Paper Details

ABSTRACT

Background: Although advances in therapy have improved outcomes for children with acute myeloid leukemia (AML), treatment-related toxicity remains a major clinical challenge. In our patient cohort (AML02 and AML08 trials), grade ≥3 gastrointestinal (GI) toxicity occurred in nearly one-third of patients during Induction 1 (30%; 121/400) which can further increase infection risk, prolong hospitalization, and treatment-related morbidity. The marked interpatient variability in severity of toxicity suggests that inherited genetic variation may influence patients’ susceptibility. We previously developed ACS10, a 10-SNP pharmacogenomic score for cytarabine predictive of induction efficacy; however, no genetic tool currently exists to predict toxicity risk in AML. In this study we focused on GWAS followed by development of a GI-PGx3, a three-SNP pharmacogenomic score to predict grade ≥3 GI toxicity during Induction 1 in pediatric AML patients treated on the multicenter AML02 and AML08 trials.

Methods: This study included 400 Pediatric AML patients enrolled on the St. Jude AML02 and AML08 trials; of these, 121 patients experienced grades ≥3 GI toxicity during Induction 1. GWAS was performed using a standard pipeline followed by testing significant SNPs for 2- or 3-SNP combination models. The SNP-SNP combination models were ranked using the Bayesian Information Criterion (BIC) and validated by 1,000 permutations. The best-performing model (lowest BIC, permutation P<0.001) was selected to construct the GI-PGx3 score by summing genotype-based points across the three SNPs. Patients were stratified into low- (0–1), intermediate- (2), and high-risk (3–5) groups.

Results: A genome-wide association study identified 77 variants associated with toxicity, and linkage disequilibrium pruning retained 45 independent lead SNPs. Among the top GWAS signals, rs7074064, an intronic BMPR1A variant, showed increasing grade ≥3 GI toxicity across TT, TC, and CC genotypes (22%, 34%, and 43%; P=1.64×10?³). The CC genotype remained independently associated with toxicity (adjusted OR, 3.55; 95% CI, 1.76–7.32; P<0.001), while the C allele was associated with reduced BMPR1A expression in stomach and esophageal mucosa. BMPR1A supports GI epithelial integrity, additionally a missense variant in LD with the top SNP has been previously linked to juvenile polyposis syndrome.
To translate the GWAS findings into a multivariant predictor of toxicity, we next evaluated the 45 independent lead SNPs in combination. The top best fit model (BIC=383.35; model weight=0.299; permutation P<0.001) was a 3-SNP combination model composed of rs56365935 (in CCDC26), rs9806895 (CMIP), and rs10906449 (FRMD4A). Among 398 patients with complete genotype and toxicity data, 185 were classified as low risk, 125 as intermediate risk, and 88 as high risk. The incidence of grade ≥3 GI toxicity increased across the GI-PGx3 groups, with 14.6% of low-risk patients (27/185), 32.0% of intermediate-risk patients (40/125), and 60.2% of high-risk patients (53/88) having grade ≥3 GI toxicity (P<0.001). Compared with the low-risk group, the incidence of grade ≥3 GI toxicity was more than twice as high in the intermediate-risk group and about four times as high in the high-risk group. In the multivariable analysis adjusting for age, race, treatment arm, and initial risk group, the GI-PGx3 score remained strongly associated with grade ≥3 GI toxicity. Each one-point increase in the score was associated with a 2.54-fold increase in the odds of grade 3 or more GI toxicity (OR, 2.54; 95% CI, 1.96–3.29; P=2.03×10?¹²).

Conclusions: GI-PGx3 stratified pediatric AML patients into groups with markedly different risks of grade ≥3 GI toxicity during Induction 1. Following external validation (ongoing in AAML0531 cohort), this pharmacogenomic score may help identify patients who could benefit from intensified supportive care and toxicity mitigation strategies.