Gemtuzumab Ozogamicin (GO) is a CD33-directed antibody conjugated to calicheamicin used in AML immunotherapy. Children's Oncology Group led pediatric AML phase II trial COG-AAML03P1 (NCT00070174) established that GO could be safely added to the standard chemotherapy regimen consisting of ara-C, daunorubicin and etoposide (ADE+GO) and improved outcome(Cooper et al., 2012). Subsequent COG-AAML0531 phase III trial (NCT00372593) randomized patients to receive either standard chemotherapy alone (ADE) or with addition of two doses of GO (ADE+GO) and showed that addition of GO improved outcome in newly diagnosed AML patients (Gamis et al., 2014). Our group previously established that CD33 genetic variation impacts GO response in AML patients (Lamba et al., 2017). Given that the antileukemic effect of GO is primarily driven by calicheamicin induced DNA damage, in this study we investigated pharmacogenomic impact of SNPs in DNA damage response (DDR) pathway genes on GO treatment response. We genotyped 132 SNPs in 42 genes involved in DNA damage repair pathway in DNA samples from 470 patients treated with standard chemotherapy in AAML0531 trial (ADE arm) and 755 patients treated with addition of GO to standard therapy in AAML03P1 and AAAML0531 trials (ADE+GO arm). Univariate analysis to test for association between OS, EFS, DFS and RR after induction 1 in both ADE+GO and ADE arms identified 20 SNPs in 16 genes that were significantly associated with at least one of the clinical endpoints tested in the only ADE+GO arm but not in ADE arm of the trials. We tested these 20 SNPs in all possible combinations with a maximum of 3 SNP/model using multivariable Cox proportional hazard models for association with EFS and OS in patients treated with ADE+GO. Drastically increased number of models arising from higher SNP combination numbers was a computational challenge thus we restricted our analysis to a maximum of 3 SNP combinations. We performed 1000 permutation tests per model to determine the likelihood of obtaining them falsely. Models were ordered according to their Bayesian Information Criterion (BIC) and weight in favor of each model. Those withleast BIC and a 1000 permutation p≤0.05 were selected for development of DDR pharmacogenomics score. DDR_PGx8 score was defined by adding the genotype scores of 8 SNPs in 7 genes (AKT1, ATR, DDB2, PARP1, PI3KCA, PTEN and RAD51) accounting for mode of inheritance (additive, dominant or recessive) and direction of their association with outcome (positive for beneficial and negative for detrimental association) Fig 1A shows overall study design. The DDR_PGx8 score ranged from -5 to 3 in patients treated with ADE+GO (n=755) or ADE alone (N=470). Based on the distribution, the scores were stratified into high (score ≥0, n=212 in ADE group and n=357 in ADE+GO group) and low score (score <0, n=241 in ADE group and n=329 in ADE+GO group) groups. The distribution of DDR_PGx8 score groups did not differ by risk groups or MRD1 status. However, it differed significantly within race with 81.81% (108/132) of black or African American patients compared to 43.75% (364/832) of white patients in the low DDR_PGx8 score group. Patients with low-DDR-PGx8 score had significantly worse EFS (HR=1.52, 95%CI (1.22-1.90), p<0.001; Fig 1B), OS (HR=1.62, 95%CI(1.24-2.11), p<0.001), DFS (HR=1.88, 95%CI(1.42-2.50), p<0.00001;), and higher RR1 (HR=1.89, 95%CI(1.40-2.40), p<0.00001) compared to high-DDR-PGx8 score patients when treated with GO (ADE+GO cohort). This impact of DDR_PGx8 was not observed in patients treated with standard chemotherapy alone (ADE arm), with no difference between low and high DDR_PGx8 score groups in EFS (Fig 1B), OS, DFS and RR (all p>0.28). In multivariable Cox proportional hazard models for DDR-PGx8 score groups, initial risk group assignment, WBC at diagnosis and age, low-DDR-PGx8 score remained a significant and independent predictor of inferior EFS (HR=1.6, 95%CI=1.21-2.02, p<0.001; Fig 1C) and OS (HR=1.6, 95%CI=1.17-2.22, p=0.003) in ADE+GO arm but not in ADE arm. We establish a DNA damage repair response-based pharmacogenomics score predicting outcome in patients treated with addition of GO to standard chemotherapy. The score was not predictive of outcome in patients treated with standard chemotherapy alone implying its impact is GO-specific. Our results in conjunction with existing CD33 SNPs provide a rationale for use of pharmacogenomics in personalizing GO treatment.

Figure 1
Figure 1.
View largeDownload slide

Disclosures

No relevant conflicts of interest to declare.

© 2021 by The American Society of Hematology
2021
Sign in via your Institution