Integrating in vitro sensitivity and dose-response slope is predictive of clinical response to ABL kinase inhibitors in chronic myeloid leukemia

BCR-ABL kinase domain mutations represent a common mechanism of resistance to ABL tyrosine kinase inhibitors (TKIs) in chronic myeloid leukemia (CML). In vitro cellular 50% inhibitory concentration (IC₅₀) values have been proposed to guide TKI treatment selection for specific mutations.¹  However, using peak concentration (C_max)/IC₅₀ as a measure of potential in vivo activity failed to show a correlation with complete cytogenetic response (CCyR) rates in CML patients.² 

Importantly, an IC₅₀ value constitutes only one point on the dose-response curve for a given drug. Most dose-response curves can be described by Hill’s equation (equation 1), which incorporates both IC₅₀ and slope (m) parameters:

Here, f_a and f_u are cell fractions affected and unaffected by treatment, respectively (f_u = 1 − f_a), and D is drug dose. Theoretical and clinical importance of evaluation of the slope in addition to IC₅₀ has already been shown for antiretroviral drug resistance in HIV infection.³ 

We report an estimation of the slope of in vitro dose-response curves for wild-type and kinase domain–mutant BCR-ABL against clinical ABL TKIs for CML and examine the value of this incorporated parameter for predicting clinical response.

Dose-response curves for imatinib, nilotinib, and dasatinib were determined previously by methanethiosulfonate-based cell viability assay in Ba/F3 cells expressing wild-type or kinase domain–mutant BCR-ABL.⁴  Because it was completely insensitive to all 3 ABL TKIs tested, the BCR-ABL^T315I mutant was excluded from our analysis.

Logarithmic transformation of the Hill’s equation reaches:

The parameters m and IC₅₀ were determined for each mutation and drug by fitting equation (2) to the respective dose-response curve using the least-square-sum criterion. Inhibitory potential at peak concentration (IPP)³  was subsequently calculated as:

Here, D is mean C_max in plasma as reported.² 

IPP and IC₅₀ values for each Ba/F3 BCR-ABL mutant were compared with previously reported CCyR rates for nilotinib⁵  and dasatinib.⁶  Response data for mutations reported in more than 2 patients was included, divided based on mutation IPP and IC₅₀ values, and CCyR rates were compared between groups by 2-tailed Student t test with unequal variance (P = .05 significance threshold). Multivariate analysis was performed by linear multiple regression and the Cox proportional hazard model using JMP-SAS version 10 software (see supplementary material on the Blood Web site for details).

We fitted Hill’s equation to Ba/F3 cell viability dose-response curves for imatinib, nilotinib, and dasatinib for wild-type BCR-ABL and each of 15 BCR-ABL kinase domain point mutants (see representative curves in supplemental Figure 1; all data reported in reference 4). Excellent goodness of fit (r² values = 0.94-0.99) was observed for all drug-mutation pairings. Resultant values of IC₅₀ and slope for each case are summarized in Table 1, along with calculated IPP values (see equation [3] in Methods). IPP provides a natural way to combine drug efficacy data in vitro (ie, IC₅₀ and slope) with clinical pharmacokinetic data and compare them with clinical outcomes.

Most P-loop mutations are reported to render a worse response to imatinib.⁷  We found that 4 of 7 P-loop mutations tested (G250E, Y253H, E255K, E255V) showed a lower dose-response slope relative to wild-type BCR-ABL in addition to high IC₅₀ (>1100 nM), whereas all other mutations showed variably increased slopes (Table 1). Consistent with particularly negative effects of these mutations on drug binding and clinical outcome with imatinib, their lower slopes indicate shallower drug efficacy over a given increase in concentration. Differences in slope values across different resistant mutations likely reflect a varied degree of inhibitor-binding destabilization (rather than binding preclusion). Furthermore, the range of IPPs for these mutations was lower than (and not overlapping with) all other mutations (0.084-1.66 vs 2.93-5.59; Student t test for range: P = 6×10⁻⁶).

Additional value of the slope parameter was particularly apparent in cases where increased in vitro IC₅₀ does not track with clinical resistance. For example, in comparing the G250E and V379I mutants, which feature comparable cellular IC₅₀ values for imatinib (1184 and 1140 nM, respectively), only G250E harbors worse clinical prognosis, arguably because of its lower dose-response slope. This is also reflected in a lower IPP value compared with V379I (Table 1). Similarly, the P-loop mutation M244V does not confer marked clinical resistance to imatinib despite increased IC₅₀ relative to wild-type BCR-ABL, possibly because of an exceptionally high slope value (m = 5 vs m = 1.87 for wild-type BCR-ABL) reflecting a very steep dose-response curve (Table 1). Although at 6000 nM the IPP values for the M244V mutant and wild-type BCR-ABL become virtually identical (5.70 and 5.77, respectively), this concentration is below imatinib C_max for some individuals but not for others,⁸  suggesting that patients with this mutation may be particularly vulnerable to consequences of unfavorable imatinib pharmacokinetic profile or reduced compliance. Indeed, one study found that clinical resistance to imatinib in 3 of 6 patients with a M244V mutation was overcome by dose increase.⁹ 

Particularly high slope values (m>4) were found for M244V and Y253F with nilotinib treatment, in contrast to lower values for both mutations with dasatinib treatment (Table 1).

Comparison of BCR-ABL mutant IPP values between nilotinib and dasatinib revealed favorable, higher IPP for dasatinib over nilotinib against the Y253H mutant; nilotinib demonstrated higher IPP for E255K and F317L compared with dasatinib. For the E255V mutation, both drugs featured low IPPs, consistent with this mutation being reported in clinical failures of both drugs.^5,6 

We next examined whether higher IPP values are predictive of better clinical response. Patient response data were initially divided according to the median mutant IPP value for each TKI. Dasatinib-treated patients with mutations resulting in IPP values above the median (IPP = 8) had a significantly higher mean CCyR rate than patients with mutation IPPs below the median (53% [range, 39-80] vs 31% [range, 7-56]; P = .038). Notably, using an IPP threshold value of 7 resulted in even better group separation of mean CCyR rate (P = .007; Figure 1A and supplemental Figure 2 for P values for a range of threshold values). In contrast, this relationship was not evident for mean CCyR rates when IC₅₀ alone was used as the comparator (P = .83 for IC₅₀ values above vs below the median IC₅₀; Figure 1B).

Upon comparison of nilotinib-treated patients by either mutation IPP or IC₅₀ value, the difference in mean CCyR rate between patients with values above or below the median approached but did not reach statistical significance (P = .055 for both cases; Figure 1C-D). Notably, however, both the number of mutations and overall number of nilotinib-treated patients with published available mutation-response data were smaller than for dasatinib (7 vs 11 mutations, 65 vs 295 patients with mutations, respectively).

Multivariate analysis by 2 methods also demonstrated that a model including IPP and TKI type significantly correlates with CCyR (P < .001) and outperforms a model that includes IC₅₀ and slope as separate variables based on the model significance, covariate significance, and Akaike information criterion (supplemental Table 1).

In conclusion, we show that in vitro dose-response slope provides important additional information regarding efficacy of ABL TKIs beyond in vitro IC₅₀ values: lower slopes are indicative of increased resistance, and very high slopes result in a steep dose-response curve, presenting potential challenges in cases of low compliance or unfavorable pharmacokinetic profile. Calculation of IPP (based on IC₅₀, slope, and clinical plasma C_max) allows for separation of patients with respect to rates of CCyR with dasatinib, unlike using in vitro IC₅₀ alone.²  Further implementation of our results for improvement of mutationally-guided TKI treatment selection in CML will require incorporation of additional variables such as plasma protein binding, drug clearance and distribution, and validation using independent large patient databases.

The authors thank Dr Ilan Roziner for his help with statistical analysis of the data.

Contribution: V.V. and B.J.D. designed research; T.O. and B.J.D. contributed all in vitro experimental data; V.V. and C.A.E. performed the research; O.S. performed statistical analysis; and V.V., C.A.E., and B.J.D. wrote the manuscript.

Conflict-of-interest disclosure: V.V. is employed by Neumedicines, Inc., as a clinical advisor. B.J.D. is currently principal investigator or co-investigator on Novartis and Bristol-Myers Squibb clinical trials. B.J.D.’s institution has contracts with these companies to pay for patient costs, nurse and data manager salaries, and institutional overhead. B.J.D. does not derive salary, nor does his laboratory receive funds, from these contracts. Oregon Health & Science University (OHSU) and B.J.D. have a financial interest in MolecularMD. OHSU has licensed technology used in some of these clinical trials to MolecularMD. B.J.D.’s potential individual and institutional conflict of interest has been reviewed and managed by OHSU.

Correspondence: Vladimir Vainstein, Department of Hematology, Hadassah Medical Center, Ein Kerem, Jerusalem, Israel; e-mail: vladimir.dr@gmail.com.