Electronic Supplementary Material
Preclinical Studies – (Online Resource 1)
Methods
MDA-MB-231 (triple negative) and MCF-7 (ER/PR+, HER2-) cells (100 µl) were seeded into 96-well plates and treated with ixabepilone and vorinostat for 72 hours. The final concentrations of ixabepilone and vorinostat were 5 nM and 0.5 µM, respectively. The half-maximal inhibitory concentration (IC50) was calculated for both drugs. The synergistic effect of these two drugs was determined with Compusyn software (ComboSyn, Inc., Paramus, NJ), based on the median effect principle as implemented by Chou and Talalay [33].
Human neuroblastoma cells (SH-SY5Y) were trypsinized and counted using a hemocytometer. Ten thousand cells were seeded onto glass coverslips and placed into 10 cm petri-dishes overnight. After confirming that cells were adhered to coverslips, ixabepilone (5 nM) and/or vorinostat (0.5 µM) were added to cells. Cells were treated for 48 hours and an immunofluorescence assay was used to detect acetylated -tubulin, based on a previously reported approach[34].
Results
Ixabepilone and vorinostat IC50 values in MDA-MB-231 cells were 5.1 ± 0.4 nM and 0.48 ± 0.06 uM, respectively, while in MCF-7 cells they were 7.9 ± 0.5 nM and 1.8 ± 0.3 uM, respectively. The synergistic effect of the combination is suggested by the combination index <1 at the higher range of effect size, see Fig 1.
Fig. 1 Synergistic effect of vorinostat and ixabepilone in MDA-MB-231 (A) and MCF-7 (B) breast cancer cells. Fa, fraction of cells affected (killed); CI, combination index; CI(ED50), combination index at the point where 50% of cells are affected (killed).
Untreated control cells (Fig. 2A) and cells treated with vorinostat (Fig. 2C) demonstrated well-developed neurites, however, few neurites were seen in cells treated with ixabepilone alone (Fig. 2B). Ixabepilone related decrease in neurite formation was attenuated by the combination of vorinostat with ixabepilone, which increased the neurite number compared with ixabepilone alone (Fig. 2D).
Fig. 2 Effect of vorinostat and ixabepilone on neurite formation (visualized by color intensity) in SH-SY5Y human neuroblastoma cells[34]. (A) Untreated control cells. (B) Ixabepilone alone. (C) Vorinostat alone. (D) Combination of vorinostat and ixabepilone.
Pharmacokinetic Studies – (Online Resource 2)
Methods
Vorinostat pharmacokinetics were evaluated on cycle 1 day 1 (vorinostat alone) and day 2 (vorinostat dosed at start of ixabepilone infusion) for schedule A and B. Blood samples were obtained at the following time-points: before and at 0.5, 1, 1.5, 2, 3, 4, 6, and 8 h after dosing. At each time point, 5 mL of peripheral blood was collected in a red-topped Vacutainer (BD, Franklin Lakes, NJ). The blood was allowed to clot at room temperature for 30 min and then was centrifuged at 2,000 × g for 15 min at 4 °C. The resulting serum was stored at −70 °C or colder until analysis.
Ixabepilone pharmacokinetics were evaluated on cycle 1 day 2 (vorinostat dosed at start of ixabepilone infusion) for schedule A and on cycle 1 day 2 (vorinostat dosed at start of ixabepilone infusion) and day 9 (ixabepilone alone) for schedule B. Blood samples (7 ml) were drawn from a vein in the arm opposite to that used for infusing the drug and collected in tubes containing freeze-dried sodium heparin. Heparinized blood samples were obtained at the following time-points: before and at 0.5, 1, 1.5, 2, 3, 4, 6, 8, 24, 48, and 72 h after start of infusion. Each sample was centrifuged at approximately 2000 × g at 4 °C for 10 min, and the resulting plasma was stored at -70 °C or colder until analysis.
Vorinostat was quantitated with a previously published, validated LC-MS/MS assay[35]. Ixabepilone concentrations were quantitated by LC-MS/MS based on a previously published assay[36]. Modifications were as follows. The LC system consisted of an Agilent (Palo Alto, CA, USA) 1200 SL autosampler and binary pump, a YMC-Pack ODS-AQ (3 µm, 100 mm x 2 mm) column, (YMC America, Inc., Allentown, PA, USA), and an isocratic mobile phase of acetonitrile/water (70/30, v/v) with 10 mM ammonium acetate, pH 5.0, pumped at a flow rate of 0.3 mL/min. The overall run time was 2.5 min. To aliquots of 50 µL of plasma, 500 µL of acetonitrile with internal standard (40 ng/mL BMS-212188) were added. After vortexing and centrifugation, 100 µL of supernatant were transferred to autosampler vials followed by injection of 5 µL. This assay was accurate (within 10%) and precise CV% within 14%) over a range of 1-1000 ng/mL.
The maximum concentration (Cmax) and the time to reach it (Tmax) were determined by visual inspection of the concentration versus time data. Other pharmacokinetic parameters were calculated non-compartmentally using PK Solutions 2.0 (Summit Research Services, Montrose, CO; Descriptive statistics were calculated with Microsoft Excel 2010. Pharmacokinetic parameters between days were compared by nonparametric 2-tailed Wilcoxon exact signed rank test as implemented by IBM SPSS statistics version 22.
Population pharmacokinetics (POP-PK) modeling of ixabepilone data was performed with NONMEM 7.3 (ICON Development Solutions, Ellicott City, MD), which was run with the PLT Tools 5.1 statistical interface (PLT Soft, San Francisco, CA), to further understand the pharmacokinetic effects of different infusion schedules. One-compartment, 2- compartment and 3-compartment models were fit to the plasma data. Between-subject variances of PK parameters were assumed log-normally distributed, and proportional residual errors were used. The model diagnostics for the goodness of fit included the NONMEM objective function (OFV; -2 x log-likelihood) and a visual inspection of graphics, such as the correlation/symmetry of observations versus predictions (population and post hoc) and conditional weighted residuals versus time and predictions. Statistical support for choosing models was obtained through comparisons of changes in OFV (DOFV); DOFV > 3.84 for additional estimated parameters, respectively (P < 0.05, chi-square). The total body weight (in kilograms), body surface area (BSA; Mosteller formula, in m2[37]) of patients were explored as individual-specific covariates of the drug clearance and distribution volumes. Inter-occasional variability was checked to evaluate drug-drug interaction between ixabepilone and vorinostat for patients in schedule B.
Limited sampling strategies were explored with ADAPT5 software for pharmacokinetic/pharmacodynamic systems analysis through D-optimality using the population mean parameters obtained from analysis of the data with NONMEM[38,39].
Results
Compartmental analysis of ixabepilone was based on 854 observations (Table 1) and was best described by a 3-compartment model, as demonstrated by a significant reduction in OFV (OFV difference=271.7) and a triphasic decline in the plasma concentrations after the maximum concentration in all 51 patients. The final basic structural model used the 3-compartment model (ADVAN 11) and first-order conditional estimation with interaction. Total body weight and BSA did not appear as significant covariates. Inter-occasion variability for drug interaction with vorinostat did not show a significant improvement in OFV. Diagnostic plots from the basic 3-compartment model are presented in Fig. 3. Predicted population and post hoc concentrations versus observed data about their respective lines of identity displayed reasonable symmetry for the majority of the concentrations examined.
Estimation of ixabepilone clearance with a single sample showed standard errors in the estimate diminished at later time points: 48 h (32.68%), 72 h (30.23%), and 96 h (29.32%). Allowing for multiple samples to be drawn out to 72 h increased the ability to accurately estimate clearance: N=1 (30.23%), N=2 (21.38%), N=3 (17.45%), and allocated all sampling times to 72 h. This suggests the importance of the contribution of the elimination phase to AUC and of getting an accurate plasma determination at the late time point. Simulation of 1000 patients at 32 mg/m2 resulted in a mean (SD) 72 h plasma concentration of 46.1 (20.6) ng/mL, amply above the assay lower limit of 1 ng/mL.
In summary, the limited sampling strategy explored based on the POP-PK parameters showed that sampling after 72 hours provided good predictive parameters. This suggests that the terminal phase is the most informative and that future sparse sampling designs should focus on 72 h blood samples. However, the time frame could be quite challenging in clinical practice with may not be feasible in larger trials.
Table 1. Population pharmacokinetic parameters.
Parameters, median (% SE)CL (L/hr) / 36.3 (11.4)
Vc (L) / 4.6 (67.6)
Q1 (L/hr) / 20.3 (22.2)
V1 (L) / 31.4 (14.6)
Q2 (L/hr) / 50.8 (21.9)
V2 (L) / 1379.3 (14.7)
IIV of CL (%) / 1.5
IIV of Vc (%) / 19.1
IIV of Q1 (%) / 4.8
IIV of V1 (%) / 1.4
IIV of Q2 (%) / 1.6
IIV of V2 (%) / 0.02
additive error / 0.53 (30.2)
proportional error / 0.33 (4.9)
Objective function / 5011.5
Fig. 3 Goodness of fit diagnostics of the final pharmacokinetic model. Observed plasma concentrations of ixabepilone versus predicted plasma concentrations of ixabepilone. The left panel shows typical predictions without the inclusion of model variability (interindividual variability), whereas the right panel represents the complete model, including interindividual variability (post hoc fit). The curvilinear plots (green) depict local regression trend smoother fits to the data.