Genomic variants of cytarabine sensitivity associated with treatment related mortality in pediatric AML: A report from the Children’s Oncology Group

Abstract

Purpose:

Cytarabine is an effective treatment for AML with associated toxicities including treatment related mortality (TRM). The purpose is to determine the clinical relevance of single-nucleotide polymorphisms (SNPs) identified through the use of HapMap lymphoblastoid cell-based models, in predicting cytarabine response and toxicity in AML.

Experimental Design:

We tested clinical significance of SNPs associated with cytarabine sensitivity in children with AML treated on Children’s Oncology Group regimens (CCG 2941/2961). Endpoints included overall survival (OS), event free survival (EFS) and TRM. Patients who received bone marrow transplant were excluded. We tested 124 SNPs associated with cytarabine sensitivity in HapMap cell lines in 348 children to determine if any associated with treatment outcomes. In addition, we tested five SNPs previously associated with TRM in children with AML in our independent data set of 385 children.

Results:

Homozygous variant genotypes of rs2025501 and rs6661575 had increased in vitro cellular sensitivity to cytarabine and were associated with increased TRM. TRM was particularly increased in children with variant genotype randomized to high dose cytarabine (rs2025501: p= 0.0024 and rs6661575 p=0.0188). In analysis of previously reported SNPs, only the variant genotype rs17202778 C/C was significantly associated with TRM (p<0.0001).

Conclusions:

We report clinical importance of two SNPs not previously associated with cytarabine toxicity. Moreover, we confirm that SNP rs17202778 significantly impacts TRM in pediatric AML. Cytarabine sensitivity genotypes may predict TRM and could be used to stratify to standard vs. high dose cytarabine regimens, warranting further study in prospective AML trials.

INTRODUCTION:

Pediatric acute myeloid leukemia (AML) accounts for 15%−20% of childhood leukemia, but is responsible for the majority of pediatric deaths caused by leukemia. (1) Intensification of chemotherapy over the past several decades has resulted in improvements in survival; however, the five year survival remains only about 50–60%. (2, 3) Intensification of dose and aggressive timing of induction therapy, as well as appropriate use of transplant, have been employed to cure a higher percentage of patients with AML. While increasing the success of therapy for some, this approach has resulted in increased toxicity in contemporary regimens. Other children still do not achieve cure due to failure to achieve remission or early relapse of the disease. The 5-year event-free survival (EFS) and overall survival (OS) were 42% ± 3% and 52% ± 4%, respectively, on Children’s Cancer Group study, CCG 2961, the study from which the patient samples discussed further in this manuscript were obtained.(2) Importantly, the overall treatment related mortality (TRM) was significant, ranging from 9 ± 4% to 17 ± 5% depending on treatment arm, largely due to infection secondary to neutropenia and mucosal barrier injury.

The backbone of contemporary AML therapy is cytarabine plus an anthracycline. (4–6) Cytarabine is currently the single most effective agent in the treatment of acute myeloid leukemia and failure to respond to cytarabine is associated with poor survival. (7–11) In one study, patient leukemia samples at diagnosis which demonstrated in vitro resistance to cytarabine predicted a four-fold increased risk of relapse. (12)

The dose of cytarabine that is used is based on a population average, however a subset of patients will experience adverse reactions while another subset will have an inadequate therapeutic effect from the drug. The potential benefit of implementing pharmacogenetics to identify patients at greatest risk of nonresponse or serious toxicity would have enormous benefit in pediatric AML. Cure is dependent on the ability to achieve remission of the leukemia. Only 30–50% of children with relapsed disease will achieve a second remission and have a chance at cure. (6) The development of a genetic model that can be used to predict inter-individual variability to chemotherapy response will allow upfront tailoring of therapy to reduce acute and chronic toxicities and cure a larger number of patients.

MATERIALS AND METHODS

Identification of SNPs to be tested

We used two different approaches to identify SNPs that would be tested for association with outcome in pediatric AML patients (Figure 1). Our first approach used a set of SNPs first identified in a genome-wide association study (GWAS) of cellular sensitivity to cytarabine in 85 European (CEU) lymphoblastoid cell lines (LCLs). (13) SNPs (n= 124) were selected from the GWAS based on their strong association with cytarabine sensitivity phenotype (p </= 1 × 10⁻⁴) and minor allele frequency in CEU population >5%. Appendix A lists the SNPs that were included in the first approach.

Figure 1: Study Design.

Shaded areas represent previously published data used to identify SNPs for testing, unshaded boxes represent data presented in this paper. The CEU1/2 population consisted of 85 lymphoblastoid cell lines as described in Hartford et al., (13) and the meta-analysis SNPs associated with cytarabine were obtained from 523 lymphoblastoid cell lines as described in Gamazon et al., (14).

CEU: refers to HapMap population samples of Northern and Western European descents; YRI: Yoruba, refers to HapMap population samples from Ibadan, Nigeria; ASW: refers to population African American of Southwestern United States; ASN: refers to HapMap population samples of Asian descent

In our second approach we focused on a small number of SNPs identified using multiple HapMap LCL populations followed by genotyping in a single institution study and found to be significant. (14) We selected five SNPs described as modifying risk of TRM in children with AML, to determine whether they replicated in this entirely new dataset.

Clinical Samples

The study cohort consists of 455 patients with de novo AML treated on Children’s Cancer Group (CCG) protocols 2941 and 2961 between 1995 and 2002. Clinical data, including age, sex, white blood cell count (WBC) at diagnosis, race, presence of chloromas, CNS status, immunophenotyping and cytogenetic analysis were collected prospectively. Cases were reviewed by central pathology and were classified by the French-American-British (FAB) Cooperative Study Group criteria. All AML subtypes, except APL, were eligible and treated on the same chemotherapy regimens.(2) Patients or legal guardians provided written consent to enrollment on the therapeutic studies after approval of the study by the IRB of each participating institution, and to submission of samples for biologic studies. Studies were conducted in accordance with recognized ethical guidelines according to Declaration of Helsinki. Patients who received bone marrow transplant (BMT) were omitted from the current analysis of EFS/OS/TRM as the transplantation confounds the outcome being studied. Stored blood samples from study patients, who gave permission for storage and future use at time of initial consent, were used for genotyping. After exclusion of patients who received BMT, SNPs derived from CEU cell lines in the first approach using OpenArray were genotyped in 348 patients. In our second approach, 385 non-BMT patient samples were genotyped for the five SNPs previously linked to TRM (Figure 1). (14) Some patient samples did not have sufficient amount of concentrated DNA to be analyzed via OpenArray platform leading to the variation in sample numbers between the two approaches.

Chemotherapy Treatment Regimen

The study design of CCG 2961 is shown in Figure 2. All patients received intensively timed induction therapy with IDA-DCTER (idarubicin, dexamethasone, cytarabine, thioguanine, etoposide and daunomycin) given on days 0 to 3 followed by DCTER (dexamethasone, cytarabine, thioguanine, etoposide and daunomycin) given on days 10 to 13 for total dose of cytarabine 1600mg/m². On recovery of white blood cell and platelet counts, patients were randomly assigned to receive consolidation therapy with Regimen A i.e. DCTER (1600 mg/m² cytarabine) or Regimen B i.e. IDA-FLAG (idarubicin, fludarabine, cytarabine and G-CSF) with total dose of cytarabine at 7590 mg/m². Intrathecal (IT) cytarabine was used for central nervous system (CNS) prophylaxis. Following consolidation, patients with a matched related donor proceeded to allogeneic marrow transplant (BMT) with ablative conditioning (busulfan and cytoxan). Those without a related donor received intensification with high-dose cytarabine (24,000 mg/m²), L-asparaginase (Capizzi II) and additional IT cytarabine. Patients on the Capizzi II arm were further randomized to receive Interleukin-2 or standard follow-up. BMT recipients were excluded from this randomization. (2) This study was suspended due to unacceptably high TRM from September 1999 to May 2000, and reopened after amendments to incorporate multiple changes including mandatory hospitalization for neutropenia, use of broad-spectrum antibiotics including vancomycin with fever, empiric anti-fungal therapy. (2) CCG 2941 was a pilot study of intensively timed IDA-DCTER/DCTER and patients were non-randomly assigned to treatment as shown for regimen A. (15) Results comparing treatment arms of IDA-DCTER and IDA-FLAG are therefore limited to patients on 2961.

Figure 2: Clinical trial CCG 2961 design.

Design schema of chemotherapy backbone CCG 2961 with cumulative doses of cytarabine in bold.

Genotyping

We used Applied Biosystems Taqman OpenArray technology to genotype the samples. A small number of SNPs were unsuitable for OpenArray analysis and were genotyped using the fluorogenic PCR-based allelic discrimination (single TaqMan) assay using the PE Applied Biosystems 7200 Sequence Detection System.

Statistical Analyses

Clinical data were collected prospectively, audited and error checked and provided for analysis by the statistical staff of the Children’s Oncology Group. Collection of follow-up data were completed for pilot study CCG-2941 on 9/1/05 and 1/1/11 for CCG-2961. The median follow-up time was 3609 days (9.9 years), [range 0–4717 days (0–12.9 years)]. Differences in induction outcome, dichotomized into complete remission or no remission by genotype, were tested with Pearson’s chi square statistic or Fisher’s exact test when data were sparse. OS was defined as the time from study entry until death. EFS was defined as a time from study entry until induction failure, relapse or death. Survival estimates of OS and EFS were reported using the Kaplan and Meier method.(16) Differences in OS and EFS were evaluated using the log-rank statistic. TRM and relapse rate (RR) were determined using methods that account for competing events (17). TRM was defined as time from study entry to death from non-progressive disease where relapses were treated as competing events. RR was defined as time from study entry to relapse where deaths from non-progressive disease were treated as competing events. Gray’s test was used to test differences between groups for RR and TRM. All estimates at 1 year were reported with their corresponding standard error. A multivariate competing risks regression was performed to assess the impact on TRM of the significant SNPs. Analyses comparing patients randomized to IDA-DCTER or IDA-FLAG were limited to the CCG-2961 study.

Evaluation of genomic location and potential functional significance of SNPs. Statistically significant SNPs were evaluated for functional significance through expression quantitative trait loci (eQTLs) and transcription factor binding analysis using 48 tissues in GTEx v7 (18) and RegulomeDB (19), respectively. The overall deleteriousness of identified SNPs was evaluated through Combined Annotation Dependent Depletion (CADD), a tool that evaluates the deleteriousness of single nucleotide, insertion and deletion variants in the human genome through the integration of annotations from more than 60 different databases into one metric. (20, 21)

RESULTS

Approach 1: Association of Candidate Genes Selected from GWAS with Clinical Outcome in Pediatric AML Cohort

Previous work identified single nucleotide polymorphisms (SNPs) that are associated with in vitro cytarabine sensitivity in a GWAS conducted in HapMap LCLs. (13) A panel of relevant SNPs from the European (CEU) population of LCLs was selected for testing, as most children enrolled on the study were of European ancestry. SNPs included in this panel were initially selected based on association with cytarabine sensitivity with a p-value less than 0.0001. We further refined the list by eliminating any SNPs with a minor allele frequency of less than 0.05 in the CEU population. An additional three SNPs were excluded due to failure to genotype on OpenArray chip, resulting in 124 SNPs successfully genotyped in 348 patients.

SNPs and increased TRM

Two SNPs (rs6661575 (T/T) and rs2025501 (A/A) were significantly associated with both increased in vitro sensitivity to cytarabine in LCL (13) and increased TRM in children (Figure 3). Children with a T/T genotype at rs6661575 (n=7) had a one year cumulative incidence of TRM of 57 ± 22% versus those who were wild type, 15 ± 3% (p=0.0077). Those carrying the homozygous variant genotype A/A at rs2025501 (n=68) had a one year cumulative incidence of TRM of 29 ±7% vs 11 ± 3%, 20 ± 4% for genotypes A/G and G/G respectively (p=0.0228). (Table 1). Of note, the heterozygous genotype had unexpectedly lower TRM than either of homozygous variants. The impact of genotype on TRM was more notable in children randomized to consolidation with IDA-FLAG which contains higher dosing of cytarabine (7590 mg/m²⁾ than IDA-DCTER (1600 mg/m²). Patients with the recessive genotype A/A rs2025501 who were randomized to IDA-FLAG had increased TRM compared to A/G and G/G (34 ± 9% vs 5 ± 3% and 19 ± 8%; p-=0.0024). A similar change was not observed with rs6661575. However, the homozygous variant genotype is rare and the number of affected children was small, limiting power. A trend towards decreased relapse rate for the homozygous variant of both SNPs was noted (Table 1). In terms of functional significance, rs6661575 is intronic to FMOD on chromosome 1 and is an eQTL for the gene with a CADD score = 13.26. In contrast, rs2025501 is in a gene desert on chromosome 9, with no known functional annotation and a very low CADD score of 0.005.

Figure 3:

(A) Homozygous variant genotypes of rs2025501 and rs6661575 associated with increased cytarabine sensitivity in 85 LCLs of European ancestry from Hartford et al (previously unpublished figures) (13) (B) Patients carrying these homozygous variants randomized to regimen B (IDA-FLAG) with higher dose cytarabine have significantly increased TRM.

Table 1.

Patient outcomes according to genotype.

rs6661575	1 year cumulative incidence TRM	TRM randomized to IDA-FLAG arm	TRM randomized to IDA-DCTER arm	3 year relapse rate (RR)
Genotype (n)	n = 343	n = 130	n = 131	n = 279
C/C (263)	15 ± 3 %	16 ± 5 %	9 ± 3 %	44 ± 3 %
C/T (73)	21 ± 5 %	14 ± 6 %	10 ± 7 %	40 ± 7 %
T/T (7)	57 ± 22 %	60 ± 26 %	50 ± 50 %	14 ± 15 %
p-value	0.0077	0.0188	0.1086	0.5441

rs2025501	1 year cumulative incidence TRM	TRM randomized to IDA-FLAG arm	TRM randomized to IDA-DCTER arm	3 year relapse rate (RR)
Genotype (n)	n = 342	n = 130	n = 130	n = 278
G/G (106)	20 ± 4 %	19 ± 8 %	12 ± 5 %	46 ± 5 %
A/G (169)	11 ± 3 %	5 ± 3 %	6 ± 3 %	45 ± 4 %
A/A (67)	29 ± 7 %	34 ± 9 %	12 ± 8 %	31 ± 6 %
p-value	0.0228	0.0024	0.7697	0.2549

We assessed whether treatment selection may mitigate adverse outcomes for children with the variant (A/A) genotype of rs2025501 by examining OS according to randomization to IDA-FLAG versus IDA-DCTER. Those carrying the variant genotype demonstrated a trend toward improved overall survival when randomized to IDA DCTER arm with less cumulative cytarabine exposure (p=0.19) (Figure 4A). The small number of patients with variant genotype limited this analysis for SNP rs6661575 as only three patients with the variant genotype were randomized to IDA-DCTER, shown in Figure 4B

Figure 4:

Survival probability for children carrying each homozygous variant genotype according to treatment randomization to either IDA-FLAG or IDA-DCTER

4A. rs2025501 (p=0.19) 4B. rs6661575 (p=0.38) 4C. rs17202778 (p=0.2)

Approach 2: Replication of Previously Published SNPs associated with TRM

We sought to replicate previous findings of SNPs that impact response to cytarabine based chemotherapy in the St Jude’s AML02 study in our Children’s Cancer Group CCG2941/2961 AML cohort. (14) We genotyped five SNPs (rs17202778, rs1533140, rs6550826, rs9883101, rs1203633) that had a significant impact on TRM in the St Jude’s study in our cohort of 385 patients. In agreement with the report from St Jude’s Children’s Hospital, the variant genotype, rs17202778 C/C was significantly associated with TRM in our independent dataset (p<0.0001) (Figure 5). One year cumulative incidences of TRM were 72.9% for C/C vs 17.0% and 13.8% for C/T and T/T genotypes, respectively. The variant genotype C/C had lower three year relapse rate of 11% than the other genotypes, 23% C/T and 49% T/T (p=0.0013). We examined the impact of treatment randomization on survival for patients carrying the variant genotype (C/C) and noted a non-significant trend towards improved survival for patients treated with lower dosing of cytarabine (p=0.2) (Figure 4C). We performed a multivariate analysis for rs17202278 and the two SNPs identified in our first approach (rs6661575 and rs2025501). Both rs17202278 and rs6661575 remained significant (p = <0.0001 and p=0.0008, respectively). Although rs17202778 is in a gene desert on chromosome 2 with no known biological significance, this SNP has a very high CADD score = 19.50 (among the 10% most deleterious variants in the genome).

Figure 5. TRM for rs17202778.

Cumulative incidence of TRM in patients treated on CCG2941/2961 by genotype. BMT patients excluded

DISCUSSION

Chemotherapeutic regimens for acute myeloid leukemia are intensive and carry a risk of TRM, thus a significant number of children will die of toxicity from the therapy. (2) We hypothesized that genetic susceptibility to the deleterious effects of chemotherapy is in part responsible for the inter-individual variability in outcomes seen in clinical trials. In the two parts of this study, we have found three SNPs associated with TRM. The variant genotypes with increased in vitro sensitivity to cytarabine in the cell-based models were also associated with increased TRM in the children treated on the CCG clinical trial. Two of the SNPs are newly identified, while one of the SNPs, also identified in cell based models, was previously reported to be associated with increased mortality on the St. Jude’s AML trial. (14)

Recessive variants of rs6661575 (T/T) and rs2025501 (A/A) had significantly increased in vitro sensitivity to cytarabine in the LCL in the prior GWAS and were associated with increased treatment related deaths in children suggesting a direct effect of cytarabine chemotherapy, as the deaths were largely due to infections that result from mucosal barrier breakdown and profound neutropenia. Children with A/A genotype at rs2025501 randomized to the IDA-FLAG arm with higher doses of cytarabine had the more notable increase in TRM. We have previously reported a similar finding of increased toxicity in the risk genotype with higher doses of cytarabine in a previous pharmacogenetic study of cytidine deaminase (CDA), an enzyme responsible for irreversibly deaminating cytarabine. (22) Genotyping of children on the same CCG 2941/CCG 2961 study revealed increased TRM for those children carrying the homozygous variant gene of CDA associated with reduced enzyme activity. This effect was more pronounced for those children who were randomized to receive higher doses of cytarabine. (22)

The expected directionality of cellular sensitivity in vitro was observed via increased chemotherapy sensitivity and subsequent toxicity in children with these SNPs identified in this approach, suggesting a plausible genotype effect that warrants validation in an independent dataset. The functional implications of these SNPs is unclear at this time as they are both intergenic. The SNP rs6661575 is located on chromosome 1q32.1 close to gene FMOD, which encodes fibromodulin. Fibromodulin is an extracellular matrix proteoglycan which has an important role in matrix organization and is necessary for tissue repair in many organs. Recent studies have shown that it may play a role in various cancers, including leukemias, through angiogenesis and suppressing apoptosis.(23) It is also possible that these SNPs are in close linkage with other functionally important genes. Neither of our newly identified SNPs were included in the data set which was genotyped on the St Jude cohort as they did not meet significance parameters after meta-analysis of the six pooled GWAS studies used to identify candidate genes in that study.(14)

The second aim of our study was to replicate and validate the findings of the St Jude’s AML clinical trial which evaluated relevant SNPs derived from the meta-analysis of six different population based GWAS studies. Five SNPs that were associated with variable cytarabine sensitivity in vitro were also associated with treatment related mortality in their pediatric cohort. In our cohort, we found only one of the 5 SNPs to be associated with clinical outcomes. SNP rs17202778 (C/C) was found to be significantly associated with markedly increased treatment related mortality in children carrying the variant alleles in our independent data set. The very high burden of mortality in these children carrying the variant genotype in our study, which represents an independent validation of a previously identified signal for mortality in children with AML and is complementary to the studies of in vitro sensitivity, underscores the potential importance of this SNP in cytarabine-based chemotherapy. This SNP should be studied in upcoming prospective clinical trials in AML to ascertain whether this genotype can be used to stratify to standard versus high dose cytarabine based chemotherapy or a less intensive chemotherapy regimen if the excessive toxicity is again noted.

Pharmacogenetic studies for AML have several limitations which must be addressed in the context of these results. Chemotherapeutic regimens for AML incorporate multiple drugs, limiting our ability to decipher whether the outcomes we observe are the direct effect of cytarabine. The congruence of the in vitro sensitivity with expected treatment outcome allows us to infer an effect of cytarabine. Additionally, multiple previous studies have linked cytarabine sensitivity to outcome which supports the importance of this drug in both clearance of leukemia and excess toxicity.(6, 12) For two of the SNPs, we demonstrated an increase in toxicity signal when patients were randomized to increasing doses of cytarabine, again suggesting a dosage effect of this drug. The large number of SNPs analyzed allows for the risk of false discovery. Our risk of false discovery was minimized by our priori hypothesis of biologic plausibility of the data that increased in vitro sensitivity to cytarabine would predict toxicity, and maintaining a strict p-value for significance (p<0.01). These findings resulted from a multicenter national study with audited data and large number of patients, but we recognize that these limitations are not insignificant. The two newly discovered SNPs should be validated in an independent dataset to validate the clinical significance of our findings. In our second aim, we were able to provide supportive data that the previously identified SNP rs17202278 impacts TRM in children with AML. The reproducibility of the clinical importance of rs17202278 C/C predicting TRM, demonstrated in two large independent clinical trials, overcomes many of the expected limitations described above.

GWAS studies are helpful in finding novel determinants of drug sensitivity if validated in clinical samples.(24–27) The toxicity profile of cytarabine is well-established and notably dose dependent, with increased toxicity with high dose cytarabine (>1g/m²).(2) Pharmogenomic studies can be complementary to molecular and cytogenetic exploration of leukemia, offering further prognostication and chemotherapy selection in children with leukemia. The number of children with these genotypes is not insignificant, as even the rare recessive genotype (rs6661575) is present in 2% of our cohort. While the two newly discovered variants warrant additional validation in an independent clinical dataset, SNP rs17202278 has a toxicity signal in two independent datasets, and should be evaluated in prospective studies as it may impact dose-specific treatment strata or the need for additional supportive care in the future to reduce the significant mortality seen in those children harboring the recessive alleles.

Statement of Translational Relevance:

Genetic polymorphisms influencing sensitivity to cytarabine, identified using genome wide approaches in cell lines, predict outcome of therapy for childhood AML. In this study, we identify two new SNPs associated with cytarabine toxicity in pediatric AML. Furthermore, the present study validated a SNP previously identified in a separate cohort, using large patient numbers and patient outcome data, to identify those at risk of treatment related mortality. Knowledge of genetic characteristics that predict either excessive toxicity, or inadequate therapy (relapse) at the time of diagnose will allow for selection of individualized cancer treatments and will improve survival. These findings can be validated prospectively in pediatric treatment protocols and ultimately used to modify treatment selection.