SAM486A (previously termed CGP 48664), a potent inhibitor of S-adenosylmethionine decarboxylase, is under clinical development for the treatment of advanced refractory malignancies. Hematological toxicity manifested by dose-dependent neutropenia has been observed in phase I studies. Population methods were used to investigate pharmacokinetics (PK) as a prognostic factor for safety end point (hematological toxicity) in patients with advanced cancers. SAM486A plasma concentrations and neutrophil counts were collected from three ascending-dose tolerability and PK studies (study 1: single 5-day continuous intravenous (IV) infusion with doses ranging from 24-700 mg/m2/cycle; study 2: 10-minute to 3-hour IV infusion once weekly with doses ranging from 16-325 mg/m2/week; study 3: 1-hour IV infusion once daily for 5 days with doses ranging from 3.6-202.8 mg/m2/day). The PK of SAM486A were best estimated by a population linear three-compartment model with NONMEM (version 5) using data from 9 patients in studies 1 through 3. The population pharmacokinetic parameters (SD) were CL = 6.2 (0.4) l/h/m2, Q2 = 15.4 (1.5) l/h/m2, Q3 = 33.6 (5.3) l/h/m2, V1 = 9.5 (1.6) l/m2, V2 = 672 (52) l/m2, and V3 = 39.9 (8.3) l/m2, and the corresponding intersubject variability was 45.4%, 74.0%, 85.3%, 80.1%, 37.0%, and 103%, respectively, where CL is total body clearance, Q2 and Q3 are intercompartmental clearances, and V1, V2, and V3 are the volumes of distribution in central and peripheral compartments, respectively. The intrasubject variability was 24.0%. The cumulative AUC before the onset of neutrophil nadir count (AUC) and the duration of exposure over threshold SAM486A concentrations in the range of 0.05 to 0.1 microM based on Bayesian PK parameter estimates significantly correlated with absolute neutrophil count nadir (< 5 x 10(9)/l). AUC showed the best correlation (R2 = 0.72) with absolute neutrophil count nadir by an inhibitory sigmoid Emax model and also correlated with percent decrease in neutrophil count from baseline to nadir by a simple Emax model (R2 = 0.53). Logistic regression analysis indicated that AUC and the duration of exposure over 0.05 to 0.1 microM, but not Cmax, were strong predictors of grade 4 neutropenia (< 0.5 x 10(9)/l). Drug exposure parameters such as AUC derived from population analysis may be used clinically as a useful predictor of drug-induced neutropenia.