GSTA1 Genotype Influences Performance of Initial Bu Prediction Methods during Conditioning before SCT

Introduction: Busulfan (Bu) is a component of conditioning before stem cell transplant (SCT). Bu has a narrow therapeutic window. High exposure was associated to higher rates of GVHD, VOD and mortality and low exposure to relapse or graft failure. Pharmacokinetic (PK) studies demonstrated high intra- and inter-patient variability. Several factors such as age, gender, weight and genetic components have been associated to this variability (McCune 2009, Vassal 1993, Ansari 2013). Certain models and nomograms have been used to predict Bu first dose, in theory precluding need of further adjustments. Recent comparison of 12 different models showed acceptable performances but high coefficients of variation (CV) and failure to reach target AUC in at least 24% of cases (Zao 2015). The present study aims to evaluate the role of GSTA1 haplotypes on different methods’ performances in predicting of Bu first dose.

Methods: A hundred and ten patients underwent a SCT after Bu based conditioning from October 2000 to October 2011 and 104 had PK data. Median of age was 6.9 (IQR 2 to 13.7) years, 55.5% were males and most had malignancies (69%). Most received Bucy (78.2%) as conditioning regimen. Four-time daily dose of Bu was calculated as previously described (Ansari 2014). Bu AUC, steady-state concentration (Css) and Clearance (Cl) were assessed for all patients after the first dose. The same parameters were predicted based on 11 other methods previously described (Zao 2015) to calculate the initial dose of Bu assuming a linear relationship between predicted doses and AUC. For each model that predicted individualized clearances, a mean prediction error (MPE) was calculated as follows: MPE=(Cl_predicted-Cl_observed)/Cl_observed. Target AUC was from 900μM.min through 1500μM.min. Two haplotypes, found previously to influence Cl were included in the analysis: haplotype *B1a (-1142G, -631G, -513C, -69T) and *A2 (-1142C, -631G, -513T, -69C) in heterozygous and homozygous state (Ansari 2013). X² test was used to compare methods’ performances and observed values. Continuous data was expressed as median and interquartile range (IQR). Multivariate linear model was used to evaluate the association of Cl MPE and the GSTA1 haplotypes.

Results: Initial doses of Bu, calculated by the different methods, varied from 0.47 to 1.6mg/kg. Predicted AUCs varied from 392 to 2,968μ.min with CV ranging from 21.5 to 37.1%. Ability to predict the target AUC was from 28.8% to 76%, with 0 to 20% of measures above the range. Bartelink’s (p<0.001) and McCune’s (p<0.001) methods demonstrated better performances: 76 and 71.2% of predicted AUC, respectively, against 39.4% of observed values. Few AUC were above the range using Bartelink’s and McCune’s approach (4.8 and 5.8%, respectively). Savic’s method, although 76% within the target, was associated to 20% of AUC above the range. Regarding the GSTA1 haplotypes, 96 patients were genotyped (87.3%). Bartelink’s and McCune’s models performed similarly in carriers of GSTA1 polymorphisms not associated with extreme clearances (78 and 76% respectively). Performances as low as 60%, with Bartelink’s and McCune’s methods, could be seen in patients homozygous for haplotype *A2, previously reported as rapid metabolizers (Ansari 2013). Levels above the target could be predicted at rates as high as 50% using Bartelink’s and McCune’s methods in carriers of haplotype *B1a, associated with lower clearances. Clearance-predictor models showed wide range of Cl MPE based on observed ones: -0.15 (IQR -0.29 to -0.06) for McCune’s, -0.07 (IQR -0.21 to 0.01) for Savic’s and -0.12 (-0.35 to 0.05) for Long-Boyde’s. In a multivariate model that included age, GSTA1 haplotype and gender, GSTA1 (p=0.003) and age (p=0.003) were independent factors to predict MPE in McCune’s model. Its clearance prediction was 30% less (95% CI, 10.1 to 48%) than the observed clearance in homozygotes for haplotype *A2.

Conclusion: Bartelink’s and McCune’s models seemed to be adequate methods to calculate the initial dose of Bu and demonstrated better performances than the method used in our center. For the clearance-predictor methods, although McCune’s seemed to be the most reliable model for most patients, it had limited performance in GSTA1 haplotypes previously associated to extreme clearances. It suggests that genetic factors such as GSTA1 genotyping should be included in a future model design to better predict the initial dose of Bu.