Association of GATA3 Polymorphisms With Minimal Residual Disease and Relapse Risk in Childhood Acute Lymphoblastic Leukemia

Abstract

Background

Minimal residual disease (MRD) after induction therapy is one of the strongest prognostic factors in childhood acute lymphoblastic leukemia (ALL), and MRD-directed treatment intensification improves survival. Little is known about the effects of inherited genetic variants on interpatient variability in MRD.

Methods

A genome-wide association study was performed on 2597 children on the Children’s Oncology Group AALL0232 trial for high-risk B-cell ALL. Association between genotype and end-of-induction MRD levels was evaluated for 863 370 single nucleotide polymorphisms (SNPs), adjusting for genetic ancestry and treatment strata. Top variants were further evaluated in a validation cohort of 491 patients from the Children’s Oncology Group P9905 and 6 ALL trials. The independent prognostic value of single nucleotide polymorphisms was determined in multivariable analyses. All statistical tests were 2-sided.

Results

In the discovery genome-wide association study, we identified a genome-wide significant association at the GATA3 locus (rs3824662, odds ratio [OR] = 1.58, 95% confidence interval [CI] = 1.35 to 1.84; P = 1.15 × 10^-8 as a dichotomous variable). This association was replicated in the validation cohort (P = .003, MRD as a dichotomous variable). The rs3824662 risk allele independently predicted ALL relapse after adjusting for age, white blood cell count, and leukemia DNA index (P = .04 and .007 in the discovery and validation cohort, respectively) and remained prognostic when the analyses were restricted to MRD-negative patients (P = .04 and .03 for the discovery and validation cohorts, respectively).

Conclusion

Inherited GATA3 variant rs3824662 strongly influences ALL response to remission induction therapy and is associated with relapse. This work highlights the potential utility of germline variants in upfront risk stratification in ALL.

Early treatment response reflects inherent leukemia cell sensitivity to chemotherapeutics in vivo and is predictive of relapse and survival in children and adults with acute lymphoblastic leukemia (ALL) (1–4). Contemporary pediatric ALL treatment regimens start with 4 to 6 weeks of induction therapy to rapidly reduce leukemia burden and achieve disease remission. Despite wide variations in remission induction treatment regimens, minimal residual disease (MRD) at the end of induction therapy is a consistently independent prognostic factor and associated with long-term outcome (5–10). MRD can be determined by flow cytometry or polymerase chain reaction for the vast majority of childhood ALL (3). As one of the strongest prognostic factors in pediatric ALL, MRD is now widely used to guide treatment stratification (11). Patients with MRD above a specific threshold after induction therapy are often reassigned to a higher-risk group and treated with more intensive chemotherapy during subsequent phases of ALL treatment.

Although MRD-directed treatment stratification has improved the outcome of pediatric ALL, the biological basis of the variability in treatment response remains largely unclear. Certain molecular subtypes (eg, Philadephia chromosome-positive [Ph⁺] ALL and Ph-like ALL) respond poorly to initial therapy and are associated with MRD positivity at the end-of-induction phase (12,13), whereas ETV6-RUNX1, high hyperdiploidy, and trisomies of chromosomes 4 and 10 confer favorable outcome and MRD negativity (14,15). In parallel, we and others have identified inherited genetic factors that affect metabolism and disposition of chemotherapeutic agents, including thiopurine methyltransferase (TPMT) (16,17). Our prior genome-wide association study (GWAS) also identified intronic variants in the interleukin-15 (IL15) gene that are strongly associated with MRD status (18). Subseqent mechanistic work confirmed that IL-15 influences leukemia microenvironment and thus leukemia sensitivity to chemotherapeutic agents. However, these studies linking inherited host factors and treatment outcomes were performed in the context of ALL therapy with notable differences compared with contemporary frontline ALL protocols (eg, pegaspargase vs Escherichia coli asparaginase, additional exposure to nonantimetabolite therapeutics during postinduction treatment) (18,19).

To that end, we conducted a GWAS in 2597 children enrolled on the Children’s Oncology Group (COG) AALL0232 clinical trial for newly diagnosed high-risk B-cell ALL (B-ALL) to systematically identify germline variants related to MRD; we then tested top single nucleotide polymorphisms (SNPs) using cases in the COG P9905 and 6 series of trials with similar presenting features as an independent validation cohort (20–23). The prognostic value of identified MRD risk variants was also explored in multivariable analyses.

Methods and Materials

Study Design and Patients

The GWAS discovery cohort included 2597 newly diagnosed ALL cases enrolled in the COG AALL0232 trial (clinicaltrials.gov ID NCT00075725) for National Cancer Institute (NCI) high-risk B-ALL (Figure 1) (20,24). This trial randomized patients in a 2-by-2 factorial design to receive prednisone or dexamethasone during induction and Capizzi methotrexate plus pegaspargase (C-MTX) or high-dose methotrexate (HD-MTX) during interim maintenance. The validation cohort consisted of 491 patients with NCI high-risk ALL enrolled in COG P9905 (NCT00005596) (22) or P9906 (NCT00005603) (23), with NCI high risk presenting clinical features similar to the discovery cohort. Patients in the P9905 trial were randomized between 2 regimens of MTX delivery and then between receiving or not receiving delayed intensification; the P9906 trial was a single-arm study for patients with high-risk ALL (21). ALL cases in the P9906 trial included here were also described in our prior genomic analyses of MRD (18).

Figure 1.

Consort flow diagram for the discovery (blue-shaded box) and validation (pink-shaded box) genome-wide association study (GWAS) cohorts. ALL = acute lymphoblastic leukemia; COG = Children’s Oncology Group; MAF = minor allele frequency; MRD = minimal residual disease; SNP = single nucleotide polymorphisms.

MRD data were assessed by standardized flow cytometry (20). An MRD level of 0.1% (1 leukemic cell/1000 normal cells) or higher was used to define MRD positivity in our study (21). In the P9905 and 9906 vs AALL0232 trials, MRD reflected response to largely similar combination chemotherapy during induction phase although with notable difference in asparaginase formulation (20,22,23). Patients with no available germline DNA specimens or missing MRD data at the end of induction therapy were excluded from this study (n = 55 for AALL0232 and n = 168 for P9905 and 6). Germline DNA was extracted from peripheral blood samples during clinical remission.

The clinical trials were approved by local institutional review boards, and informed consents for trial enrollment and banking of specimens for future research were obtained from parents, guardians, and/or patients, as appropriate. This study was approved by the St Jude Children’s Research Hospital institutional review board.

Genotyping and Quality Control

Genome-wide SNP genotyping was performed with the Affymetrix Human SNP Array 6.0 for cases from the AALL0232 and P9905 cohorts; the Affymetrix 500 K SNP array was used for those from the P9906 cohort. Genotype calls (coded as 0, 1, and 2 for AA, AC, and CC genotypes, respectively) were determined by the Birdseed v2 (Affymetrix SNP 6.0) algorithm. Imputed SNPs were subjected to quality control where prior to GWAS, SNPs with a minor allele frequency less than 0.5% or a call rate less than 95% and those without annotation were excluded.

We also performed exome-centric genome-wide SNP genotyping for the majority of cases in the discovery cohort (n = 2332), using the Illumina Infinium HumanExome Beadchip (Supplementary Figure 1, A, available online).

Determining Genetic Ancestry and Population Structure

Genetic ancestry was determined using STRUCTURE (version 2.2.3) on the basis of genotypes at 30 000 SNPs randomly selected from the Affymetrix SNP arrays (25). HapMap samples from the CEU (n = 60), YRI (n = 60), CHB and JPT (n = 90) panels, and Native American references (26) (n = 105) were used to represent European, African, Asian, and Native American ancestry groups, respectively. We assumed that the proportions for these 4 ancestry groups summed to 100% in each genotyped individual. European Americans, African Americans, and Asians were defined as having more than 95% European genetic ancestry, more than 70% African ancestry, and more than 90% Asian ancestry, respectively. Hispanics were individuals for whom the proportion of Native American ancestry was more than 10% and was greater than the proportion of African ancestry (27).

Genome-Wide Association, Outcome Correlation, and Statistical Analyses

The association between MRD status and SNP genotype was tested. First, MRD was treated as a dichotomous variable (ie, MRD ≥ 0.1% or < 0.1%) and modeled with logistic regression using the SNP as a predictor and adjusting for ancestry and treatment strata (ie, dexamethasone vs prednisolone in the discovery cohort; treatment protocol in the validation cohort). Secondly, MRD was treated as an ordinal variable (<0.1%, ≥0.1%-<1%, ≥1%), and its association with SNP was measured using a Spearman-type correlation stratified by ancestry and treatment arms. In the discovery GWAS analyses, we applied a P value cut-off of 5 × 10^-8 and sought to verify SNPs meeting this threshold of P = 5 × 10^-8 in an independent validation series (P9905 and 6). We also considered additional candidate SNPs using a relaxed threshold of 1 x 10^-5 in the discovery cohort for validation to explore variants with potentially less robust but reproducible associations. Variants were considered validated if P was less than .05.

We first performed GWAS focusing on genotypes from Human SNP Array 6.0 array and a separate exome-based GWAS focusing on variants examined using the HumanExome Beadchip. For GATA3 SNP rs3824662, we also tested its association with MRD (as a dichotomous variable) using multivariable regression analyses, including age at diagnosis (<10 years vs ≥10 years), white blood cell count at diagnosis (<50 x 10⁹/L vs ≥50 x 10⁹/L), and DNA index (<1.16 vs ≥1.16).

MRD-related GATA3 SNPs were tested for their association with ALL relapse in patients eligible for postinduction evaluation in the COG AALL0232 and P9905 and 6 cohorts (n = 2289 and 455, respectively). Relapse was defined as the detection of leukemic cells in the bone marrow and/or extramedullary space after a period of remission and was treated as a time-to-event variable. Association of genotype with cumulative incidence of relapse was determined using the Fine and Gray hazard-rate regression model in both cohorts, with death in remission, induction death, and second malignancy as competing events. Multivariable analysis adjusted for ancestry, treatment strata (28,29) (5 strata for the AALL0232 cohort: rapid early response—dexamethasone/C-MTX, prednisolone/C-MTX, dexamethasone/HD-MTX, prednisolone/HD-MTX; slow early response—C-MTX or HD-MTX; 5 strata for the validation cohort: P9905 Regimens A-D and P9906) and other presenting features (age at diagnosis [<10 years vs ≥10 years], white blood cell count at diagnosis [<50 x 10⁹/L vs ≥50 x 10⁹/L], and DNA index [<1.16 vs ≥1.16]). Association was first evaluated for all patients in each cohort and repeated for those who were MRD negative (ie, <0.1%) at the end of induction therapy. Associations of GATA3 SNP genotype with ALL presenting features were analyzed by Fisher test or χ² test. R (version 3.4.2) statistical software was used for analyses unless otherwise stated, and all statistical tests were 2-sided.

Results

GWAS of End-of-Induction MRD Level

In the discovery GWAS, we systematically evaluated the association of SNP genotype with MRD status at the end of induction therapy as a dichotomous and then as an ordinal variable in 2597 children on COG AALL0232 (Table 1). The final GWAS dataset included 863 370 SNPs (Figure 1). After adjusting for genetic ancestry and induction treatment strata (dexamethasone vs prednisolone), both analyses identified a single locus as the genome-wide top hit: rs3824662 within the GATA3 gene (risk allele frequency = 41.4% in MRD positive vs 30.5% in MRD negative; odds ratio [OR] = 1.58, 95% confidence interval [CI] = 1.35 to 1.84; P = 1.15 × 10^-8 in the dichotomous variable analysis; risk allele frequency = 30.5% in MRD <0.1%, 40.4% in MRD ≥0.1% to <1%, 42.5% in MRD ≥ 1%; P = 2.49 × 10^-8 in the ordinal variable analysis) (Figure 2). MRD positivity increased progressively with each additional copy of the risk allele (8.1%, 14.3%, and 17.3% in patients with CC, CA, and AA genotype, respectively), consistent with an additive genetic model. In fact, the A allele at this SNP was also statistically significantly associated with increased frequency of both intermediate MRD (≥0.1% to <1%) and high MRD (≥1%) (Figure 3). Two other SNPs at this locus, rs3781093 and rs477461, were statistically significant (OR = 1.50, 95% CI = 1.28 to 1.75; P = 3.85 × 10^-7; and OR = 1.48, 95% CI = 1.25 to 1.76; P = 6.95 × 10^-6, respectively). Of these 2 SNPs, rs3781093 was in strong linkage disequilibrium with rs3824662 (r² = 0.84, D’ = 0.95 in HapMap CEU), and rs477461 was not. In addition to the GATA3 SNPs, we identified variants in SNORD74, LINC00366, AKAP6, NOS1, DOCK2, GALNTL2, RP6-114E22.1, ANO2, RP11-649A16.1, AC078821.1, AKAP2, and UCN3 (P ≤ 1 × 10^-5; Supplementary Table 1, available online). Quantile-quantile plots of logistic regression test for the GWAS indicated inflation only at the tail of the distribution (λ  =  0.999 for dichotomous analysis, and λ  =  1.022 for ordinal analysis), suggesting that population structure was adequately controlled (Supplementary Figure 2, available online).

Table 1.

Patient characteristics and end-of-induction MRD status in the discovery and validation cohorts

Patient characteristics	Discovery cohort (n = 2597)			Validation cohort (n = 491)
	MRD ≥ 0.1%	MRD < 0.1%	P a	MRD ≥ 0.1%	MRD < 0.1%	P a
	No. (%)	No. (%)		No. (%)	No. (%)
Sex
Male	278 (19.3)	1166 (80.7)	.11	71 (27.9)	183 (72.1)	.09
Female	194 (16.8)	959 (83.2)		50 (21.1)	187 (78.9)
Age at diagnosis, y
<10	112 (12.6)	777 (87.4)	< .001	28 (16.2)	145 (83.8)	.001
≥10	360 (21.1)	1348 (78.9)		93 (29.3)	225 (70.7)
Genetically defined race
White	243 (17.1)	1181 (82.9)	.12	49 (16.5)	248 (83.5)	.18
Black	18 (14.1)	110 (85.9)		4 (12.1)	29 (87.9)
Hispanic	141 (21.0)	531 (79.0)		23 (21.9)	82 (78.1)
Asian	16 (22.9)	54 (77.1)		3 (25.0)	9 (75.0)
Other	54 (17.8)	249 (82.2)		3 (6.8)	41 (93.2)
WBC at diagnosis, ×109/L
<50	247 (17.0)	1204 (83.0)	.09	73 (27.8)	190 (72.2)	.09
≥50	225 (19.6)	921 (80.4)		48 (21.1)	180 (78.9)
DNA index
<1.16	419 (19.7)	1711 (80.3)	< .001	92 (24.0)	292 (76.0)	.37
≥1.16	53 (11.5)	407 (88.5)		29 (28.7)	72 (71.3)
Total	472	2125	N/A	121	370	N/A

^aAssociations of patient characteristics with end-of-induction MRD status were analyzed by χ² test. MRD = minimal residual disease; N/A = not applicable; WBC = white blood cell count.

Figure 2.

Genome-wide association study (GWAS) of end-of-induction minimal residual disease (MRD) in acute lymphoblastic leukemia. A) GWAS of specimens from patients with end-of-induction MRD. B) GWAS of specimens with MRD level categorized by ordinal analysis. The association between genotype and end-of-induction MRD was evaluated using a logistic regression model for 863 370 single nucleotide polymorphisms (SNPs) in 2597 acute lymphoblastic leukemia cases. Association P values (y axis) are plotted against the respective chromosomal position of each SNP (x axis). Points above the horizontal line indicate SNPs achieving genome-wide significant association (P < 5 × 10^-8). The GATA3 locus is indicated at 10p14.

Figure 3.

Association of the GATA3 SNP rs3824662 genotype with end-of-induction minimal residual disease (MRD). A) Discovery and (B) validation cohorts. Early treatment response measured by MRD at the end of induction was related to genotype at rs3824662 in both the AALL0232 (A) and Children’s Oncology Group P9905 and 6 (B) cohorts, with P values estimated by Spearman rank test of differences in MRD positivity among genotype groups. The A allele was statistically significantly associated with worse treatment response in both cohorts.

We then tested the above 19 SNPs in the validation cohort (491 NCI high-risk ALL patients from COG P9905 and 6). GATA3 SNP rs3824662 again showed a statistically significant association with MRD in both the dichotomous variable and the ordinal variable analyses (P = .003 and P = .01, respectively), whereas other SNPs did not achieve a nominal level of statistical significance of a P value equal to .05 (Supplementary Table 1, available online; Figure 3). Exome-based GWAS of MRD in the discovery cohort identified MOB3C rs12029680 and ANKHD1 rs1051309 as the top loci (Supplementary Figure 1, B, available online). However, neither of these variants showed a statistically significant association in the validation cohort.

Importantly, GATA3 SNP rs3824662 remained statistically significantly associated with MRD after adjusting clinical variables (ie, age, white blood cell count, and DNA index) in the discovery cohort (OR = 1.55, 95% CI = 1.28 to 1.87; P < .001) and also in the validation cohort (OR = 1.51, 95% CI = 1.02 to 2.25; P = .04), as shown in Supplementary Table 2 (available online). The frequency of risk allele A at this SNP varied by race and ethnicity and was highest in Hispanics. (Supplementary Figure 3, available online).

Associations of GATA3 Genotype With Presenting Features and Relapse Risk

In the discovery cohort, the carrier state of GATA3 rs3824662 allele A was statistically significantly associated with a higher cumulative incidence of relapse (hazard ratio [HR] = 1.34, 95% CI = 1.11 to 1.61; P = .002) after adjusting for ancestry and treatment strata (Figure 4). Even when we restricted the analyses to patients with MRD less than 0.1% at end of induction therapy, GATA3 SNP genotype remained prognostic (HR = 1.35, 95% CI = 1.09 to 1.68; P = .007) (Figure 4). Adding age at diagnosis, presenting white blood cell count, and DNA index to the multivariable model, we again observed a statistically significant association of GATA3 rs3824662 with relapse in the entire discovery cohort (HR = 1.22, 95% CI = 1.01 to 1.47; P = .04) and patients with MRD less than 0.1% (HR = 1.27, 95% CI = 1.02 to 1.59; P = .04) (Table 2). The association between GATA3 rs3824662 and relapse was then confirmed in the validation cohort. With adjustment for ancestry and treatment strata, GATA3 rs3824662 allele A conferred a higher cumulative incidence of relapse in the entire validation cohort (HR = 1.40, 95% CI = 1.08 to 1.81; P = .01) and in those with MRD less than 0.1% (HR = 1.45, 95% CI = 1.02 to 2.07; P = .04) (Figure 4). Similar to the observations in the discovery cohort, the prognostic value of GATA3 SNP remained statistically significant even when we added presenting features such as age at diagnosis, white blood cell count, and DNA index in the validation cohort (HR = 1.43, 95% CI = 1.10 to 1.85; P= .007 in the entire cohort, and HR = 1.51, 95% CI = 1.05 to 2.18; P = .03 in the MRD < 0.1% subset) (Table 2). Adding ALL molecular subtypes (ie, ETV6-RUNX1, MLL-rearrangement, E2A-PBX1, B-other) to these multivariable models only slightly diminished the prognostic significance of the GATA3 SNP, suggesting its potential independent effects on ALL treatment outcome (Supplementary Table 3, available online). Interestingly, in a subset of patients in the AALL0232 cohort (n = 684) for whom Ph-like status (somatic) and GATA3 genotype (germline) were both available, the inclusion of Ph-like ALL seemed to diminish the association of rs3824662 with MRD (Supplementary Table 4, available online). Although this observation implies that GATA3 SNP influences MRD at least partly through its association with Ph-like ALL, our sample size is limited to draw definitive conclusions. The GATA3 rs3824662 A allele was associated with older age at diagnosis and nonhyperdiploid ALL (DNA index < 1.16) in the discovery cohort but not with any presenting features in the validation cohort (Supplementary Table 5, available online).

Figure 4.

Cumulative incidence of relapse by GATA3 SNP rs3824662 genotype (AA, AC, or CC). A) Discovery cohort (entire cohort, P =.002 [left]; patients with MRD < 0.1%, P =.007 [right]) and (B) validation cohort (entire cohort, P =.01 [left]; patients with MRD < 0.1%, P =.04 [right]). The cumulative incidence of relapse is plotted by genotype at rs3824662 in the AALL0232 (A) and COG P9905 and 6 (B) cohorts, with P value estimated by the hazard regression test of differences in relapse rate among genotype groups with ancestry and treatment strata as covariates. The A allele was statistically significantly associated with relapse in both cohorts.

Table 2.

Multivariable analysis of GATA3 rs3824662 genotype for association with relapse risk in the discovery and validation cohorts

Patient characteristics	Discovery cohort				Validation cohort
	All patients		Patients with MRD <0.1%		All patients		Patients with MRD <0.1%
	(n = 2289)		(n = 1987)		(n = 455)		(n = 320)
	Hazard Ratios (95% CI)	P a	Hazard Ratios (95% CI)	P a	Hazard Ratios (95% CI)	P a	Hazard Ratios (95% CI)	P a
Age at diagnosis, y, <10 y as reference	2.33 (1.67 to 3.23)	< .001	2.22 (1.49 to 3.22)	< .001	1.08 (0.59 to 1.96)	.80	0.87 (0.34 to 2.22)	.76
WBC at diagnosis, ×109/L, <50 as reference	2.22 (1.64 to 2.94)	< .001	2.43 (1.69 to 3.57)	< .001	1.32 (0.74 to 2.33)	.35	1.10 (0.43 to 2.78)	.84
DNA index,b ≥1.16 as reference	2.03 (1.30 to 3.04)	.001	1.59 (1.02 to 2.48)	.04	2.03 (1.24 to 3.33)	.005	2.05 (1.02 to 4.11)	.04
Ancestry, European as reference
Native American	1.90 (1.15 to 3.16)	.01	1.92 (1.05 to 3.53)	.04	1.76 (0.78 to 3.96)	.17	1.63 (0.49 to 5.37)	.42
Asian	0.75 (0.32 to 1.77)	.51	0.45 (0.12 to 1.61)	.22	1.77 (0.65 to 4.80)	.26	2.12 (0.53 to 8.48)	.29
African	2.71 (1.62 to 4.52) 1.62-4.52	< .001	3.04 (1.72 to 5.36)	< .001	1.72 (0.82 to 3.61)	.15	3.14 (1.29 to 7.62)	.01
GATA3 rs3824662 genotypec	1.22 (1.01 to 1.47)	.04	1.27 (1.02 to 1.59)	.04	1.43 (1.10 to 1.85)	.007	1.51 (1.05 to 2.18)	.03

^aAssociation of SNP genotype with relapse was determined using the Fine and Gray hazard rate regression model. CI = confidence interval; GWAS = genome-wide association study; MRD = minimal residual disease; WBC = white blood cell count.

^bDNA index of ≥1.16 is used to define leukemia hyperdiploidy.

^cHazard ratio is estimated for every additional copy of the risk allele.

Discussion

Published candidate-gene association studies and GWASs of pediatric ALL have identified germline variants associated with disease susceptibility (IKFZ1, PIP4K2A, CEBPE, ARID5B, ETV6, GATA3, CDKN2A, TP63, ELK3, LHPP) (30–40), drug pharmacokinetics or pharmacodynamics (TPMT, NUDT15, ABCC2, ABCC4, SLCO1B1, MTHFR, DHFR, ATF5) (16,41–46), and disease relapse (GSTM1, TYMS, PYGL, PDE4B, ABCB1, IL15, ARID5B) (18,29,38,47,48). Despite the central role of MRD in pediatric ALL risk and treatment stratification, the association between inherited genetic factors and MRD or treatment outcome has not been extensively investigated. Here, we identified a single genome-wide significant risk locus within the GATA3 gene on chromosome 10p14.

MRD-guided therapy has substantially improved the treatment outcome of pediatric ALL. Studies in adult and pediatric ALL indicated that MRD level is a reflection of multiple risk factors, including NCI risk group, leukemic genomic signature (eg, BCR-ABL1 fusion, TCF3-HLF fusion, IKZF1 mutation and/or deletion, CRLF2-rearrangement, MLL-rearrangement) (11), and antileukemic drug disposition (49–53). Inherited germline variants in patients also predict MRD and outcome. Allelic variants at the TPMT locus lead to decreased enzymatic activity and enhanced cytotoxic effects of thiopurines (16). Analysis of patients in the ALL-BFM 2000 protocol confirmed a 2.9-fold reduction in risk for postconsolidation MRD positivity in those patients who were heterozygous at the TPMT locus. Based on a candidate-gene approach to assess the validation cohort, the chemokine receptor 5 genotype was associated with end-of-induction MRD status after adjusting for race, NCI-designated risk group, TEL/AML1 status, and ploidy (54). Taking a GWAS approach, we previously reported intronic variants in IL15 associated with MRD in St Jude and COG cohorts (18). These variants, however, were not related to MRD in the larger AALL0232 or P9905 and 6 cohorts. Many factors could have contributed to the differences in the hits from these 2 GWASs, especially the variation of the ALL treatment regimens and possibly the patient characteristics. For example, the discovery and validation cohorts of the current study are exclusively those of NCI high-risk features, whereas our previous GWAS included both high- and standard-risk patients. Also, 65% of patients in the 2009 study received induction therapy on the St Jude Total XV protocol, which consisted of additional chemotherapeutic agents for a longer period of time (ie, 46 days of induction with 7 drugs) compared with the 4-week induction therapy with 4 drugs in the COG protocols. In fact, the risk allele at IL15 variant rs17007695 identified in our prior study exhibited a trend for association with MRD in the AALL0232 cohort but did not reach statistical significance (P = .08). These studies highlight the role of host pharmacogenomic variants in determining early treatment response.

GATA3 encodes a crucial transcription factor during lymphopoiesis, especially for development and differentiation of T cells (55,56). Enforced GATA3 expression suppresses the maturation of natural killer cells and CD8⁺ T cells (57,58). Growing evidence also indicates that GATA3 is implicated in tumorigenesis and prognosis of breast cancer, Hodgkin lymphoma, T-cell leukemia-lymphoma, and B-ALL (40,59–63). Our top GATA3 SNP in the MRD GWAS, namely rs3824662, has been associated with susceptibility to ALL, particularly Ph-like ALL (33,40,62). The presence of the rs3824662 risk allele is also correlated with an increased risk of relapse and inferior outcome (40). The overrepresentation of Ph-like ALL in patients with GATA3 risk allele raises the possibility that its association with MRD is driven by somatic features related to Ph-like ALL (Supplementary Table 4, available online). Future studies with larger sample sizes are warranted to further investigate these correlations. It should also be noted that the per-allele effect of the GATA3 variant on relapse is lower than certain clinical variables (Table 2), but patients with homozygous genotype at rs3824662 might have a substantial increase in the likelihood of poor survival to justify treatment intensification.

Our study further examined the prognostic value of the GATA3 SNP rs3824662 risk allele by defining it as the only variant associated with end-of-induction MRD with genome-wide significance. Thus, host germline variants should be integrated with other various clinicopathologic factors during upfront treatment stratification for pediatric ALL. It is known that rs3824662 SNP functions as a cis-acting regulatory element of GATA3 transcription (40); however, the exact mechanism that links GATA3 to leukemogenesis and worse treatment response remains elusive. The risk allele is also associated with high GATA3 expression and higher levels of DNase I hypersensitivity at the locus, supporting its role in altering chromatic accessibility and epigenetic regulation (40,64). Transcriptome analysis suggested that activating the JAK/STAT pathway in the GATA3 regulatory network is a plausible mechanism linking GATA3 overexpression, leukemogenesis, and now treatment response (64). A possible role of GATA3 in L-asparaginase resistance was also suggested in a previous study linking leukemia gene expression profiles to drug sensitivity in vitro (65).

One limitation of our study is that only a minority of patients (approximately 25%) have both comprehensive somatic and germline genomic data available. Hence, with our limited sample size, this study is not powered to draw definitive conclusions about how the association of Ph-like ALL and GATA3 SNP influences MRD. These limitations can be overcome in future studies of larger cohorts for whom we also perform somatic profiling.

Through performing GWAS on large pediatric frontline ALL trials, we identified GATA3 SNP rs3824662 as an inherited variant that statistically significantly predicted MRD positivity at the end of induction therapy and subsequent disease relapse. These findings may be of value in ALL risk stratification and warrant further mechanistic study to understand the role of GATA3 in determining treatment resistance in pediatric ALL.

Funding

This work is supported by US National Institutes of Health Grants No. CA21765, CA98543, CA114766, CA98413, CA180886, CA180899, CA197695, CA11476, GM92666, GM115279, and GM097119; and the American Lebanese Syrian Associated Charities. HZ is supported as a St Baldrick’s International Scholar (581580). SHRL is supported by National Medical Research Council Singapore Research Training Fellowship (003/008-258).

Notes

Role of the funder: The study sponsors were not directly involved in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript.

Disclosures: CGM receives research funding from Abbvie and Pfizer, and consulting fees for Illumina. The other authors declare no conflict of interest related to this work.

Author contributions: JJY is the principal investigator of this study, has full access to all the study data, and takes responsibility for the integrity of the data and the accuracy of the data analysis. HZ, AP-YL, SHRL and JJY wrote the manuscript; HZ, XC, DP, WY, CC performed data analysis; MD, MB, BW, JMG-F, YD, ER, EL, NW, WPB, SK, PLM, WLC, C-HP, CGM, WEE, SPH, MVR, MLL contributed reagents, materials, and/or data; HZ, AP-YL, CC, JJY interpreted the data and the research findings. All the coauthors reviewed the manuscript.

Data availability

The GWAS data are deposited in the NIH dbGAP (https://www.ncbi.nlm.nih.gov/gap/) under phs000638.v1.p1 and phs000637.v1.p1.