Associations between AT-rich Interactive Domain 5 B gene Polymorphisms and Risk of Childhood Acute Lymphoblastic Leukemia : a Meta-analysis

Acute lymphoblastic leukemia (ALL), the most common malignancy in children (Linabery and Ross, 2008; Jiang et al., 2013), is a heterogeneous disease, with respect to its underlying cellular and molecular biology (Papaemmanuil et al., 2009), acquired genetic abnormalities, and social clinical responses to combination chemotherapy (Pui et al., 2004). The mechanisms involved in the etiology of childhood ALL remain obscure. It is hypothesized that both environmental and genetic factors contribute to the initiation of leukemogenesis (Greaves, 2006). Previous genome-wide association studies (GWAS) have implicated the associations of several single nucleotide polymorphisms (SNPs) in the AT-rich interactive domain 5B (ARID5B) gene with childhood ALL (Papaemmanuil et al., 2009; Trevino et al., 2009). Importantly, the ARID5B gene showed specificity for B lineage ALL and B-hyperdiploid subtype (Papaemmanuil et al., 2009; Trevino et al., 2009) ARID5B, the gene encoding the ARID5B protein, which located on chromosome 10q21.2, is one of the common susceptibility loci associated with childhood ALL risk (Papaemmanuil et al., 2009; Trevino et


Introduction
Acute lymphoblastic leukemia (ALL), the most common malignancy in children (Linabery and Ross, 2008;Jiang et al., 2013), is a heterogeneous disease, with respect to its underlying cellular and molecular biology (Papaemmanuil et al., 2009), acquired genetic abnormalities, and social clinical responses to combination chemotherapy (Pui et al., 2004).The mechanisms involved in the etiology of childhood ALL remain obscure.It is hypothesized that both environmental and genetic factors contribute to the initiation of leukemogenesis (Greaves, 2006).Previous genome-wide association studies (GWAS) have implicated the associations of several single nucleotide polymorphisms (SNPs) in the AT-rich interactive domain 5B (ARID5B) gene with childhood ALL (Papaemmanuil et al., 2009;Trevino et al., 2009).Importantly, the ARID5B gene showed specificity for B lineage ALL and B-hyperdiploid subtype (Papaemmanuil et al., 2009;Trevino et al., 2009) ARID5B, the gene encoding the ARID5B protein, which located on chromosome 10q21.2, is one of the common susceptibility loci associated with childhood ALL risk (Papaemmanuil et al., 2009;Trevino et

Search strategy and selection criteria
The literature retrieval was performed by two independent reviewers (Hui Zeng, Ning-hua Cui).All studies included in the meta-analysis were selected by searching the PubMed, EMBASE databases up to May 2014 using the following keywords: " (childhood acute lymphoblastic leukemia or childhood ALL) and (AT-rich interactive domain 5B gene or ARID5B) and (polymorphism or genotype or variant)".All references in these studies were examined to identify additional research that was not indexed by the databases.Articles that reported results from more than 1 population were considered as separate studies.We selected only published articles written in English.

Selection criteria for articles
The selected studies were required to meet all the following criteria: (1) evaluating the associations between ARID5B SNPs rs10994982, rs7089424 and childhood ALL risk; (2) case-control design; (3) studies with cases younger than 15 years; (4) results with sufficient published data to estimate odds ratios (ORs), confidence interval (CI) and P value.The exclusion criteria were as follows: (1) review articles, abstracts, case reports, reports with incomplete data; (2) studies about adult ALL; (3) studies based on acute myeloid leukemia (AML) and/or Burkitt lymphoma.For duplicate articles, only the most recent or largest data set was selected.

Data extraction
Data were extracted from eligible articles independently by two of the authors (Xue-bin Wang, Tuo Zeng), with any disagreement resolved by consensus.The following information was collected in a predefined data collection form: first author's name, publication year, country, ethnicity, total number of cases and controls, genotyping methods, Hardy-Weinberg equilibrium (HWE) and quality assessment of studies.Quality assessment was performed in each of the acceptable studies in duplicate by independent reviewers using the Newcastle-Ottawa Quality Assessment Scale (Stang, 2010) for all casecontrol studies.

Statistical analysis
The deviation of HWE for the genotype distribution in controls was assessed using the chi-squared test.The strength of association was expressed as pooled ORs along with the corresponding 95%CIs.In primary, we mainly examined the overall effects for each polymorphism.Firstly, allelic comparison for each polymorphism (A vs G for SNP rs10994982, C vs A for SNP rs7089424) was used to detect overall differences.Secondly, we compared each genotype with addictive model (homozygote comparison and heterozygote comparison).Thirdly, recessive model and dominant model were assessed for each polymorphism.We also planned subgroup analyses to determine the effects of ethnicity, ALL subtypes (B-lineage ALL, B-hyperdiploid ALL) on the summary risk estimate.Bonferroni correction (Shi et al., 2012) was used to control for the multiple testing in view of ten comparisons under investigation (significance was set at 0.05/10=0.005).
Heterogeneity between studies was assessed by χ 2based Q-tests and I 2 tests, where I 2 (%)>50% or p<0.10 was considered significantly heterogeneous (Higgins et al., 2003).The random-effects model (DerSimonian-Laird) (DerSimonian and Kacker, 2007) was used to assess pooled ORs when significant heterogeneity was observed.Otherwise, the fixed-effects model (Mantel-Haenszel) (Higgins and Thompson, 2002) was used.Further, Galbraith plot analyses were used to visualize the impact of individual studies on the overall homogeneity, which identified the outlier as possible major sources of heterogeneity (Huy et al., 2010).Sensitivity analyses were conducted to evaluate the stability of the combined results by omitting one study at a time or excluding those studies that deviated from HWE (Yang et al., 2014).Publication bias was assessed graphically by funnel plots (Langan et al., 2012) and formally by Egger's test (Egger et al., 1997) and Begg's test (Begg and Mazumdar, 1994).The power of meta-analysis for each polymorphism to detect some effect size was estimated according to the method recommended by Hedges and Pigott (Hedges and Pigott, 2004), given a significant value of 0.05.All statistical analyses were conducted using RevMan 5.2 (The Nordic Cochrane Centre, The Cochrane Collaboration) & STATA 12.0 (Stata, College Station, TX, USA).

Study characteristics
A total of 41 relevant articles were initially identified from PubMed, EMBASE.After titles and abstracts were screened, 14 articles were excluded because of irrelevant or duplicate data.The full texts of the remaining 27 records were carefully reviewed.Among these articles, five articles were about other SNPs (Han et al., 2010;Yang et al., 2010;Wang et al., 2013;Guo et al., 2014) or overlapped data (Xu et al., 2012), two articles were excluded because of adult ALL (Peyrouze et al., 2012) or AML (Rudant et al., 2013), three articles were excluded due to case reports (Paulsson et al., 2010) or review papers (Levine, 2009;Qian et al., 2012)(Figure 1).Another article (Chokkalingam et al., 2013) examined SNPs rs10994982 and rs7089424, but detailed genotype counts could not be obtained despite attempts to contact the authors.Thus, this article was excluded as well.Two of the eligible articles contained data from 4 independent studies and we treated them independently (Papaemmanuil et al., 2009;Prasad et al., 2010).Therefore, 14 articles including 16 studies were included in the present meta-analysis.Here, eight articles (including 3684 patients and 25085 controls) dealt with SNP rs10994982 (Trevino et al., 2009;Healy et al., 2010;Orsi et al., 2012;Gutierrez-Camino et al., 2013;Linabery et al., 2013;Ross et al., 2013;Xu et al., 2013;Emerenciano et al., 2014).Among the included articles, two articles only reported allele contrasts (Orsi et al., 2012;Xu et al., 2012).For SNP rs7089424, nine articles including 11 studies (totally 4378 patients and 7928 controls) were included in the analyses (Papaemmanuil et al., 2009;Healy et al., 2010;Prasad et al., 2010;Vijayakrishnan et al., 2010;Pastorczak et al., 2011;Lautner-Csorba et al., 2012;Orsi et al., 2012;Gutierrez-Camino et al., 2013;Lin et al., 2014), but one article only provided allele contrast (Orsi et al., 2012).For studies only provided allele contrasts, the DOI:http://dx.doi.org/10.7314/APJCP.2014.15.15.6211Associations between AT-rich Interactive Domain 5B genePolymorphisms and Risk of Childhood ALL: a Meta-analysis deviation of HWE for the genotype distribution in controls was not performed because of insufficient data.The study characteristics and genotype distributions included in the meta-analysis were listed in (Table 1).

Meta-analysis about SNP rs10994982
For SNP rs10994982, at the effect size of their respective OR values, the power of meta-analysis for all five comparisons was greater than 0.999.Summaries of the ORs for different comparisons were provided in Table 2 and Figure 2A.In brief, the robust associations between SNP rs10994982 and children ALL were detected using all genetic models even after Bonferroni correction (for A vs G, OR=1.44, 95%CI=1.31-1.58,p<0.00001; for AA vs GG, OR=1.98, 95%CI=1.67-2.34,p<0.00001; for AG vs GG, OR=1.33, 95%CI=1.14-1.56,P=0.0007; for AA vs AG+GG, OR=1.61, 95%CI=1.42-1.82,p<0.00001; for AA+AG vs GG, OR=1.55, 95%CI=1.34-1.79,p<0.00001).Subgroup analyses restricted to the Caucasian populations were performed.Significant associations were also identified (Table 2).In subtype analyses, the differences in all genetic models were highly significant when analyses were performed between B-lineage ALL cases and non-ALL controls (Table 3).So the primary impact of variation defined by SNP rs10994982 was for B-lineage ALL.In subtype analyses of B-lineage ALL, there were higher differences in the genotype distributions of SNP rs10994982 between B-hyperdiploid ALL cases and non-ALL controls in all genetic models (Table 3).When B-hyperdiploid ALL cases and cases with all other B-lineage ALL were compared, statistical differences were observed in allelic comparison, homozygous comparison and the recessive model, although they were all not sufficient robust to withstand the Bonferroni correction (Table 3).In general, the frequency of the risk allele A at SNP rs10994982 was significantly different in patients with B-hyperdiploid ALL than in all other patients with B-lineage ALL and than in non-ALL controls.

Figure 1. Flow Chart of Article Selection in our Meta-Analysis
Moderate heterogeneity was observed in the allelic comparison, and the Galbraith plot analyses indicated that the study from Orsi et al. (Orsi et al., 2012) was mainly responsible for the observed heterogeneity (Figure 3A).When we excluded this study, the moderate heterogeneity was significantly decreased and the association was still significantly different (for A vs G, OR=1.39, 95%CI=1.31-1.48,p<0.00001,P for heterogeneity=0.30,I 2 =17%).Sensitivity analyses indicated that the pooled ORs for SNP rs10994982 were consistently significant in childhood ALL risk by omitting one study at a time.We found no evidence of publication bias through both funnel plots    2).

Meta-analysis about SNP rs7089424
For SNP rs7089424, at the effect size of their respective OR values, the power of meta-analysis for all five comparisons was greater than 0.999.Details of each comparison were provided in (Table 2 and Figure  2B).In brief, associations with SNP rs7089424 were revealed in all genetic comparisons (for C vs A, OR=1.64, 95%CI=1.46-1.83,p<0.00001; for CC vs AA, OR=2.43, 95%CI=1.99-2.97,p<0.00001; for CA vs AA, OR=1.67, 95%CI=1.52-1.83,p<0.00001; for CC vs CA+AA, OR=1.84, 95%CI=1.57-2.16,p<0.00001; for CC+CA vs AA, OR=1.83, 95%CI=1.60-2.09,p<0.00001).In the subgroup analyses by ethnicity, significant associations were found among Caucasians but not among Asians in all genetic models (Table 2).In further analyses, the SNP rs7089424 CC genotype frequencies in non-ALL controls showed a statistically significant decreasing trend (p<0.0001,Chi-square test, data not shown) from Caucasian subgroup to Asian subgroup, while there was a concomitant increase in SNP rs7089424 AA genotype frequencies.These differences may contribute to racial differences in childhood ALL incidences.
In subtype analyses, all the comparisons revealed significant differences between B-lineage ALL cases and non-ALL controls (Table 3).Subtype analyses of B-lineage ALL were not performed because of limited data.

Discussion
According to our meta-analysis, there was a significant association between SNP rs10994982 and childhood ALL risk.In addition, subtype analyses of B-lineage ALL provided evidence that SNP rs10994982 was highly associated with the risk of developing B-hyperdiploid ALL.Although, only 3 studies were performed to assess the association between SNP rs10994982 and B-hyperdiploid ALL risk.However, power analyses in our pooled B-hyperdiploid ALL patients and non-ALL controls indicated there was 91-100% power to detect the observed associations.The mechanism of ARID5B on disease susceptibility remains abstruse.A pivotal piece of evidence is that ARID5B homozygous null mice show transient immune abnormalities including a reduction in early B-cell progenitors (Lahoud et al., 2001;Sherborne and Houlston, 2010).Besides, expression of ARID5B have associated with recombination activating gene 1 (RAG1) expression in bone marrow (Jensen et al., 2010), which supports a role for this gene in early B-cell development.
We have shown that a significant effect of ARID5B SNP rs7089424 on childhood ALL susceptibility.When stratified by ethnicity, SNP rs7089424 presented as a risk factor for childhood ALL in Caucasian populations, but not in Asians.It was notable that the frequency of the CC genotype was higher in the control populations of Asian subgroup than in controls of Caucasian subgroup (p<0.0001,Chi-square test, data not shown), while there was a concomitant decrease in SNP rs7089424 AA genotype frequency.An explanation was that the racial differences in childhood ALL incidence could partly be attributed to differences in genotype frequencies between populations of Asian and Caucasian ancestries at ARID5B loci.However, due to the limited number of relevant studies among Asian populations included in this meta-analysis, the pooled results in Asian subgroup may have insufficient statistical power to detect a slight effect or have generated a fluctuated risk estimate.Besides, in Asian subgroup, the genotype distributions in the control population of Lin's study (Lin et al., 2014) was deviated from HWE.Because of the reasons of disequilibrium, the results of genetic association studies might be spurious (Trikalinos et al., 2006).So more well designed studies about racial difference were required in the future.
Moderate heterogeneity was observed in allelic comparison for SNP rs10994982 and in the most comparisons for SNP rs7089424.Heterogeneity is a potential problem when interpreting the results of a metaanalysis, and finding the sources of heterogeneity is one of the most important goals of meta-analysis (Ioannidis et al., 2007).So first, we attempted to use a random-effects model to adequately capture the trade-off between the association estimates in comparisons with significant heterogeneity.Second, Galbraith plot analyses were performed to identify the outliers which might be the sources of the heterogeneity and we found that all the heterogeneity was effectively eliminated or decreased after excluding the outliers.Third, sensitivity analyses were carried and removal of each study did not alter the associations with childhood ALL risk.All these procedures suggested the reliability of these results.
Despite our efforts in performing a analysis, some limitations also existed in our meta-analysis.First, this meta-analysis was conducted based on case-control studies, which might encounter recall and selection bias.Second, in subgroup analyses by ethnicity (Asian subgroup), subtype analyses, the number of studies were relatively small, although there was sufficient power (≥80%) to detect the observed associations.Third, lack of the original data of available studies limited our further evaluation of potential associations, such as gender-specific associations.Finally, given the fact that ALL subtypes show considerable differences in terms of clinical features, treatment response, etc (Iqbal, 2014), the associations between these two ARID5B polymorphisms and other ALL subtypes (such as T-lineage ALL) should be concerned in future studies.
In conclusion, this meta-analysis confirmed the associations between two ARID5B hot spot polymorphisms and the risk of childhood ALL.In addition, subtype analyses of B-lineage ALL provided evidence that SNP rs10994982 was highly associated with the risk of developing hyperdiploid.We believe that our findings will be useful for future studies in childhood ALL.

Figure 2 .
Figure 2. Forest Plots of Pooled OR with 95%CI for Association between Two ARID5B Gene Polymorphisms and Childhood ALL Risk.(A) Forest plots of pooled OR with 95%CI for association between SNP rs10994982 and childhood ALL risk (A vs G, randomeffects model); (B) Forest plots of pooled OR with 95%CI for association between SNP rs7089424 and childhood ALL risk (C vs A; random-effects model) DOI:http://dx.doi.org/10.7314/APJCP.2014.15.15.6211Associations between AT-rich Interactive Domain 5B genePolymorphisms and Risk of Childhood ALL: a Meta-) and Egger's test and Begg's test (Table

Figure 3 .
Figure 3. Galbraith plots of association between two ARID5B gene polymorphisms and childhood ALL risk.(A) Galbraith plots of association between SNP rs10994982 and childhood ALL risk (A vs G); (B) Galbraith plots of association between SNP rs7089424 and childhood ALL risk (C vs A)

Table 2 . Total and Subgroup Analyses of ARID5B gene two Polymorphisms on Childhood ALL
*N number of studies; P (Q) P value of Q test for heterogeneity; P (Z) P value of Z test for significant test; P (Egger) P value of Egger's test for publication bias; P (Begg) P value of Begg's test for publication bias; a R: random-effect model; F: fixed-effect model

Table 3 . Comparison of SNP rs10994982 and SNP rs7089424 Genotypes between B-lineage ALL Cases and Controls, B-Hyperdiploid ALL Cases and Controls, B-Hyperdiploid ALL Cases and Other B-Lineage ALL Cases in This Meta-analysis
*N number of studies; P (Z) P value of Z test for significant test