Germline genetic variation and treatment response on CCG-1891

Abstract

Background

Recent studies suggest that polymorphisms in genes encoding enzymes involved in drug detoxification and metabolism may influence disease outcome in pediatric acute lymphoblastic leukemia (ALL). We sought to extend current knowledge by using standard and novel statistical methodology to examine polymorphic variants of genes and relapse risk, toxicity, and drug dose delivery in standard risk ALL.

Procedure

We genotyped and abstracted chemotherapy drug dose data from treatment roadmaps on 557 patients on the Children’s Cancer Group ALL study, CCG-1891. Fourteen common polymorphisms in genes involved in folate metabolism and/or phase I and II drug detoxification were evaluated individually and clique-finding methodology was employed for detection of significant gene-gene interactions.

Results

After controlling for known risk factors, polymorphisms in four genes: GSTP1*B (HR=1.94, p=0.047), MTHFR (HR=1.61, p=0.034), MTRR (HR=1.95, p=0.01), and TS (3R/4R, HR=3.69, p=0.007), were found to significantly increase relapse risk. One gene-gene pair, MTRR A/G and GSTM1 null genotype, significantly increased the risk of relapse after correction for multiple comparisons (p=0.012). Multiple polymorphisms were associated with various toxicities and there was no significant difference in dose of chemotherapy delivered by genotypes.

Conclusions

These data suggest that various polymorphisms play a role in relapse risk and toxicity during childhood ALL therapy and that genotype does not play a role in adjustment of drug dose administered. Additionally, gene-gene interactions may increase the risk of relapse in childhood ALL and the clique method may have utility in further exploring these interactions. childhood ALL therapy.

Introduction

Although the majority of children treated for acute lymphoblastic leukemia (ALL) are cured, disease recurrence and treatment related toxicities remain important, and often unpredictable, clinical issues. Recently, evaluation of multiple biologic characteristics of both the patient and the blast cells has improved prediction of treatment outcomes.[1] A number of studies have shown that response to chemotherapeutic agents, as well as drug-related toxicity, can vary with polymorphism frequency.[2–4] Most of the work in this area has centered on single-nucleotide polymorphisms (SNPs) on the genes encoding thiopurine methyltransferase (TPMT), thymidylate synthase (TS), and methylenetetrahydrofolate reductase (MTHFR).

We sought to validate and extend these previous pharmacogenetic analyses by examining the relationship between germline genetic polymorphisms, chemotherapy dose delivered, disease relapse, infection, and toxicity risks in the Children’s Cancer Group trial CCG-1891, a phase III trial for standard risk ALL. We have previously shown in this sample set that the MTHFR C677T allele was associated with a statistically significantly increased risk of relapse.[5] This result, along with the work of other investigators, led us to undertake a more comprehensive genotyping and statistical analysis to determine the role of host genetic polymorphisms in determining treatment response.

While prior studies, including our own, have evaluated odds ratios, relative risks, and hazard ratios associated with genotype status, these metrics have well described limitations in the clinical prediction of treatment outcomes.[6] Specifically, clinicians treating patients with ALL need to know the probability of relapse for a given genotype rather than an odds ratio which describes the relative increase in risk. To address this gap in research knowledge, we estimated the positive predictive value (PPV) for relapse for all genotypes under consideration.

Furthermore, prior investigations have evaluated gene-gene interactions that would determine relapse and toxicity risk. However, these searches have not employed an exhaustive method that would evaluate all possible interactions while appropriately controlling for multiple comparisons. We sought to address this gap by applying the clique methodology with appropriate control for multiple comparisons to comprehensively evaluate all two-way gene-gene interactions in our data.[7] While multiple methods exist for searching for gene-gene interactions including classification and regression trees (CART)[8], multi-factor dimensionality reduction[9], random forests[10], support vector machine[11], neural networks[12] and others, these approaches have important limitations. The clique methodology has been shown to be superior to CART in simulation studies[7], and is less sensitive to assumptions of underlying biologic model and hierarchy. Moreover, the clique methodology accommodates heterogeneity in the etiology of the outcome of interest.

For these analyses, we obtained DNA samples from children with standard-risk ALL in a well-defined cohort in a Children’s Cancer Group study and abstracted 90,000 drug administration doses from patient roadmaps. We analyzed 14 candidate genes that were selected based on their known involvement in either folate metabolism or Phase I and II drug detoxification. We hypothesized that germline genetic variation would be associated with relapse risk and that we would be able to determine the PPV for genotypes of interest and determine relevant gene-gene interactions using the clique methodology.

Materials and Methods

Patient Population and Study Design

CCG-1891 enrolled 1204 NCI standard risk ALL patients in a three-armed randomized trial that has been previously reported.[13] Patients were treated with modified Berlin-Frankfurt-Munster (BFM) therapy and were randomized at diagnosis to a single delayed intensification, two delayed intensifications, or a single delayed intensification with compressed vincristine/prednisone pulses in maintenance. The doses of mercaptopurine (6-MP) and methotrexate (MTX) prescribed were identical among treatment arms. At time of study enrollment, patient guardians gave permission for the use of archived samples in future studies. The Children’s Oncology Group (COG) scientific council, CTEP, the Children’s Hospital of Philadelphia Institutional Review Board and the University of Pennsylvania Institutional Review Board approved this research.

A nested case-control design was employed as the study design. The primary outcomes of interest were disease relapse, occurrence of toxicity, and infectious complications. For relapse risk analyses, patients were considered cases if they experienced disease relapse at any site and controls if they remained in continuous remission as defined by < 5% blasts on bone marrow aspirate or absence of documented extramedullary disease. For toxicity and infection analyses, cases were defined as patients who experienced the toxicity or infection under analysis and controls were patients who did not experience the toxicity or infection under analysis. All patients with available genotype and toxicity data were included in the analyses.

Sample and Data Collection

Data on patient and leukemia characteristics, toxicities, and infectious complications were collected prospectively by local institutions using CCG case report forms. Data on disease relapse are reported in an ongoing manner. Toxicity and infection data were graded according to the CCG toxicity criteria which are similar to CTC v 2.0. These included, but were not limited to, grade III and IV central nervous system (CNS), diarrhea, hyperbilirubinemia, peripheral neuropathy, and transaminitis. Infectious complications included fever with neutropenia, sepsis, interstitial pneumonia, and other infections. Data on 6-MP and MTX doses were abstracted from treatment roadmaps and entered into a relational database in MS Access. The relational database was programmed with data validation rules and all recorded drug doses were checked against the paper records by a second data entry operator.

Laboratory Analyses

The COG Biopathology Center provided marrow and peripheral blood slides on patients. DNA extraction was performed in a dedicated DNA extraction area, as previously reported. Each extraction was performed with a separate set of gloves and consumable items and the extraction area was cleaned with 70% ethanol between extractions. Genotyping was performed with different methodologies: DHFR, TS-6 bp insertion, MTRR, MTHFR, RFC, TPMT, C3A4, C3A5, GSTP, and UGT genotypes were assayed on the Pyrosequencing platform, SHMT genotypes were assayed by direct sequencing, and TS-28 bp insertion, GSTT1, and GSTM1 were assayed by standard RFLP techniques.[3,14] All primer sequences are available from the corresponding author. The technician performing the assays was blind to the case/control status of the sample. All PCR reactions were set up in a dedicated PCR area with dedicated pipettes and reagents. All RFLP digest and Pyrograms were reviewed by 2 individuals who were blind to the case/control status of the sample. All genotype calls were checked again prior to statistical analysis.

Statistical Analysis

The primary outcomes of interest were disease relapse, occurrence of toxicity, and infectious complications. All patients with genotype data were analyzed. Univariate analyses were performed with standard summary statistics and a chi-squared test. A t-test was used to compare drug dose delivered between groups of interest. Relapse free (RFS) survival was estimated by the Kaplan-Meier method and the log-rank test was used to compare time to relapse. Cox proportional hazards modeling accounting for clinically important covariates (age, gender, ethnicity, and day 7 bone marrow response) was used to model relapse risk.

Time dependent positive and negative predictive values (PPV, NPV) for genotypes were estimated as described by Moskowitz and Pepe.[15,16] Statistical significance for PPV and NPV comparisons was determined by bootstrapping simulations performed in STATA version 10 (STATA Corporation, College Park, TX).

Clique analysis was performed as previously described.[7] In brief, the analysis evaluated the effect of all two-way and three-way genotyping combinations among the 14 SNPs studied. For each set, an odds ratio was computed that estimated the risk associated with that particular SNP combination. The 10 combinations with the largest protective and adverse effects were ranked, and the pathways contained in these patterns were examined. Correction for multiple comparisons was performed as described by Benjamini and Hochberg.[17]

Results

Table I presents the characteristics of all patients on CCG-1891. There were 557 genotyped patients and 647 patients who did not have DNA available for genotyping. The 557 patients who were genotyped were representative of the entire study population. Specifically, there were no differences in age, gender, treatment arm, and day 7 bone marrow status between genotyped and non-genotyped patients. African-American patients were over-represented in the patients experiencing relapse. However, the frequency of other ethnic groups was similar between patients with and without relapse.

Table I.

Patient characteristics

Age	Not Genotyped		Genotyped
	4 years		4 years
Gender
Male	437	67.5%	381	68.4%
Female	210	32.5%	176	31.6%
Ethnicity
White	494	76.4%	420	75.4%
Black	16	2.5%	31	5.6%
Hispanic	99	15.3%	67	12.0%
Asian	14	2.1%	13	2.3%
Other	24	4.9%	26	6.2%
Regimen
A	226	34.9%	180	32.3%
B	216	33.4%	186	33.4%
C	205	31.7%	191	34.3%
Day 7 BM
M1	295	45.7%	253	45.4%
M2	133	20.6%	130	21.9%
M3	140	21.9%	122	10.8%
Unknown	78	11.8%	52	10.8%
Relapse
Bone marrow	44	47.8%	77	58.8%
CNS	36	39.1%	43	32.8%
Testicular	9	9.8%	10	7.6%
Other	3	3.3%	1	0.8%

Supplemental Table I summarizes overall and ethnic group genotype frequencies. Observed genotype frequencies were consistent with previously reported frequencies and were in Hardy-Weinberg equilibrium.

Genotype, relapse free survival estimates, and Cox proportional hazards for the most significant polymorphisms are presented in Table II. The complete list of relapse free survival estimates by genotype is found in Supplemental Table II. After adjustment for known prognostic factors, one gene (GSTP1) responsible for Phase II drug metabolism was associated with a statistically significantly increased hazard rate of leukemia relapse. Patients who were homozygous for the GSTP variant allele showed a significant decrease in RFS as compared to patients with the wild- type allele (HR=1.94; 95% CI, 1.010–3.732; P= .047). Three genes (MTHFR, MTRR, TS) involved in folate metabolism were also associated with a decreased RFS: MTHFR C677T (HR=1.61; 95% CI, 1.036–2.505; P=.034), MTRR G/A (HR=1.95: 95% CI, 1.175–3.227, P=.01) and TS 3R/4R (HR=3.69; 95% CI, 1.436–9.481; P=.007). Furthermore, the association of two genes (GSTT1 and TPMT) approached, but did not reach, statistical significance after correction for multiple testing.

Table II.

Genotype frequencies and relapse free survival^*

Gene	Genotype	Total N	Relapse N	RFS	95% CI	HR*	95% CI HR*	P value*
TS 28 bp insertion	2R/2R	83	15	0.80	.691–.877	1	reference	--
	2R/3R	196	38	0.79	.726–.845	1.68	0.863–3.255	0.13
	3R/3R	103	27	0.73	.631–.806	1.87	0.942–3.721	0.074
	3R/4R	20	7	0.60	.342–.789	3.69	1.436–9.481	0.007
MTHFR C677T	C/C	224	41	0.81	.747–.854	1	reference	--
	C/T	187	46	0.74	.667–.798	1.61	1.036–2.505	0.034
	T/T	72	20	0.70	.578–.798	1.36	0.763–2.411	0.3
MTRR	A/A	144	26	0.80	.720–.860	1	reference	--
	G/A	225	62	0.72	.650–.770	1.95	1.175–3.227	0.01
	G/G	128	23	0.81	.724–.867	1.23	0.672–2.241	0.51
GSTP1*B	A/A	147	26	0.81	.738–.869	1	reference	--
	G/A	176	42	0.75	.678–.809	1.47	0.889–2.442	0.13
	G/G	55	16	0.68	.529–.789	1.94	1.010–3.732	0.047
TPMT (460 and 719)	Normal	465	98	0.78	.733–.812	1	reference
	Heterozygote	31	9	0.69	.486–.828	1.86	.932–3.725	0.078
	Variant	1	1	--	--	--	--	--
GSTT1	Present	433	100	0.76	.712–.794	1	reference	--
	Null	65	9	0.85	.740–.922	0.54	0.271–1.078	0.081

^*Adjusted for age, gender, ethnicity, and day 7 bone marrow response

Analysis of genotype positive predictive value was performed for all genotypes that were significantly associated with relapse risk. Figure I illustrates the PPV for the TS 28 bp insertion which had the largest effect size (HR = 3.69) of the significantly associated genotypes. The positive predictive value of the TS 28 bp genotype was not statistically significant (p > 0.5). That is, knowledge of the TS 28 bp genotype did not significantly improve the prediction of relapse risk over the baseline prediction ability based on well-established risk factors for relapse (Figure 1).

Figure 1.

Positive predictive value of TS 28bp insertion and risk of relapse.

The analysis of genotypes associated with altered leukemia-free survival and non-hematologic and infectious toxicities are presented in Table III. All toxicity data that was reported for patients enrolled on CCG1891 was collected. However, analysis was limited to those toxicities believed to be related to drugs whose metabolism may have been affected by those polymorphisms under investigation.

Table III.

Toxicities by genotype

Gene	Genotype	n	N with toxicity	toxicity	OR	95% CI	P value
TPMT (416 and 719)	Normal	465	304	Fever/Neutropenia	1		0.0051
	Heterozygote	31	27	Fever/Neutropenia	3.85	1.307–11.320
	Variant	1	1	Fever/Neutropenia
TPMT (416 and 719)	Normal	465	264	Infection	1		0.015
	Heterozygote	31	24	Infection	2.73	1.147–6.515
	Variant	1	0	Infection
TPMT (416 and 719)	Normal	465	179	Transaminitis	1		0.0004
	Heterozygote	31	3	Transaminitis	0.17	0.051–0.568
	Variant	1	0	Transaminitis
MTHFR C677T	C/C	224	78	Transaminitis	1		0.019
	C/T	187	83	Transaminitis	1.52	1.015–2.271
	T/T	72	21	Transaminitis	0.7	0.388–1.272
TS 28 bp insertion	2R/2R	83	8	Neuropathy	1		0.015
	2R/3R	196	13	Neuropathy	0.64	0.261–1.591
	3R/3R	103	1	Neuropathy	0.09	0.012–0.768
	3R/4R	20	1	Neuropathy
GSTM1	Present	185	78	Transaminitis	1		0.021
	Null	311	99	Transaminitis	0.64	0.439–0.935
UGT1A1 TA insertion	6/6	224	7	Hyperbilirubinemia	1		0.045
	5/6	5	1	Hyperbilirubinemia
	5/7	1	0	Hyperbilirubinemia
	6/7	206	7	Hyperbilirubinemia	0.79	0.292–2.146
	6/8	2	0	Hyperbilirubinemia
	7/7	62	7	Hyperbilirubinemia	3.27	1.154–9.256
	7/8	3	2	Hyperbilirubinemia
GSTT1	Present	433	18	Hyperbilirubinemia	1		0.047
	Null	65	7	Hyperbilirubinemia	2.74	1.079–6.984

The MTHFR C677T variant allele showed an increased risk of transaminitis as compared to the wild-type allele (OR=1.52; 95% CI, 1.015–2.271; P=.019). Patients who were heterozygous for TPMT variants were found to have an increased risk of fever and neutropenia (OR=3.85; 95% CI, 1.307–11.320; P=.0051) and infection (OR=2.73; 95% CI, 1.147–6.515; P=.015). In addition, the UGT1A1 TA insertion polymorphism was associated with an increased risk of hyperbilirubinemia with an OR = 3.27 (95% CI 1.154–9.256, p = 0.045) for the UGT1A1 7/7 TA insertion. Other significant associations were observed for transaminitis with RFC-1 and GTSM1, and infection with the TS-6 bp insertion.

Data on the percent of expected chemotherapy dose actually prescribed throughout maintenance is presented in Table IV. Delivered doses of methotrexate range between 69.3 and 77.8% and those of 6-MP were 71.8–81.2% of prescribed doses. The differences in exposure to 6-MP and MTX are relatively modest for all genotypes studied. In contrast, the delivered doses of prednisone and vincristine were greater than 90% of prescribed doses.

Table IV.

Average chemotherapy exposure by genotype

		Methotrexate		Prednisone		6-MP		Vincristine
		Mean	sd	Mean	sd	Mean	sd	Mean	sd
TS 28 bp insertion	2R/2R	74.4	18.6	97.5	3.5	76.3	15.4	96.4	8.0
	2R/3R	73.2	18.3	95.6	5.9	74.4	15.9	94.8	9.8
	3R/3R	76.1	18.6	96.1	7.8	79.0	16.5	96.2	11.9
	3R/4R	77.8	17.2	96.0	5.2	81.2	11.8	96.1	4.6
MTHFR C677T	C/C	73.4	18.8	95.9	5.2	75.3	16.1	95.4	9.7
	C/T	75.4	18.9	96.5	5.1	77.2	15.9	98.5	23.2
	T/T	71.8	20.0	95.1	8.7	74.1	16.5	93.4	15.3
MTRR	A/A	76.1	18.7	96.3	5.6	76.6	17.0	96.7	7.3
	G/A	73.7	18.8	95.2	7.8	75.6	15.4	96.5	22.5
	G/G	72.7	18.3	96.5	5.1	75.8	15.7	94.7	12.7
GSTP1*B	A/A	73.2	17.3	95.8	7.3	76.5	14.1	96.0	7.9
	G/A	74.2	18.7	95.5	7.2	74.8	16.0	96.5	25.7
	G/G	75.8	15.8	96.5	4.9	78.0	13.7	95.5	10.8
TPMT (416 and 719)	Normal	73.9	18.2	96.1	6.8	75.9	15.6	96.5	18.7
	Heterozygote	69.3	21.1	94.3	6.1	71.8	17.5	91.2	14.0
	Variant	--	--	--	--	--	--	--	--
GSTT1	Present	74.3	18.3	96.0	6.7	75.9	15.7	96.2	18.1
	Null	71.9	21.5	95.8	5.7	75.9	16.5	96.3	5.4

Clique analysis revealed multiple gene combinations that were associated with altered relapse risk. After correction for multiple comparisons and collapsing across minimum support levels, only the combination of the MTRR A/G and GSTM1 null genotype two-way gene-gene interaction was associated with a significantly increased risk of relapse, p = 0.012 at a minimum support level of 30.

Discussion

Although the current treatment regimens available for pediatric ALL provide a survival of greater than 5 years for the majority of children diagnosed with the disease, approximately 15–20% of patients will relapse.[1] Genetic variations in certain key enzymes in chemotherapy metabolism may play a key role in determining relapse risk and treatment related toxicity.[18] Our results confirm previous reports, particularly the impact of TS on relapse risk, and extend our knowledge of the role of other polymorphisms and gene-gene interactions on childhood ALL.

Our data confirm the report of Krajinovic et al. that the TS 28 bp genotype is associated with altered relapse risk.[3] Despite a hazard ratio of 3.7, the TS 28 bp genotype did not significantly improve risk prediction. Despite the lack of statistical significance, the analysis of positive predictive value is informative in several ways. First, the point estimate for the TS 28 bp 3R/4R PPV for relapse is approximately 40%, which is approximately twice the baseline risk for all patients on study. Moreover, the positive predictive value of relapse is much more informative for discussions of altering treatment than the hazard ratio which does not provide any direct information on the absolute magnitude of risk for a group of patients. In this way, this result exemplifies the limitations of standard analyses using measures of association to identify clinically applicable prognostic factors.[19,20] For these reasons, we suggest that the Moscowitz-Pepe methodology for PPV estimation should be included in genotype analyses that evaluate the clinical applicability of germline genetic variation.

Our data are consistent with prior observations that patients with TPMT variants are at increased risk for hematopoietic and infectious complications from ALL therapy.[21] While a statistically significant association between relapse risk and TPMT heterozygous genotype was not observed, the p-value approached statistical significance, p = 0.078. This result is similar to the report from Stanulla et al.[22], who demonstrated an increased risk of minimal residual disease in TPMT heterozygous patients. The absence of statistical significance in our analysis may be due to the expectedly small number of patients with TPMT variant genotypes.

As reported previously by our group and others, these data support the role of MTHFR polymorphisms in determining treatment related outcomes.[5,23,24] Although other investigators have not demonstrated an association between methionine synthase reductase (MTRR) and ALL treatment response,[25,26] we observed a significant association between the MTRR G66A variant allele and relapse risk. This association may be a false positive, or may be a true positive that was not detected by other groups because of more modest study sample sizes.

Other groups have demonstrated conflicting results for the role of GSPT1 variants and ALL treatment response. Krajinovic and Gatedee found no association between GSTP1 genotype and response, while Stanulla et al. reported a decreased risk of relapse that approached statistical significance in patients with the GSTP1 Val/Val genotype.[27–29] Our data support the hypothesis that GSTP1 variants may alter relapse risk, although the direction of effect observed in our data is opposite in direction from that of Stanulla et al.[28] The reason for this difference is unclear, although the sample size from CCG 1891 was substantially larger than the 64 cases and controls reported by the BFM group. Furthermore, differences in therapy among the various patient populations analyzed across the studies may be a factor in the varied findings.

There are multiple methods that can be employed in order to evaluate for gene-gene interactions. While the optimal method is not yet known, we used the clique method which has been used by other research groups extensively. We evaluated second order gene-gene interactions to facilitate the interpretation of results and minimize computational requirements. These analyses discovered one gene-gene interaction that had a substantial effect size and retained statistical significance after correction for multiple comparisons. The interpretation of the MTRR and GSTM1 interaction is not intuitively clear. Neither showed significant effect on univariate analysis and they involve different drug detoxification and metabolic pathways. Thus, this result may be a false positive despite appropriate correction for multiple comparisons. However, the significance after correction for multiple comparisons indicates that this gene-gene interaction should be tested in other sample sets, and if validated, may lead to a better understanding of gene-gene interactions in determining treatment response in ALL.

We also evaluated the effect of various genotypes on the effect of cumulative chemotherapy dose delivered to the patients. Consistent with clinical experience, prednisone and vincristine doses were not substantially adjusted while methotrexate and 6-MP had 20–25% dose reductions in cumulative exposures. While this result is not unexpected, to our knowledge this is the first comprehensive report of chemotherapy dose adjustments in a large cooperative group clinical trial. Interestingly, dose adjustment was not associated significantly with genotype, although no patients had homozygous variant genotypes for TPMT. These results indicate that dose adjustment and the genotypes assessed in this study are not correlated. This provides further confirmation of the results seen with TS 28 bp positive predictive value analysis, namely that statistically significant genotype-outcome associations can be difficult to translate into clinical practice.

This study has several strengths that support the validity of these findings. This was a national prospective clinical trial of standard-risk ALL conducted through the Children’s Cancer Group. As such, patients were on an identical treatment protocol and received standardized supportive care per uniform care guidelines. Moreover, since all patients were enrolled on a CCG trial, the chance of reporting and detection biases are likely less than other study populations. Furthermore, to the best of our knowledge, this is the first study reporting the effect of genotype on relapse risk and drug toxicity with extensive data on drug dose prescribed.

This study also has limitations. We did not genotype all of the functional polymorphisms in all of the genes in the folate drug metabolism pathways. Thus, the candidate genes studied here do not account for all of the genetic variation in folate metabolism. Unfortunately, the limited amount of available DNA prevented more comprehensive genotyping. Furthermore, while substantive, the sample size was not large enough to detect modest effect sizes or to provide very narrow confidence intervals around relapse-free survival estimates. As in any pharmacogenetic study, false positive findings are a concern. We sought to minimize the risk of false positives by limiting the number of subgroup analyses that were performed (e.g. not evaluating CNS versus testicular relapse) and correcting for multiple comparisons in the gene-gene interaction analysis. Likewise, we addressed the issue of population stratification by including ethnicity in the Cox proportional hazards models.

In summary, this report provides additional data supporting the impact of germline genetic variation in ALL relapse and toxicity risk. Our study confirms several associations seen by other groups, and this confirmation lends support to the overall validity of our findings. Furthermore, our report extends the current state of pharmacogenetic research in pediatric ALL in several ways. First, we have applied novel statistical methods to evaluate the clinical utility of incorporating genotype data in treatment stratification. Such evaluation will be critical in the establishment of the clinical utility of genetic variation data. Second, we have demonstrated that the clique analysis can identify gene-gene interactions that withstand correction for multiple comparisons. Additional studies will be needed to confirm these findings, and to determine how germline genotype data may be more broadly used in pediatric ALL therapy.