Background
FMS-like tyrosine kinase 3 (FLT3) is a class III receptor tyrosine kinase. Normally involved in haematopoietic stem cell development, the protein also plays a significant role in acute myeloid leukaemia (AML). Activating mutations in the gene encoding FLT3, and overexpression of FLT3, are commonly associated with AML. These mutations often involve duplications that occur at (rarely) or upstream of (more often) the kinase domain (internal tandem duplications, ITDs [
1,
2]). Such insertions increase the flexibility of the protein’s hinge domain, which in turn drives the mutated kinase towards an active state [
3,
4]. Inhibition of FLT3 is one of the approaches that are considered in treating the disease in such cases [
5]. Indeed, three FLT3 inhibitors were approved for AML treatment: midostaurin (Rydapt
®), gilteritinib (Xospata
®) and quizartinib (Vanflyta
®, approved in Japan).
The paradigm of targeted cancer therapy calls for the use of highly specific drugs that maximise therapeutic benefits and minimise the risk for toxicity. This works well in cancers such as chronic myeloid leukaemia (CML), where a single molecular target is decisive for tumour proliferation [
6]. The situation appears to be more complex in AML, where drug combinations are suggested to be used even when FLT3 is overexpressed and mutated [
7]. The FLT3 inhibitor midostaurin is widely used as a drug in FLT3-mutated AML. In spite of its capability to inhibit multiple targets, midostaurin is not used as a single-agent but together with chemotherapy. Apparently, midostaurin is not efficient enough in itself due to binding to plasma proteins [
8], and its usefulness is due to inhibiting multiple kinases. Multiple kinase inhibitors act on multiple targets, which might be an asset as such inhibitors can be more effective. Unfortunately, this often comes with the associated cost of more severe side effects since even house-keeping kinases are affected. To make things even more complex, some cellular proteins may even have a protective role against the tumour; such proteins should of course not be inhibited [
9].
Gilteritinib is a kinase inhibitor with a narrower range of targets, and is used as monotherapy for refractory FLT3-mutated AML [
7]. Quizartinib has been developed as a specific FLT3 inhibitor and was found to be highly effective [
10]. Unfortunately, resistance is quickly developed in AML patients when treated with quizartinib, often due to secondary mutations in the gene encoding FLT3 [
11]. Pexidartinib (Turalio
®) has shown efficacy against FLT3-ITD carrying the F691L resistance mutation, but other secondary mutations make the tumour resistant to pexidartinib as well [
12]. Biophysical and computational studies could explain the reason behind resistance to quizartinib and pexidartinib [
13‐
16] and suggest new means of inhibition [
17]. Furthermore, efforts are being made to develop additional FLT3 inhibitors [
18]. One of the interesting ideas in this respect is the development of dual-specificity inhibitors that target not only FLT3 but also other molecular targets [
19]. However, in this case there is a risk for toxicity and even reduced efficacy if biological mechanisms that act against the tumour are inhibited. It may be possible to tune the use of multiple inhibitors based on biomarkers [
20‐
22], while also considering pharmacokinetics [
23]. However, this requires precise knowledge on the expression of multiple targets and the degree of their inhibition by drugs, which is not commonly available.
Recently, it has been suggested that the emergence of drug resistance could be postponed by use of drug rotation rather than drug combination. The underlying idea is that by using two drugs that have a high affinity to the same target but non-overlapping resistance mutation profile (i.e., there are resistant mutations that are unique to each drug), it should be possible to postpone the emergence of resistance that would manifest in the clinic. A knowledge-based computational study suggested that the right combination of drugs and rotation times would indeed be useful to this aim in CML [
6,
24]. Here, we examine whether a rotation between quizartinib and pexidartinib might be useful for postponing the emergence of resistance in AML, using two AML cell lines that express FLT3-ITD (MOLM-14 and MV4-11). Knowledge based computer simulations are used to interpret the results and survey the mutational landscape in the tumour.
Materials and methods
Reagents
Quizartinib was purchased from AdipoGen Life Sciences. Pexidartinib was purchased from Selleck Chemicals LLC. MTS, (3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium) was purchased from Promega. Dimethyl sulfoxide (DMSO), fetal bovine serum (FBS), Gibco cell culture media (RPMI-1640 and IMDM) and antibibiotics (1% penicillin-streptomycin, Pen-Strep) were bought from Thermo Fisher Scientific.
Cell cultures
MV4-11 and MOLM-14 were a generous gift from Prof. Stefan Fröhling, National Center for Tumor Diseases, Heidelberg, Germany. MV4-11 was maintained in 10% FBS and 1% Pen-Strep in IMDM while MOLM-14 was maintained in 10% FBS and 1% Pen-Strep in RPMI-1640. Cells were maintained at \(0.5 \times 10^{6}\) to \(1.5 \times 10^{6}\) cells/mL and split saturated culture every 2 to 3 days by diluting for 2 to 3 times. Incubation was performed at T = 37\(^{\circ }\)C and 5% CO\(_2\).
Cell viability assays
Exponentially growing cells were seeded at a density of 50000 (MV4-11) or 30000 (MOLM-14) cells per well in a 96-well plate and treated with DMSO (control) or a dose titration of inhibitors (pexidartinib and quizartinib). MTS was added to the cells 48 hours following the addition of inhibitors. In living mammalian cells, MTS is converted to a formazan compound that has an absorbance maximum at 490 nm. The resulting signals at 490 nm were measured after two hours using a fluorescence plate reader. Viability was calculated as a percentage of cells treated with DMSO. Numerical IC50 values were calculated with non-linear best-fit regression analysis using the Prism 5 software. The model was:
$$\begin{aligned} f(x) = max - \frac{max-min}{1+\left( \frac{x}{IC50}\right) ^\alpha } \end{aligned}$$
(1)
where
x is the concentration of the drug,
f(
x) is the measurement,
max and
min are the maximum and minimum values that are measured, and
\(\alpha\) is the Hill slope.
The MTS assay was performed according to the supplier’s recommended protocol. The dye was kept frozen and thawed by leaving it at room temperature for 90 min. 20 \(\upmu\)L of MTS solution was added to each well and left for incubation in the dark for 1–4 h, before measuring the absorbance at 490 nm.
Drug rotations in MOLM-14 and MV4-11 cell lines
To study the effect of inhibitors on MOLM-14 and MV4-11 cells, these cell lines were grown in the presence of quizartinib and pexidartinib at a concentration that matched the IC50 value for the drugs (quizartinib: 0.384 nM for MOLM-14, 0.313 nM for MV4-11; pexidartinib: 51.4 nM for MOLM-14, 37.9 nM for MV4-11). In addition, we studied the effect of a higher concentration of pexidartinib (163.1 nM for MOLM-14 and 530.2 nM for MV4-11, corresponding to the IC90 values) on the cells. To differentiate between the two treatments, i.e., between pexidartinib (IC50) and pexidartinib (IC90) we refer to the latter as pexidartinib*. Of note, a phase I/II clinical study pointed out that achieving such degree of inhibition in patients is possible with pexidartinib [
25].
Adaptation experiments were run in 24-well plates, which were set up as shown in Table
1. Each well was set up with
\(10^5\) cells in 1 mL medium with the appropriate inhibitor. The drugs were added once, at the start of each 6-day period, to which we refer as “generation”. The number of cells was counted every two days. After 6 days, the medium in all wells was replaced with fresh medium; in the control wells, the inhibitors were unchanged, and in the rotation wells, the inhibitors were changed as shown in Table
1. At the same time, the culture was diluted such that, at the start of each generation, each well contained
\(10^5\) cells in 1 mL medium. Each set of cells was grown in duplicate separate wells, and sampling was performed twice for each well, yielding four points of data for each measurement. Growth rates were calculated at day 3 and day 6 using the
ratrack
tool [
26],
https://github.com/Sandalmoth/ratrack.
Table 1
Schedule for treatment-rotation experiments
1 | Pexidartinib | Pexidartinib* | Quizartinib | Pexidartinib* | Quizartinib | Pexidartinib* |
2 | Pexidartinib | Pexidartinib* | Quizartinib | Quizartinib | Pexidartinib* | Quizartinib |
3 | Pexidartinib | Pexidartinib* | Quizartinib | Pexidartinib* | Quizartinib | Quizartinib |
4 | Pexidartinib | Pexidartinib* | Quizartinib | Quizartinib | Pexidartinib* | Pexidartinib* |
5 | Pexidartinib | Pexidartinib* | Quizartinib | Pexidartinib* | Quizartinib | Quizartinib |
6 | Pexidartinib | Pexidartinib* | Quizartinib | Quizartinib | Pexidartinib* | Quizartinib |
Figures | | | | | | |
A Luna-II Automated Cell Counter was used together with counting slides manufactured by Logos Biosystems for all cell counts. The cell counting procedure was as follows. First, the culture medium in the well was gently mixed. Thereafter, 10 μL of culture medium was extracted from the well, and mixed with 10 μL trypan blue in a 1.5 mL Eppendorf tube. After mixing the medium and trypan blue, 10 μL of the mixture was deposited onto a disposable cell counting slide, which was used in the cell counter.
Simulations
Simulations of cell population developing resistance were run using the
wollsey
package [
6], which is freely available for download at
https://github.com/Sandalmoth/wollsey-public. The protocol was the same as in [
6]. Briefly, given a starting population of cells that do not carry any resistance mutations and a table listing mutations and their IC50 values, the development of mutations under treatment is followed stochastically. A treatment protocol is also included. Here, the protocol matched the experiments (treatment by either of the drugs alone or rotations thereof).
Given that we model an active form of cancer, the number of cells in the simulation was kept roughly constant at
\(1.0 \times 10^6\) cells, which enables to follow on the development of resistance. The probability for mutation was set to
\(1.0 \times 10^{-7}\) per base pair, consistent with estimations in blood cancers and about two order of magnitude larger than the mutation rate in normal cells. The simulations were run
\(1.0 \times 10^5\) times to ensure sufficient statistics. The growth rate was set to once per day and the death rate to 1/10 of the growth rate (there is a spring parameter that ensures that the population of cells do not overgrow). Relative IC50 values per mutation are given in Table
2. Rotation between the drugs was set to have the same period as in the experiments (6 days).
Table 2
Fold-IC50 values for FLT3 mutations relative to FLT3-ITD
None | 1 | 1 |
F691L | 329 | 3 |
D835A | 10 | 18 |
D835E | 6 | 9 |
D835F | 1474 | 415 |
D835G | 10 | 13 |
D835H | 45 | 40 |
D835I | 718 | 1937 |
D835N | 7 | 10 |
D835V | 563 | 320 |
D835Y | 183 | 206 |
D835Del | 320 | 121 |
D839G | 6 | 27 |
Y842C | 106 | 48 |
Y842H | 58 | 49 |
Discussion
The aim of this study was to examine whether rotating between two drugs with a different (yet partially overlapping) mutational profile could postpone the emergence of resistance in FLT3
\(^+\)-AML. To this end, we developed and applied a rotation protocol in AML cell lines, whereby drugs are periodically switched. Postponing the emergence of drug resistance is highly desired in many cancers, and is a matter of urgency in AML. Aggressive chemotherapy (cytarabine and either daunorubicin or idarubicin) is the current standard of care for AML, often together with the kinase inhibitor midostaurin for patients that carry activating FLT3 mutations [
32]. Unfortunately, whereas AML is more likely to occur at older age, older patients are at higher risk when undergoing such intensive treatment. In addition, the cancer may become refractory to chemotherapy, leaving FLT3-inhibitors such as gilteritinib or quizartinib the only treatment option [
8]. Development of resistance mutations is therefore disastrous.
Examination of total growth and growth rates revealed that AML cells quickly became resistant to quizartinib and pexidartinib, as shown by their growth rates increasing with time (Figs.
2,
3). Resistance to pexidartinib was particularly fast to develop. The process was slower with a higher concentration of pexidartinib, but the cells gradually overcame the drug and their growth rates were increased. Resistance to pexidartinib at IC90 was clearly generation-dependent, indicating that mutants that confer resistance became increasingly more common in the cell population.
From an evolutionary standpoint, treating a tumour leads to very strong impairment of its fitness [
33,
34], as cells are not able to divide and the population shrinks. Mutations that provide resistance to the drug suddenly gain a large benefit when it comes to adaptation. When considering kinase inhibitors, gate-keeper mutations, that involve an amino acid residue at the nucleotide binding site of a kinase are observed quite often [
29]. Examples include T790M in EGFR and T315I in Abl1. The corresponding mutation in FLT3 is F691L. This common resistance mutation makes the tumour insensitive to quizartinib, whereas treatment with pexidartinib might still be effective. The presence of mutations that confer resistance only to one drug but not to the other suggests that both treatment options could be considered. Whereas it is counter-productive to treat patients with two drugs that target the same molecular target in the same way, a study of CML therapy suggested that rotating between two drugs might be effective to postpone resistance [
6]. Following our study, this approach is not recommended for treating AML as none of the rotation protocols performed very well. What is more, longer treatment with quizartinib made the cells more resistance to high concentration of pexidartinib, compared to quizartinib-naïve cells.
The results of a Phase I/II clinical study of pexidartinib have recently been reported [
25]. As a single agent, pexidartinib was shown to be less effective than quizartinib or gilteritinib and the authors suggested that this might be attributed to prior treatment with one of these agents. Our study shows that resistance develops rapidly against pexidartinib and that despite the relative potency of pexidartinib against FLT3-ITD/F691L, development of resistance mutations is likely a major reason for relapse. Interestingly, the simulations suggested that it is not a single mutation that is observed but rather a combination thereof (Fig.
9), in all cases. Thus, patients that relapse on quizartinib are unlikely to benefit from pexidartinib.
Encouragingly, simulations of cell growth agreed very well with the experiments. Analysis of the simulations revealed that the cell developed resistance under all conditions and that rotation protocols did not yield any additional benefits. Using a higher concentration of the drug seemed to increase the median WT\(_{1/2}\), that is the median time until resistance mutations (of any kind) dominated the population. In principle, this might indicate that aggressive treatment is best even when considering targeted therapies. While this may indeed be the case, the dosage of drugs is often limited by toxicities, so that the clinician is seldom at liberty to choose a higher dose.
Considering the mutations themselves, the simulations suggest F691L to be common even for pexidartinib. This appears to be puzzling as there are more efficient resistance mutations especially against pexidartinib. Of note, the molecular mechanism causing mutations is blind to the effect of these mutations, and any increase in the fold-IC50 may be enough for the mutation to gain a fitness advantage. There are multiple ways to modify the TTT
codon for Phe in residue Phe\(^{691}\) to Leu, making the mutation likely to occur more often by chance (TTA
, TTG
and CTT
all encode for Leu). In addition, it should be mentioned that there are multiple mutations in residues Asp\(^{835}\) that lead to resistance; taken together, mutations in Asp\(^{835}\) are more common than F691L.
Conclusions
Experiments with two AML cell lines confirm the notion that monotherapy with highly specific FLT3 inhibitors alone is subject to resistance. Resistance was more pronounced with each generation, revealing an evolutionary mechanism that likely depends on resistance mutations. Different rotation protocols were not successful in driving the cells towards a treatable state. Simulations of cell growth agreed with the experimental measurements and predicted that the gatekeeper mutation F691L is the most common single resistance mutation to be observed upon using any protocol involving quizartinib, and that it will commonly be observed even under treatment with pexidartinib. Other measures, such as combination therapy may be more promising; for such treatment to succeed it is however necessary to consider different pathways that promote cell proliferation. Finally, given the central role of FLT3 in AML, novel inhibitors that will be more robust to resistance mutation (such that only very few mutations would lead to resistance) may also be of great value. This is since, as the computational modelling shows, when more mutations are possible resistance develops faster.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.