Candidate gene association study in pediatric acute lymphoblastic leukemia evaluated by Bayesian network based Bayesian multilevel analysis of relevance

Abstract

Background: We carried out a candidate gene association study in pediatric acute lymphoblastic leukemia (ALL) to identify possible genetic risk factors in a Hungarian population.

Methods: The results were evaluated with traditional statistical methods and with our newly developed Bayesian network based Bayesian multilevel analysis of relevance (BN-BMLA) method. We collected genomic DNA and clinical data from 543 children, who underwent chemotherapy due to ALL, and 529 healthy controls. Altogether 66 single nucleotide polymorphisms (SNPs) in 19 candidate genes were genotyped.

Results: With logistic regression, we identified 6 SNPs in the ARID5B and IKZF1 genes associated with increased risk to B-cell ALL, and two SNPs in the STAT3 gene, which decreased the risk to hyperdiploid ALL. Because the associated SNPs were in linkage in each gene, these associations corresponded to one signal per gene. The odds ratio (OR) associated with the tag SNPs were: OR = 1.69, P = 2.22x10(-7) for rs4132601 (IKZF1), OR = 1.53, P = 1.95x10(-5) for rs10821936 (ARID5B) and OR = 0.64, P = 2.32x10(-4) for rs12949918 (STAT3). With the BN-BMLA we confirmed the findings of the frequentist-based method and received additional information about the nature of the relations between the SNPs and the disease. E.g. the rs10821936 in ARID5B and rs17405722 in STAT3 showed a weak interaction, and in case of T-cell lineage sample group, the gender showed a weak interaction with three SNPs in three genes. In the hyperdiploid patient group the BN-BMLA detected a strong interaction among SNPs in the NOTCH1, STAT1, STAT3 and BCL2 genes. Evaluating the survival rate of the patients with ALL, the BN-BMLA showed that besides risk groups and subtypes, genetic variations in the BAX and CEBPA genes might also influence the probability of survival of the patients.

Conclusions: In the present study we confirmed the roles of genetic variations in ARID5B and IKZF1 in the susceptibility to B-cell ALL. With the newly developed BN-BMLA method several gene-gene, gene-phenotype and phenotype-phenotype connections were revealed. We showed several advantageous features of the new method, and suggested that in gene association studies the BN-BMLA might be a useful supplementary to the traditional frequentist-based statistical method.

Figure 1

Illustration of different dependency relations between certain SNPs in ARID5B and IKZF1 genes, gender and ALL susceptibility. Top panel: The “averaged structure” of the Bayesian networks including ALL susceptibility (red node), gender (purple node) and the SNPs of ARID5B (blue nodes) and IKZF1 (orange nodes). The width of the edges is proportional to their a posteriori probability. The probability of the edges is computed by averaging over the Bayesian networks visited by the MCMC process. See Methods. Bottom panel: The posterior probability of strong relevance (blue columns), edge (direct strong relevance, red columns), pure interaction (green columns), association (purple columns), transitive association (yellow columns) and confounded association (brown columns) of the variables to ALL susceptibility according to the BN-BMLA method.

Figure 2

Overall survival rate of the ALL patients by survival time.

Figure 3

Overall survival rates according to the risk groups (LR = low risk; MR = medium risk; HR = high risk).

Figure 4

Redundancies and interactions according to the BN-BMLA method. The figure shows the magnitude of redundancies (blue curved lines) and interactions (red curved lines) between the variables in the whole dataset (i.e. ALL susceptibility, A panel), in the T-cell lineage sample group (B panel) and in the hyperdiploid sample group (C panel) according to the BN-BMLA method. See Methods for the computation of interaction and redundancy. The width of the curved lines is proportional to the strength of the effect. The a posteriori probability of the strong relevance of the variables is proportional to the length of the dark red columns next to the variable in the inner gray colored ring. The corresponding genes and chromosomes of the SNPs are shown on the outer ring.

Figure 5

Subgraphs of the strongly relevant variables in event-free and overall survival. Left: The “averaged structure” of the subgraphs of the strongly relevant variables in event-free survival according to the BN-BMLA method. Right: The “averaged structure” of the subgraphs of the strongly relevant variables in overall survival according to the BN-BMLA method. The width of the edges is proportional to their a posteriori probability. Edges are shown only if their a posteriori probability exceed 0.5. The probability of the edges is computed by averaging over the Bayesian networks visited by the MCMC process. See Methods. Target variables are indicated with red color, phenotypic variables with purple color, and SNPs with orange color.

Figure 6

Interactions in event-free and overall survival according to the BN-BMLA method. The figure shows the magnitude of interactions (red curved lines) between the variables in the event-free survival (A panel), and in the overall survival (B panel) according to the BN-BMLA method. See Methods for the computation of interaction. The width of the curved lines is proportional to the strength of the effect. The a posteriori probability of the strong relevance of the variables is proportional to the length of the dark red columns next to the variable in the inner gray colored ring. The corresponding genes and chromosomes of the SNPs are shown on the outer ring.