Genome-wide association study of childhood B-cell acute lymphoblastic leukemia reveals novel African ancestry-specific susceptibility loci

Abstract

B-cell acute lymphoblastic leukemia (B-ALL) is the most common pediatric malignancy. Given racial/ethnic differences in incidence and outcomes, B-ALL genome-wide association studies among children of African ancestry are needed. Leveraging multi-institutional datasets with 840 African American children with B-ALL and 3360 controls, nine loci achieved genome-wide significance (P < 5 × 10^-8) after meta-analysis. Two loci were established trans-ancestral susceptibility regions (IKZF1, ARID5B), while the remaining novel loci were specific to African populations. Five-year overall survival among children carrying novel risk alleles was significantly worse (83% versus 96% in non-carriers, P = 4.8 × 10^-3). Novel risk variants were also associated with subtype-specific disease (P < 0.05), including higher susceptibility for a subtype overrepresented in African American children (TCF3-PBX1) and lower susceptibility for a subtype with excellent prognosis (ETV6-RUNX1). Functional experiments revealed novel B-ALL risk variants had allele-specific differences in transcriptional activity (P < 0.05) in B-cell and leukemia cell lines. These findings shed insights into ancestry-related differences in leukemogenesis and prognosis.

Fig. 1. Top B-ALL risk loci in African American children and comparison of genetic ancestry-specific B-ALL polygenic risk scores.

Panel A shows a Manhattan plot with meta-analysis p-values (y-axis) combining association test statistics from logistic regression models in the discovery and replication samples by variant genomic position (x-axis), with the red horizontal line signifying the genome-wide significance p-value threshold (P < 5 × 10⁻⁸). Novel B-ALL risk loci are annotated with nearest gene names. Odds ratios (ORs, circles) and corresponding 95% confidence intervals (CIs, whiskers) from comparable logistic regression models for increasing deciles of a B-ALL polygenic risk score (PRS) including replicated risk variants achieving GWS after ADMIRAL meta-analysis in African ancestry children are shown in the entire study sample (AFR PRS), with a reference group with median risk or less (panel B, N = 4200). ORs (circles) with 95% CIs (whiskers) are shown for a PRS including suggestively associated lead variants from the ADMIRAL discovery GWAS (10 variants with P < 5 × 10⁻⁶; AFR PRS) and a PRS based on the largest European B-ALL meta-analysis to date (Vijayakrishnan et al.; EUR PRS) are shown in the replication study sample (panel C, N = 920) where both PRSs omit overlapping trans-ancestral loci (ARID5B, IKZF1), also using a reference group with median risk or less. Decile increments of the African ancestry-specific PRS had a dose-response relationship with B-ALL risk in independent ADMIRAL replication data (P_trend =1.2 × 10⁻⁶) while the European counterpart did not (P_trend= 0.20). All presented p-values are from two-sided statistical tests and are not adjusted for multiple testing.

Fig. 2. Novel African ancestry-specific childhood B-ALL risk loci at CNTN4 (left) and FAM174A (right).

LocusZoom plots of the genomic region surrounding index variants are shown with African ancestry linkage disequilibrium patterns, where variants are color-coded by their magnitude of linkage disequilibrium (LD, r) with the index variant. All presented p-values are from two-sided statistical tests (logistic regression models) and are not adjusted for multiple testing. Window sizes around index variants for CNTN4 (A) and FAM174A (B) are 50 and 200 kilobases, respectively. Peaks from CEBPA ChIP-seq of B-cell precursor leukemia BLaER1 cell line and DNase-seq of NAMALWA Burkitt’s lymphoma B-lymphocyte cell line are provided for the same genomic window. Effect allele frequencies (EAFs) in African (red) and European (green) 1000 Genomes continental ancestral groups are also provided.

Fig. 3. Admixture and relevance of global and local African ancestry on B-ALL risk alleles.

Contrasts in risk allele frequencies for index variants in African ancestry cases (filled) and controls (unfilled) stratified by global African ancestry proportions at novel African ancestry-specific loci (CNTN4, FAM174A) versus known trans-ancestral loci (ARID5B, IKZF1) are shown in panel A. Panel B shows the distribution of a B-ALL PRS from the largest European meta-analysis to date (Vijayakrishnan et al.) among ADMIRAL B-ALL cases with lower (<40%, green) versus higher (>80%, orange) African global ancestry proportions. The p-value (P = 3.8 × 10⁻⁶) is from a two-sided t-test to assess the difference in means between groups. Differences in the local ancestry tract composition in genomic regions overlapping index variants (dashed line) for African ancestry ADMIRAL participants with at least one effect allele (EA) at novel African ancestry-specific loci (CNTN4, FAM174A) versus known trans-ancestral loci (ARID5B, IKZF1) are illustrated in panel C.

Fig. 4. Dual-luciferase reporter assay activity for prioritized B-ALL GWAS variants and overall B-ALL risk loci evidence tally.

Panel A shows relative dual-luciferase reporter assay activity comparing non-effect (NEA) and effect (EA) alleles at prioritized African ancestry-specific B-ALL GWAS risk variants and known B-ALL risk variants (IKZF1, ARID5B) with significant allele-specific differences in at least one cell line (ns=not statistically significant). Corresponding relative luciferase expression in B-cell lymphoblastoid (GM12878, green) and leukemia (697, blue) cell lines are shown. Boxplots show median, interquartile range, and minimum and maximum values. In total, 6 biological replicates representing independent transfections were performed for each SNP allele. Biological replicates for each SNP were tested using two independent 96-well plates and the activity for each biological replicate was calculated by taking the average of 3 technical replicates. Statistical significance from two-sided paired t-tests are annotated in each sub-panel. Panel B illustrates the overall tally of evidence across study resources for candidate B-ALL risk loci, where novel African-ancestry specific risk loci are bolded.

Fig. 5. Overall survival after B-ALL diagnosis by novel risk allele carrier status among African American children in Children’s Oncology Group (COG) clinical trials.

In all panels, survival curves for participants carrying at least one candidate African ancestry-specific risk allele are shown in red (line) while those who do not are shown in black (line), with censoring status (plus signs) and shaded areas in lighter red and gray representing corresponding 95% confidence intervals (CIs). Panel A shows all COG participants in the ADMIRAL study with clinical data, while panels B and C show subgroups stratified by enrollment in standard versus high risk treatment protocols, respectively. Differences in survival curves were evaluated with two-sided log-rank tests (p-value shown in the lower right quadrant for each panel). In the overall data, the hazard ratio (HR) adjusted for sex, genetic ancestry, and treatment risk stratification is provided along with corresponding 95% CIs.