in pediatric patients, not detected in adults. This association was not reproduced with the recessive model, possibly because of the limited number of studies and their contradictory results. Only one study29 compared WT1 SNPs in both age subgroups directly, in which AG genotype was associated with shorter OS and EFS in pediatric patients, but not in adults. These results are contradictory with ours, but it should be noted that this study included a small cohort of Arabic patients undergoing allogeneic stem cell transplant.

Another factor that could explain result variability among studies could be the variability in AML induction therapy. Chemotherapy schemes used in eligible studies were based on cytarabine and anthracyclines. Previous studies did not find differences in the type of anthracycline used and the effect of WT1 polymorphisms. We obtained a significant higher OS with G allele in patients treated with idarubicin, not observed in daunorubicin treatments. Similar results were found with etoposide schemes. These studies mixed different chemotherapy agents, even both anthracyclines, therefore it is difficult to clearly differentiate the drug effect of the schemes employed.

Heterogeneity could difficult the interpretation of the pooled estimations of meta-analyses. We repeated the meta-analysis using the random effects model when heterogeneity was present, and significant results were lost in most of these analyses, therefore limiting their clinical implications. This model involves an assumption that the effects being estimated in the different studies are not identical, but follow some kind of distribution, and it is the appropriate model when heterogeneity is present. In our results, heterogeneity was identified in one of the statistical significant results, which was introduced by two studies17, 35 that only included CN-AML patients in their analysis. After excluding these studies in a post hoc analysis, the heterogeneity decreased while maintaining the OS significant results. Our subgroup analysis of CN-AML patients reveals that WT1 polymorphism does not influence OS or CR. Unfortunately we could analyze the impact of this SNP in other cytogenetic risk groups, such as high-risk or core binding factor leukemia. The ethnic and chemotherapy scheme subgroup analyses showed similar heterogeneity, probably related with the two previously cited studies.17, 35

Some limitations of this meta-analysis should be addressed. First, the role of WT1 in the development of AML and its effect in treatment outcomes are not completely understood. Second, our results are based on unadjusted estimates, whereas a more precise analysis should be conducted if more detailed individual data were available. Third, we found that genotype distributions were not in agreement with Hardy–Weinberg equilibrium in three studies27, 30, 31 and unknown in other two studies.24, 25 Fourth, regarding the subgroup analysis, we find novel associations with G allele of rs16754 and OS that generate new hypothesis but that we cannot explain properly. Finally, subgroup analyses regarding other variables (sex, AML and mutation status, and so on) were not performed due to insufficient data available in the original studies.

WT1 rs16754 polymorphism has never been formerly meta-analyzed and individual studies have shown inconclusive results. Our meta-analysis consisted in the use of extensive search strategies of the literature together with selection criteria. Effectiveness variables were extracted and their pooled effects were estimated with appropriate statistical analyses. Likewise, results obtained were demonstrated to be robust in posterior sensitivity analyses.

In summary, this meta-analysis showed an association between rs16754 polymorphism and OS with fixed effects, but the statistical significance was lost with random effects model. This effect was especially shown in subgroups of Caucasian patients, pediatric patients, and those receiving combined idarubicin-etoposide treatments. The polymorphism showed no association with CR. Future studies based on larger populations and using different age, ethnic and cytogenetic groups should clarify the effect of WT1 in AML outcomes, as well as its influence in different chemotherapy schemes. Moreover, further studies that investigate gene–gene and gene–environment interactions may help understand the role of WT1 in the chemotherapy effectiveness.