Exposure–Response Analysis of Alvocidib (Flavopiridol) Treatment by Bolus or Hybrid Administration in Newly Diagnosed or Relapsed/Refractory Acute Leukemia Patients

Alvocidib (flavopiridol) is a synthetic flavone that was derived from a naturally occurring compound found in the plant Dysoxylum binectariferum (1). Flavopiridol prevents cell-cycle progression via cyclin-dependent kinase inhibition, inhibition of transcription elongation, and induction of apoptosis (2–4). Clinical–laboratory studies were used to derive a schedule of flavopiridol followed in a timed sequential manner by Ara-C and mitoxantrone (FLAM; refs. 4–8). The sequential addition of flavopiridol to the combination regimen is intended to induce cell-cycle arrest, kill noncycling cells, and then restart the cell cycle in the remaining cells, leaving them susceptible to the effects of Ara-C and mitoxantrone. Recent clinical studies have provided evidence to support the efficacy of the FLAM for poor-risk, newly diagnosed acute myelogenous leukemias (AMLs) as well as relapsed/refractory AML or acute lymphocytic leukemias (ALL; ref. 9).

Flavopiridol was initially evaluated clinically using extended continuous intravenous infusion of 24 to 72 hours (10–13) or short bolus intravenous infusion (8, 13, 14) both of which resulted in limited efficacy. For relapsed and primary refractory patients with AML treated with FLAM, the 1-hour bolus infusion MTD for flavopiridol was established at 50 mg/m²/day by dose-limiting toxicities (DLTs) of profound and prolonged neutropenia at 60 mg/m²/day (8). Byrd and colleagues noted that there was differential protein binding between fetal bovine and human serum which may account for the lack of translation of in vitro activity to the clinic in chronic lymphocytic leukemia (CLL; ref. 15). To overcome the protein-binding differential, Byrd and colleagues developed a hybrid administration consisting of a 30-minute intravenous bolus followed immediately by a 4-hour intravenous infusion, which has been successfully applied to refractory CLL (16). Given the improved clinical activity, the hybrid schedule was incorporated into the FLAM regimen for relapsed and refractory AML or ALL with an overall complete remission (CR) rate of 39% (17). In this study, the hybrid MTD was established at 30/60 mg/m²/day with DLTs of tumor lysis, hyperbilirubinemia, and mucositis at 30/70 mg/m²/day (17). A three-arm randomized trial comparing FLAM and two other regimens for patients with relapsed and primary refractory AML demonstrated relative superiority of FLAM over the other regimens with 28% CR rate overall and 40% CR rate in patients aged 60 years or less, the age difference being associated with heightened mortality in the older cohort (9, 18). In newly diagnosed patients with AML, there was no difference between the hybrid schedule and short bolus intravenous infusion flavopiridol as part of the FLAM with regards to efficacy and toxicity with the bolus schedule resulting in higher maximum total or unbound concentrations (18). On the basis of the equivalency between the two dosing schedules and the less complicated administration of bolus versus hybrid schedule, the bolus schedule was used for a randomized phase II trial comparing FLAM to the standard-of-care regimen of cytarabine/daunorubicin (7+3) in newly diagnosed AML with nonfavorable features (NCT01349972; ref. 20). In this trial, FLAM induced CR in 70% versus 46% with 7+3 (P = 0.0003) or 7+3 followed by 5+2 for residual AML on day 14 of therapy (CR 57%; P = 0.08). Because of the reproducibly salutary effects of FLAM in diverse adult AML populations, follow-up studies are planned for both newly diagnosed and relapsed/refractory cohorts. In this context, it is important to determine the optimal schedule for flavopiridol delivery for future clinical trials.

This report describes the results of a population pharmacokinetic analysis of total and unbound flavopiridol concentrations delivered using bolus or hybrid administration to patients with acute leukemia with either a newly diagnosed or relapsed/refractory AML. The effect of administration scheme on flavopiridol pharmacokinetic parameters and flavopiridol plasma protein binding was assessed. Exposure–response relationships between unbound flavopiridol exposure measures and clinical response or toxicities were explored along with the effect of disease history on these relationships.

Data from patients enrolled in three clinical trials were used to study the pharmacokinetics and pharmacodynamics of flavopiridol as the initial component in polytherapy comprising flavopiridol, followed sequentially by Ara-C and mitoxantrone (FLAM; refs. 17–19). Two of the studies enrolled patients with newly diagnosed, poor-risk acute myelogenous leukemias [AML; Study 1 (19) and 3 (18)], while the third study enrolled patients with relapsed and refractory acute leukemias [Study 2 (17)]. All patients were adults with pathologically confirmed disease associated with ECOG performance status from 0 to 2. All studies were approved by an appropriately constituted institutional review board. All subjects were enrolled only after giving informed consent to their participation.

A 9-day cycle of FLAM treatment consisted of flavopiridol dosing for 3 days, followed by a 72-hour intravenous infusion of cytosine arabinoside at 2 gm/m² beginning at day 6, and a 60- to 120-minute intravenous infusion of mitoxantrone at 40 mg/m² on day 9. Flavopiridol bolus dosing consisted of daily intravenous administration over a 1-hour period. Flavopiridol hybrid dosing comprised daily intravenous dosing beginning with a bolus dose consisting of 30% to 43% of the total daily dose administered over a 30-minute period followed by the remainder of the daily dose administered over a 4-hour period.

The pharmacokinetic sampling schemes were largely standardized across the three studies used in the analyses. However, the schemes differed slightly to accommodate the particular administration scheme used. Samples for pharmacokinetic assessment were collected predose on days 1, 2, and 3 regardless of the drug administration schedule. For bolus dosing (Studies 1 and 3), samples were collected on day 1 at the end of the 60-minute bolus administration and again 4 hours later (5 hours postdose). For hybrid dosing (Studies 1 and 2), samples were collected on day 1 at the end of the 30-minute bolus administration and again after the 4-hour infusion (4.5 hours postdose). Additional samples specific to Study 1 were drawn and included day 3 postdose samples at 30 minutes and 4.5 hours for hybrid and 60 minutes and 5 hours for bolus (18). All patients in Study 1 had samples drawn 24 hours and 48 hours after the last dose. A total of 780 total and unbound flavopiridol concentrations were available across the three studies. In nearly all cases, there was an unbound flavopiridol concentration corresponding with a total flavopiridol concentration except for two cases, one with a missing total concentration and one with a missing unbound concentration.

The bioanalytic methods used for the measurement of total flavopiridol in plasma were consistent across the three studies and were previously described (8, 11, 17). Total flavopiridol was separated using reverse-phase high-performance liquid chromatography and detected using tandem mass spectroscopy. The lower limit of quantification (LLOQ) for total flavopiridol was 0.1 ng/mL. Unbound flavopiridol concentrations were measured using a microequilibrium dialysis method (17). All methods were appropriately validated and showed acceptable levels of performance.

Population pharmacokinetic analysis was performed using standard techniques after exploratory data analysis that included concentration versus time profiles and regression of total and unbound flavopiridol concentrations. Structural models with two or three compartments were fit to the total flavopiridol concentration versus time data. Additive, proportional, and additive/proportional residual variability models were assessed. Individual patient concentration differences from the population mean (ϵ) were described in the model using a normal distribution with a mean of zero and a variance of sigma (∑). Model and data support for the estimation of between-subject variability (BSV) was systematically tested. Individual patient parameter differences from the population mean (η) were described in the model using a log-normal distribution with a mean of zero and variance of omega (Ω). Covariate data (body surface area, gender, and age) were tested for the ability to explain variability in the flavopiridol pharmacokinetics. Given the availability of data from daily dosing across three days in Study 1 (bolus dosing), the ability to estimate interoccasion variability was assessed. Model development was guided by minimization of the -2LL objective function and likelihood ratio testing, reasonableness of parameter estimates, estimate precision, as well as standard goodness of fit plots including observed concentrations (DV) versus population predicted (PRED) or individual predicted (IPRED) concentrations. Bootstrapping using 200 replicates after stratification by dosing schedule was used to obtain final precision estimates for the model. A visual predictive check (VPC) using 1,000 replicates was performed to provide an overall assessment of the agreement between observed and predicted data. Nonlinear mixed-effects modeling used Phoenix NLME v 1.4 with its Quasi-Random Parametric Expectation Maximization (QRPEM) estimation algorithm (Certara, L.P.).

A goal of the analysis was to determine whether there were any differences in pharmacokinetic parameters associated with use of bolus versus hybrid intravenous dosing. In particular, there was interest in assessing whether the fraction unbound, F_u, differed between the dose administration schemes. Three approaches were used to explore this: (i) a linear regression comparison of the bound versus unbound flavopiridol concentrations stratified by dosing scheme, (ii) testing dosing scheme as a covariate on the F_u parameter, and (iii) comparing the reestimated model parameters after stratifying the dataset by dosing scheme.

To understand whether there were any differences in pharmacokinetics for newly diagnosed patients versus relapsed/refractory patients, the dataset was stratified by disease history and then model parameters for each patient group were estimated and compared.

Samples for pharmacokinetic characterization were only collected in cycle 1 of treatment. Accordingly, for the current analyses, clinical response and adverse event (AE) data were compiled for the three studies using data from the first treatment cycle. A cycle was defined as completion of all doses of flavopiridol, Ara-C, and mitoxantrone. Some patients underwent a subsequent cycle of treatment, but these data were excluded. The intent was to explore potential relationships between the pharmacokinetics and the pharmacodynamics of the same cycle.

Complete remission (CR) was defined as follows: bone marrow showing less than 5% myeloblasts with normal maturation of all cell lines, an absolute neutrophil count (ANC) of at least 1,000/μL, and a platelet count of 100,000/μL, absence of blasts in peripheral blood, absence of identifiable leukemic cells in the bone marrow, clearance of disease-associated cytogenetic abnormalities, and clearance of any previously existing extramedullary disease (20). Complete remission with incomplete blood count recovery (CRi) was defined the same as for CR but without achievement of ANC of at least 1,000/μL and/or platelet count of 100,000/μL (20). The clinical response was converted to a binary variable where a CR or CRi was coded as 1, and any lesser response was coded as 0. There was a total of 115 responses available from the 129 patients in the three studies.

A list of 11 adverse events of interest was compiled: elevated alanine aminotransferase (ALT), elevated aspartate aminotransferase (AST), hyperbilirubinemia, diarrhea, nausea, vomiting, mucositis, rash, elevated creatinine, tumor lysis syndrome (TLS), and cardiac-related adverse events. To focus the analyses on only the most clinically relevant adverse events, the datasets from each study were systematically processed to retain only those adverse events (i) from cycle 1 of treatment, (ii) with an attribution of possibly, probably, or definitely related to the investigational agents, (iii) with a severity grade of 3 or higher, and (iv) from patients in the pharmacokinetic population. Adverse events (AE) were graded by NCI Common Toxicity Criteria under the version active during the conduct of the trial.

Pharmacokinetic/pharmacodynamic analyses were performed with unbound flavopiridol exposures based on a mechanistic assumption (i.e., it is only the unbound flavopiridol that is able to interact with receptors and cause downstream effects). Post hoc unbound flavopiridol pharmacokinetic profiles from 0 to 96 hours using the final pharmacokinetic model were summarized as AUC for use in exposure–response analyses.

Data for clinical response and AE incidence were stratified by quartiles of flavopiridol-unbound AUC, disease history, or dosing schedule. Differences across stratification factors were tested by Fisher exact test for categorical variables and ANOVA for continuous variables, except for unbound AUC and C_max that were tested with the nonparametric Kruskal–Wallis test. Graphical analyses included plots of clinical response probability (with 95% confidence interval) versus quartiles of flavopiridol-unbound AUC (h*ng/mL), first for all patients and then stratified by disease history.

Recursive partitioning was used to explore flavopiridol-unbound AUC, disease history, and dosing scheme as predictor variables for clinical response. This technique iteratively selected the most important predictor variable (and threshold level for continuous variables) and used it to make binary splits of the data into branches. The iterative splits continued until there were too few values for additional splitting (i.e., 20 in a node) or an additional split would not improve the overall fit (21). A pruning procedure was subsequently used on the raw partitioning tree in an attempt to cut the tree back to the point where overall predictive accuracy was maximized, thus preventing overfitting of the data (22).

Using the flavopiridol pharmacokinetic model, stochastic simulations were performed (n = 500 subjects per dosing scenario; without residual variability) to generate 96-hour unbound flavopiridol pharmacokinetic profiles for bolus 50 mg/m²/day and hybrid 20/30, 25/35, 30/40, 30/50, 30/60, and 30/70 mg/m²/day dosing. For each pharmacokinetic profile, the unbound flavopiridol AUC from 0 to 96 hours was calculated. The clinical response probabilities obtained from the recursive partitioning results at the specific decision nodes were used to generate an overall probability of clinical response for each dosing scenario by summarizing the mean of assigned individual probabilities from the 500 simulations.

Statistical and graphical analyses used the R software package (23).

Baseline characteristics for all patients are presented in Supplementary Table S1. Bolus flavopiridol was administered to 56 patients (43% of the overall population) who received the previously determined MTD of 50 mg/m²/day (8). Hybrid flavopiridol was administered to 73 patients (57% of the overall population) who received total daily doses ranging from 50 to 100 mg/m²/day with the initial bolus component of the schedule comprising from 30% to 43% of the total daily dose. The overall population of 129 patients was equally represented with respect to gender and had a median age of 58 years. Nearly two-thirds of the patients had poor-risk, newly diagnosed AML while the remaining one-third of patients had relapsed or refractory acute leukemia.

Observed total flavopiridol concentration versus time profiles for bolus and hybrid administrations are shown in Supplementary Fig. S1. The relationship between total and unbound flavopiridol using all samples is presented in Figure 1. The relationship appeared linear across the range of observed total flavopiridol concentrations, from just above the LLOQ to nearly 5,000 ng/mL, with values clustered relatively tightly around the regression line.

Both two- and three- compartment structural models were assessed for flavopiridol but poor parameter estimate precision suggested that either there was not enough information content in the dataset to support three compartments or alternatively that a three-compartment model was not the best representation for flavopiridol pharmacokinetics. Of note, previous flavopiridol models reported in the literature have all used two-compartment models even when richer sampling schemes were used (8, 24–26). A two-compartment structure was therefore adopted to describe flavopiridol pharmacokinetics.

Between-subject variability (BSV) was systematically tested in the base model to assess the extent to which overall variability could be assigned to differences between patients. The best model included BSV on clearance (Cl), volume of distribution (V), inter-compartmental clearance (Q), and peripheral volume of distribution (V₂). The addition of BSV on F_u was assessed but did not improve the model objective function. The BSV estimate for F_u was low (6.71 %), but because the η-shrinkage was high (71.6 %), the value was suspect. Therefore, BSV on F_u was ultimately not incorporated into the model. Inspection revealed a high correlation between etas (η), in particular between the ηCl and ηV pair (correlation 0.847) and the ηQ and ηV₂ pair (correlation 0.675). To deal with the correlation, a full block omega structure was implemented and significantly improved the model by reducing the objective function 78 units. Post hoc–estimated concentration versus time profiles were analyzed and revealed no apparent occasion-based differences.

Body surface area, age, and sex were simultaneously added to the model to assess whether they could help explain a portion of the variability in the pharmacokinetic parameters. The covariates were only added to parameters for which there was a mechanistic rationale for a relationship. Body surface area was added to Cl, V, Q, and V₂. Age and sex were added only to the Cl parameter. Body surface area and age were centered on the population median values for these covariates. Considering the 95% confidence intervals for parameter estimates, the only covariate relationship with an effect distinguishable from zero was between BSA and V. BSA ranged in the study population from 1.4 m² to 2.7 m² with a median of 1.9 m². The covariate relationship suggested that a patient with the lowest BSA would have a V approximately 85% of the population mean V. A patient with the highest BSA would have a V approximately 120% of the population mean V. Although statistically significant, the relationship between BSA and V was likely not of clinical importance. None of the tested covariates were incorporated in the final model.

The parameter estimates from the final flavopiridol pharmacokinetic model are shown in Table 1. The diagnostic plots in Supplementary Fig. S2 showed good agreement between observed and either population or individual predicted total and unbound flavopiridol concentrations. Local regression (LOESS) lines fit to the data were nearly superimposable with the lines of unity. The residual plots displayed in Supplementary Fig. S3 did not reveal any untoward trends, indicating the sufficiency of the proportional residual variability models used. The visual predictive check shown in Supplementary Fig. S4 revealed an overall good fit between the observed and simulated total and unbound flavopiridol concentrations. Supplementary Fig. S5 presents representative individual plots of PRED and IPRED versus DV. These plots portray the relatively good fit of the model and also highlight the different pharmacokinetic profile shapes associated with bolus versus hybrid dosing.

Simple linear regression of the bound versus unbound flavopiridol concentrations stratified by dosing scheme showed a slope-based unbound fraction of 11.4% for bolus dosing and 10.3% for hybrid dosing (P <0.0001; Fig. 1). However, more comprehensive analyses using the pharmacokinetic model showed a universal unbound fraction of 10.9% (Table 1) with no difference in the unbound fraction between the dosing schemes. Dosing scheme added as a covariate on the F_u parameter did not improve the model objective function, indicating no difference in F_u for bolus versus hybrid dosing. When separate pharmacokinetic analyses were performed for each dosing scheme (results not shown), all the structural parameters were similar between the two schemes, including F_u, again providing evidence for the universal F_u of 10.9%. The mean F_u (95% confidence interval) for bolus dosing was 11.0% (10.4%–11.6%) and for hybrid dosing was 10.8% (10.1%–11.5%).

All the structural pharmacokinetic parameter estimates were similar for newly diagnosed versus relapsed/refractory patients except for the V₂ parameter. Newly diagnosed patients had a V₂ of 89.1 L, while relapsed/refractory patients had a V₂ of 139 L. The V₂ estimates for the two groups had nonoverlapping 95% confidence intervals. Simulations of the typical pharmacokinetic profiles associated with the population parameter estimates from the newly diagnosed and relapsed/refractory models showed only minimal changes (data not shown). A single pharmacokinetic model was therefore subsequently used to describe the pharmacokinetics of all patients regardless of disease history.

Unbound flavopiridol concentrations were used to assess pharmacokinetic/pharmacodynamic relationships on a mechanistic basis, but because there was a constant relationship between unbound and bound concentrations in the pharmacokinetic model, the results would have been the same had total flavopiridol concentrations been used in the analyses.

Higher flavopiridol-unbound AUC was associated with a higher probability of clinical response in an analysis of all patients (Fig. 2A). The same analysis stratified by patient disease history confirmed that newly diagnosed patients generally responded better than relapsed/refractory patients (Fig. 2B).

Recursive partitioning using unbound AUC and disease history showed that AUC ≥ 780 h*ng/mL was associated with better response for newly diagnosed patients while relapsed/refractory patients required an AUC ≥ 1,690 h*ng/mL to achieve similar efficacy (Fig. 3). When dosing scheme (bolus or hybrid) was added to the partitioning model as a potential explanatory variable, the partitioning algorithm assessed it of low importance and did not include it in the classification tree.

The incidence of 11 preselected AEs (ALT, AST, bilirubin, cardiac-related, creatinine, diarrhea, mucositis, nausea, rash, TLS, and vomiting) with severity grade of ≥ 3 were relatively rare with AST elevations showing the highest incidence (8.7%). There were no trends detected for increasing incidence of any of the AEs across increasing quartiles of flavopiridol-unbound AUC exposure (Table 2).

Using the recursive partitioning results, newly diagnosed patients were assigned a 38% probability of clinical response for an AUC < 780 h*ng/mL and 77% for an AUC ≥ 780 h*ng/mL. Relapsed/refractory patients were assigned a 38% probability of clinical response for an AUC < 1,690 h*ng/mL and 77% for an AUC ≥ 1,690 h*ng/mL. Mean clinical probabilities over 500 simulations from the assigned baseline probabilities were generated over the different dosing scenarios. The bolus MTD of 50 mg/m²/day was associated with clinical response probabilities of 42% or 63% for relapsed/refractory or newly diagnosed patients, respectively (Fig. 4). Using the hybrid MTD of 30/60 mg/m²/day, clinical response probabilities of 56% or 74% were simulated for relapsed/refractory or newly diagnosed patients, respectively (Fig. 4).

Several flavopiridol pharmacometric models have been published. Karp and colleagues documented a two-compartment population pharmacokinetic model that used total flavopiridol concentrations from patients with acute leukemias (8). Efforts from Ji and colleagues, Ni and colleagues, and Phelps and colleagues modeled flavopiridol and the glucuronide metabolite of flavopiridol in patients with chronic lymphocytic leukemia (24–26). These models identified relationships between the glucuronide metabolite and tumor lysis syndrome, the role of transporter polymorphisms on disposition and response, and relationships between flavopiridol and response or cytokine-release syndrome.

The analyses herein focused on the pharmacokinetics and pharmacodynamics of total and unbound flavopiridol in acute leukemias and aimed to assess the impact of bolus versus hybrid administration schedules as well as explore differences between newly diagnosed and relapsed/refractory patients. The parameter estimates for the final flavopiridol pharmacokinetic model were comparable with those in the previously published models. Although a statistically significant relationship between BSA and central volume of distribution was detected, no covariates with a clinically meaningful impact on flavopiridol pharmacokinetics were identified or added to the model.

The flavopiridol hybrid dosing schedule was designed by Byrd and colleagues using in vitro and clinical pharmacokinetic data, presumably with the expectation of achieving flavopiridol exposures more efficacious than those achievable with either a bolus, a continuous infusion, or an infusion with a true loading bolus dose (16). This approach was hypothesized to overcome the high plasma protein binding of flavopiridol. Our results showed that there was no evidence of binding saturation over the range of exposures achieved. This is consistent with the in vitro work from Myatt and colleagues that showed binding site saturation was not reached even at 20-fold molar excess of flavopiridol over human serum albumin (HSA; ref. 27). Flavopiridol concentrations at the upper end of the observed range (∼5,000 ng/mL) were equivalent to approximately 0.012 mmol/L, while HSA typically circulates at approximately 0.6 mmol/L. Importantly, our results also showed that the estimated fraction unbound did not change for bolus versus hybrid dosing. Serum albumin, a prognostic biomarker in newly diagnosed and relapsed/refractory AML (28, 29), was not explicitly tested as a pharmacokinetic covariate. However, protein binding, the most plausible mechanism for an effect of albumin on flavopiridol pharmacokinetics, showed low variability between subjects.

Patients with relapsed/refractory disease appeared to require greater flavopiridol exposure (AUC ≥ 1,690 h*ng/mL) than newly diagnosed patients (AUC ≥ 780 h*ng/mL) to achieve the same maximal level of clinical response (77%) in the patient population. In the patients we studied, relapsed/refractory patients were, on average, treated with a higher daily dose of flavopiridol (∼1.4-fold) that resulted in a greater range of flavopiridol unbound AUC (Supplementary Table S2). There were no meaningful differences in pharmacokinetic parameter estimates between newly diagnosed and relapsed/refractory patients suggesting that the differences in response between these patient groups does not have a pharmacokinetic basis.

In newly diagnosed patients, administration of flavopiridol at the maximum tolerated bolus dose of 50 mg/m²/day should be used. Simulations using a 50 mg/m²/day bolus dose showed a median unbound AUC of 927 h*ng/mL, a value in excess of the efficacy improvement threshold of 780 h*ng/mL (Fig. 3) for newly diagnosed patients and predicted to result in a 63% clinical response rate (Fig. 4). In this population of patients, consistent with the results of the phase II FLAM trial directly comparing bolus versus hybrid dosing, there may be little advantage in either efficacy or safety associated with use of the hybrid dosing schedule (Supplementary Table S3). Use of bolus dosing importantly avoids the added inconvenience of the hybrid schedule and the additional energy and effort it requires of the pharmacy, nursing staff, and patients.

Relapsed/refractory patients should use the maximum tolerated hybrid flavopiridol dose of 30/60 mg/m²/day. Simulations using the hybrid MTD showed a median unbound AUC of 1,631 h*ng/mL, an exposure essentially at the efficacy improvement threshold of 1,690 h*ng/mL for relapsed/refractory patients (Fig. 3). In this population, the added burden of the hybrid schedule is warranted as it improves the likelihood of clinical response for these patients from 42% to 56% (Fig. 4).

This work was associated with some limitations. AE information was only available as worst grade per cycle. This prevented a strict time-matching of adverse events to each of the specific sequentially administered drugs in the FLAM cycle.

A population model was developed that robustly characterized flavopiridol pharmacokinetics and the constant linear relationship between total and unbound flavopiridol concentrations. Exposure–response analyses helped define the utility of bolus flavopiridol dosing for newly diagnosed patients and hybrid dosing for relapsed/refractory patients to maximize efficacy in each population. Future studies evaluating these insights are recommended. A randomized phase II study of FLAM (flavopiridol hybrid 30/60 mg/m²/day) versus cytosine arabinoside/mitoxantrone (AM) treatment in relapsed/refractory AML patients is currently under development.

Alvocidib is currently being developed by Tolero Pharmaceuticals, Inc. Tolero Pharmaceuticals, Inc. is a client of QuintilesIMCTM. J. Karp is a consultant/advisory board member of the DMSC (chairperson) for an ongoing alvocidib trial via QuintilesIMCTM. This current work was supported by the National Cancer Institute and not Tolero Pharmaceuticals, Inc. No potential conflicts of interest were disclosed by the other authors.

The contents of this study are solely the responsibility of the authors and do not necessarily represent the official view of the Johns Hopkins ICTR, NCATS or NIH.

Conception and design: V. Ivaturi, J. Karp, M.A. Rudek

Development of methodology: V. Ivaturi, J. Gobburu, A. Doyle, M.A. Rudek

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.M. Greer, J.J. Wright, J. Karp, M.A. Rudek

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. LaCerte, V. Ivaturi, J. Gobburu, J. Karp, M.A. Rudek

Writing, review, and/or revision of the manuscript: C. LaCerte, V. Ivaturi, A. Doyle, J. Karp, M.A. Rudek

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Ivaturi, J.J. Wright, M.A. Rudek

Study supervision: A. Doyle, M.A. Rudek

Other (monitored clinical trial(s) of patients whose samples were evaluated): J.J. Wright

The authors thank the Johns Hopkins Sidney Kimmel Cancer Center nursing staff for superb medical care, and the patients and their families, without whose partnership we could never have conducted the trial and from whom we have learned critical information that will help us to improve the treatment of these diseases. The authors also thank Dr. Gary Rosner for his scientific input.

This work was supported in part by NCI Cooperative Agreement U01 CA070095, and UM1CA186691 (to JEK and M.A. Rudek) and by the Analytical Pharmacology Core of the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins (NIH grants P30 CA006973 and UL1 TR 001079). Grant number UL1 TR 001079 was from the National Center for Advancing Translational Sciences (NCATS) a component of the NIH, and NIH Roadmap for Medical Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.