Sage Journals: Discover world-class research

Abstract

Over the past years, improvements in high-throughput screening (HTS) technology and compound libraries have resulted in a dramatic increase in the amounts of good-quality screening hits, and there is a growing need for follow-on hit profiling assays with medium throughput to further triage hits. Here the authors present such assays for the colony-stimulating factor 1 receptor (CSF1R, Fms), including tests for cellular activity and a homogeneous assay to measure affinity for inactive CSF1R. They also present a high-throughput assay to measure target residence time, which is based on competitive binding kinetics. To better fit k_off rates, they present a modified mathematical model for competitive kinetics. In all assays, they profiled eight reference inhibitors (imatinib, sorafenib, sunitinib, tandutinib, dasatinib, GW2580, Ki20227, and J&J’s pyrido[2,3-d]pyrimidin-5-one). Using the known biochemical selectivities of these inhibitors, which can be quantified using metrics such as the selectivity entropy, the authors have determined which assay readout best predicts hit selectivity. Their profiling shows surprisingly that imatinib has a preference for the active form of CSF1R and that Ki20227 has an unusually slow target dissociation rate. This confirms that follow-on hit profiling is essential to ensure that the best hits are selected for lead optimization.

Keywords

cell function assays cell-based assays kinases enzyme assays or enzyme kinetics structure-activity relationships

Introduction

Colony-stimulating factor 1 receptor (CSF1R; also known as Fms) is an important kinase in the inflammatory process, and it regulates homeostasis and differentiation of dendritic cells and macrophages.^1,2 CSF1R is seen as a good drug target for rheumatoid arthritis as it was observed that knockout and CSF1-deficient (op/op) mice are resistant to induction of arthritis by collagen¹ and because CSF1 can exacerbate arthritis in such animal models. Moreover, CSF1 receptor (CSF1R) inhibitors have shown a positive response in arthritis models.^3–6 Because of the role of macrophages in tumor–stroma interactions and in tumor-induced immune suppression, CSF1R is also considered a good drug target in oncology.^3,6

CSF1R is a transmembrane tyrosine kinase with cKit, PDGFR, and Flt3 as its closest homologs. CSF1R is activated by CSF1 binding, which leads to receptor autophosphorylation and downstream signaling, including extracellular signal-regulated kinase (ERK) activation.⁷ Quite a few inhibitors have been developed that target the cytoplasmic kinase domain of CSF1R, such as the clinically approved, spectrum-selective compounds imatinib, dasatinib, nilotinib, pazopanib, sorafenib, and sunitinib.^3,4,8 More selective inhibitors are tandutinib⁹ (MLN-518, phase II); PLX-5622, PLX-3397, and JNJ-28312141 (all in phase I); and the tool compounds GW-2580,⁵ Ki-20227,⁶ and a pyrido[2,3-d]pyrimidin-5-one from Johnson & Johnson (Spring House, PA), labeled here as J&J37¹⁰ ( Table 1 ). Nevertheless, selective drug-like CSF1R inhibitors that do not, for instance, also inhibit cKit remain rare ( Table 2 ).⁸ For this reason, we initiated a screening campaign on CSF1R.

Table 1.

Chemical Structures of Colony-Stimulating Factor 1 Receptor (CSF1R) Reference Inhibitors

Table 2.

Activities of Colony-Stimulating Factor 1 Receptor (CSF1R) Reference Inhibitors in the Described Assays

	IMAP Biochemical EC₅₀s (nM)						Selectivity (290 Kinases^a)			Cellular EC₅₀s (nM)
Trivial Name	CSF1R	PDGFRβ	AurA	cKit	Flt3	Abl	#pEC₅₀ <3 µM	Selectivity Entropy²²	Rank (Entropy)	NFS-60	NS0	pCSF1R	pERK
Imatinib	21	282	>10 000	79	>10 000	28	19	1.05	2	3890	>10 000	229	759
Sorafenib	15	13	>10 000	78	34	2340	52	2.15	5	537	>10 000	123	490
Dasatinib	2	20	3240	1	>10 000	<10	82	3.21	6	49	316	1	6
Sunitinib	5	91	>10 000	3	<10	4900	166	2.05	4	1122	1550	49	65
Tandutinib	52	257	>10 000	55	89	>10 000	18	1.58	3	>1000	>10 000	309	219
GW-2580	13	>10 000	>10 000	>10 000	>10 000	>10 000	4	0.26	1	1950	>10 000	38	30
Ki-20227	3	46	3020	10	2290	>10 000	ND	ND	ND	110	>10 000	5	20
J&J-37	2	129	4900	5	9	960	ND	ND	ND	141	1320	2	1

ND, not determined. Selectivity values were taken from literature.^8,22

The 290 kinases comprise the panel in Karaman et al.,⁸ excluding mutants.

Over the past years, the field of high-throughput screening (HTS) has substantially progressed through increased quality of compound libraries, more robust assay formats, and advances in the science of screening.¹¹ This has resulted in a dramatic increase in the number of high-potency hits that can be found in regular screening campaigns, such as on CSF1R. Although more hits give more choice of which chemistry to progress, hit properties need to be discriminated by further data from orthogonal assays—for instance, on cellular activity, selectivity, and binding kinetics. In this way, hits can be triaged on an empirical basis. Ideally, a completed screening campaign therefore delivers not only an EC₅₀ of target inhibition but also a data package, after which progressible chemical series are selected ( Fig. 1 ).

Fig. 1.

Flowchart for the colony-stimulating factor 1 receptor (CSF1R) screening and hit profiling campaign.

Because large numbers of hits are sometimes found (~1000), secondary assays need to be medium throughput and labor extensive. However, for many important hit parameters, such assays are not available. For instance, it is important to identify highly selective scaffolds early, but the “normal” way to assess selectivity is by profiling in a panel of more than 100 separate biochemical assays, which is cost-prohibitive for more than a few hits.¹² Because selective compounds often preferentially bind to unique kinase conformations, it has been proposed that a high binding affinity to the inactive kinase form can predict selectivity.^13–15 Although this is easier to test in high throughput, the predictive value of such a readout is only supported by anecdotal successes.

Another parameter that is difficult to assess in higher throughput fashion is compound residence time, which is related to the dissociation rate of the compound from its target (k_off). A slow dissociation rate may lead to improved in vivo efficacy and can compensate for unfavorable pharmacodynamics.^16,17 It can also be the basis for compound selectivity.¹⁶ Therefore, an unusually long residence time is considered a beneficial hit property.¹⁸ However, binding kinetics are usually measured by surface plasmon resonance, nonequilibrium reaction kinetics, or changes in intrinsic fluorescence, which are low throughput, require access to specialist technologies, or are not generally applicable.^16–18 There remains a need for a more high-throughput and universal way of measuring compound disassociation rates.

Here we present a CSF1R screening campaign and the orthogonal assay panel that followed the primary assay, which incorporates medium-throughput assays for cellular activity, binding to inactive CSF1R, and compound residence time. Validation of this assay panel with reference inhibitors indicates which readout best predicts selectivity. This CSF1R assay portfolio is a general case study on how to increase hit value by follow-up screening.

Materials and Methods

Compound synthesis

All compounds used in this study were obtained from commercial sources, except J&J’s pyrido[2,3-d]pyrimidin-5-one (J&J37). This compound was prepared according to procedures described.¹⁰

CSF1R kinase activity assay

For HTS, we used a biochemical kinase activity assay (IMAP format; Molecular Devices, Sunnyvale, CA). In each well of a black 384-well Costar plate (Corning, Corning, NY), 5 µL compound dilution was mixed with 5 µL recombinant CSF1R (cytoplasmic domain; PV3249; Life Technologies, Carlsbad, CA) and preincubated for 30 min, after which 5 µL adenosine triphosphate (ATP) and 5 µL substrate peptide (TAMRA-KKKSPGEYVNIEFG; Molecular Devices) solutions were added. The reaction was allowed to proceed for 120 min at room temperature (reaction progress curve linearity during the incubation time was ensured during assay setup). Component concentrations during the reaction were as follows: 5 nM CSF1R, 50 nM TAMRA-peptide, 112.5 µM ATP (equal to K_M-ATP), 10 mM Tris (pH 7.2), 10 mM MgCl₂, 0.01% Tween-20, 0.05% NaN₃, 1 mM dithiothreitol (DTT), 2 mM MnCl₂, 0.1% bovine serum albumin (BSA), and 0.5% DMSO. The signal from the mixture was developed by adding 40 µL IMAP solution (75% buffer A, 25% buffer B; Molecular Devices), incubating for 60 min, and reading the fluorescence polarization signal (excitation 531 nm, emission 595 nm). Dose–response curves were measured as 10 points in duplicate on two plates (n = 2, N = 2). Sigmoidal curves were fitted using Prism 4.0 (GraphPad Software, La Jolla, CA).

Selectivity profiling

Compound inhibitory activity was measured in a small panel of kinases that frequently cross-react with CSF1R or are important antitargets. The assay setup was similar to the CSF1R biochemical assay with the following differences. The cKit assay uses, in final concentrations, the following: 2 nM cKit (14-559; Millipore, Temecula, CA), 50 nM substrate peptide FAM-GEEPLYWSFPAKKK, and 90 µM ATP. The PDGFRβ assay uses 30 nM PDGFRβ (PV3591; Life Technologies), 50 nM FAM-QEEEYVFIE, and 2.5 µM ATP (note that in this assay, substrate and enzyme concentration are very similar). The Abl assay uses 0.5 nM Abl (P3094; Life Technologies), 50 nM 5FAM-KKGEAIYAAPFA, and 0.4 µM ATP. The Aurora A assay uses 1 nM Aurora A (PV3612; Life Technologies), 50 nM 5FAM-GRTGRRNS, and 15 µM ATP. The Flt3 assay uses 8 mU (~0.2 nM) Flt3 (14-500; Millipore), 50 nM 5FAM-KKKKEEI YFFFG, and 24 µM ATP. In all these assays, the compound preincubation step was 60 min. All ATP concentrations are equal to K_M-ATP.

M-NFS-60 and NS0 proliferation assays

Batches of frozen mouse myeloid MNFS-60 (ATCC no. CRL-1838) cells were prepared to limit the amount of (laborious) cell culture. For each experiment, a batch was thawed and cultured overnight in the absence of mouse–CSF1. The culture medium was RPMI60, supplemented with 10% fetal bovine serum (FBS), 2 mM L-glutamine, penicillin, and streptomycin. The following day, cells were stimulated by adding 11.1 ng/mL mouse–CSF1. In each well of a sterile black 96-well plate, 90 µL (25 000 cells) was dispensed, followed by 10 µL of compound dilution in medium (final DMSO concentration 1%). After 48-h incubation at 37 °C and 5% to 7% CO₂, 10 µL of 50^v/_v% Alamar blue (Serotec, Raleigh, NC) was added, and the cells were further incubated. After 24 h, the fluorescence signal was read (emission: 545 nm, excitation: 590 nm). Dose–response curves were measured as 10 points in duplicate on two plates (n = 2, N = 2) and fitted using Prism (GraphPad Software).

As a control, the proliferation of mouse myeloid NS0 cells (ATCC no. CRL-1827), which grow independently of CSF1, was measured in an assay similar to the M-NFS-60 proliferation. The only difference was that 15 000 cells/well were seeded, and no CSF1 ligand was added.

CSF1R and ERK target engagement assays

Human monocytic THP1 cells (ATCC no. TIB-202) were stimulated with CSF1, after which CSF1R autophosphorylation and ERK1/2 kinase activation were measured. THP1 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM)/F-12 medium (without phenol red) and supplemented with L-glutamine, penicillin, streptomycin, and 10% heat-inactivated FBS. In each well of a 96-well culture plate, 250 000 cells were dispensed (at 2·10⁶ cells/mL) and rested overnight at 7 °C and 6% CO₂. The next day, compound was added to a final DMSO concentration of 0.1%, followed by 30 min preincubation and stimulation with 500 ng/mL of human CSF1 for 5 min, after which the cells were put on ice. Lysis buffer was added (from the PerkinElmer [Waltham, MA] AlphaScreen p-ERK Surefire kit), and cells were lysed for 10 min on a shaker. The lysates were frozen at −80 °C for further use.

For detection of CSF1R phosphorylation, a human phospho-CSF1R duoset kit was used (DYC3268; R&D Systems, Minneapolis, MN). MaxiSorb 96-well plates (Nunc, Roskilde, Denmark) were coated with mouse antihuman CSF1R, blocked, and incubated with 100 µL of THP1 cell lysate for 2 h at room temperature. After washing, the plates were incubated with horseradish peroxidase (HRP)–labeled anti-pTyr. Signal was detected by incubating with chromogen for 20 min and then reading absorbance at 450 nM on an EnVision plate reader (PerkinElmer).

For detection of phosphorylated ERK1/2 levels, 12.5 µL lysate was thawed out and mixed in a 384-well OptiPlate (PerkinElmer) with 12.5 µL of SureFire pERK1/2 (cat. no. TGRES10K; PerkinElmer) bead mix that was prepared according to the manufacturer’s instructions. The mixture was shaken and placed in the dark for 2 h at room temperature. The fluorescent signal, which results from the proximity of two labeled antibodies against ERK1/2 and phoshorylated ERK1/2(pThr202/Tyr204) and diffusion of activated oxygen between them, was quantified in an EnVision plate reader.

For checking specificity of the signal, we ran Western blots in a standard fashion. For primary antibodies, we used anti-phosphorylated ERK1/2 (cat. no. 4377; Cell Signaling, Danvers, MA) or anti-ERK1/2 (cat. no. 700012; Life Technologies); for secondary antibody, we used goat-anti-rabbit HRP (cat. no. W4011; Promega, Madison, WI). Signal was generated using ECL mix (Pierce, Rockford, IL) and captured on a gel imaging system ( Fig. 2 ).

Fig. 2.

Cellular assays for colony-stimulating factor 1 receptor (CSF1R). (A) GW-2580 effects on M-NFS-60 and NS0 cell proliferation. (B) Sorafenib effects on CSF1R and extracellular signal-regulated kinase (ERK) phosphorylation after CSF1 stimulation of THP1 cells. (C) AZD6244 effects in the same assays as panel B. The large difference in EC₅₀ reflects that AZD-6244 is an MEK inhibitor. (D) Top panel: Western blot of lysed THP1 cells with an anti-phosphoERK1/2 antibody (phosphoERK1/2 bands are indicated by labels ERK1 and ERK2). Lane 1: no stimulus. Lane 2: CSF1 stimulus inhibited by GW-2580. Lane 3: CSF1 stimulus only. Lane 4: empty. Lane 5: marker. Bottom panel: control for total ERK. The ERK1 band (not indicated) is poorly visible just above the ERK2 band. Lanes in the bottom blot were digitally shuffled to match the lanes in the top blot.

Sources of active and inactive CSF1R

Active CSF1R was purchased from Life Technologies (cat. no. PV3249) and Millipore (cat. no. 14-551M). Inactive CSF1R was obtained in a small quantity as a gift from Life Technologies; a larger batch was purchased from Millipore. Both inactive batches had equal specific activity after a test activation. Inactive and active CSF1R contain the cytoplasmic part (amino acids 538–910) and are produced in the baculovirus expression system. The inactive material is directly harvested, whereas active material is obtained after incubating the inactive material for 1 h with 5 mM ATP on ice. The phosphorylation status of CSF1R was checked on Western blot using anti-phoshorylated tyrosine antibody ( Fig. 3 ). A rapid activity test showed that the inactive material has ~1% phosphorylation activity of the active material.

Fig. 3.

Compounds have preference for active and inactive colony-stimulating factor 1 receptor (CSF1R). (A) Top panel: Western blot of recombinant CSF1R with anti-phosphotyrosine. Bottom panel: Ponceau S staining of the same blot. Lanes 1, 6: molecular weight markers. 2: CSF1R, activated by incubation with adenosine triphosphate (ATP) and subsequently YopH phosphatase treated. 3: CSF1R activated with ATP. 4: CSF1R harvested from insect cells and YopH phosphatase treated. 5: CSF1R harvested from insect cells. (B) Competitive binding assay of GW-2580 with active (panel A, lane 3) and inactive (panel A, lane 5) CSF1R. (C) Competitive binding assay results of reference compounds. The unit on the y-axis is the ratio of the K_ds (at equilibrium) of binding to active and inactive CSF1R. The scale is logarithmic. The numbers in the panel give the K_d for binding to active and inactive CSF1R, respectively. (D) Direct comparison of imatinib and GW-2580 ligand binding affinities with inactive CSF1R using the affinity mass spectrometry assay system (ALIS). The CSF1R concentration was fixed at 2.5 µM. As the ligand concentration increases, the amount of receptor–ligand complex present will approach the total receptor concentration, producing steep hyperbolic curves. In this method, receptor and ligand concentrations are of similar magnitudes, so ligand depletion effects are explicitly considered by fitting the titration curve data by nonlinear regression to yield direct estimates of K_d (see Supplementary File S1 and references therein).

Competitive binding assay

Compound binding to CSF1R was assessed by a competitive binding assay (LanthaScreen; Life Technologies) based on fluorescence resonance energy transfer (FRET) between a labeled antibody bound to the His-expression tag on CSF1R and a small-molecule tracer, which consists of a kinase inhibitor conjugated to a label (no. 236, cat. no. PV5592). The FRET signal is proportional to the amount of kinase–tracer complex in the well and decreases when an inhibitor displaces the tracer.

CSF1R (active or inactive) was mixed with antibody in reaction buffer (10 mM HEPES [pH 7.5], 2 mM MgCl₂, 0.2 mM EGTA, 0.002% Brij-35), and 5 µL was dispensed in a black 384 shallow-well plate (cat. no. 3676; Corning). Next, 10 µL of inhibitor and tracer was added. Final concentrations of CSF1R and tracer were 5 nM and 30 nM, respectively. The FRET signal was followed in an EnVision plate reader (excitation: 320 nm, emission: 615 nm), and the value at equilibrium was taken for calculating EC₅₀s.

The K_d of tracer binding was established by testing a concentration range of tracer in the absence of inhibitor and fitting a sigmoidal dose–response curve in Prism. The tracer K_d is 32 nM for the active and 346 nM for the inactive CSF1R material. Inhibitor K_ds were derived from EC₅₀s and the tracer K_d by use of the Cheng-Prusoff equation ( Fig. 3 ).

Affinity selection–mass spectrometry (ALIS) assay

Ligand binding to inactive CSF1R was independently evaluated with the affinity selection–mass spectrometry (ALIS) system¹⁹ (see Suppl. File S1). A complex of 2.5 µM inactive CSF1R complexed to inhibitor was prepared in 50 mM HEPES (pH 7.5), 100 mM NaCl, 0.05 mM EDTA, 2 mM MgCl₂, 1 mM DTT, and 5% glycerol, and the mixture was incubated for 30 min at room temperature. The amount of inhibitor bound was subsequently analyzed by size exclusion chromatography followed by mass spectrometry (AS-MS analysis). For full experimental details, see Supplementary File S1.

Off-rate profiling through competitive binding kinetics

For measuring binding kinetics, we used the LanthaScreen competitive binding assay as described above, with a tracer concentration of 15 nM. The compound/tracer mixture was dispensed first, and the reaction was started by adding preequilibrated CSF1R/antibody mixture. Only active CSF1R was used. After rapid mixing, the emerging FRET signal was followed. The time between mixing and the first read (always smaller than 30 s) was tracked on a stopwatch and used to correct the time intervals registered by the reader ( Fig. 4A ).

Fig. 4.

High-throughput target residence time profiling. (A) Fluorescence resonance energy transfer (FRET) signal of the binding complex between tracer and enzyme (LanthaScreen format) in time. The signal is in competition with either Ki-20227 (black) or sunitinib (gray) at the indicated concentrations. Both compounds are equally potent because they reach the same equilibrium signal, but they follow different time courses toward it. (B) k_off parameters for a set of reference inhibitors, determined by competitive binding kinetics. Italic numbers indicate target residence half-lives, as calculated by t_0.5 = ln 2/k_off.

First, we established the binding kinetics of the tracer to active CSF1R. A concentration range of tracer was mixed with 5 nM CSF1R in the absence of inhibitor. The resulting curves were globally fitted as an exponential binding curve in Prism,²⁰ k_on,tracer = 1.0 s⁻¹M⁻¹, k_off,tracer = 0.015 s⁻¹.

For inhibitor kinetics, we fixed the tracer concentration and added inhibitors in a dose range around their K_d. The time-dependent competitive binding curves were fitted using equations proposed by Motulsky and Mahan.²⁰ However, the algorithms in Prism frequently failed to find a solution. To facilitate the numerical optimization, we rearranged the competitive binding equations (see Suppl. File S2).

Y = Q \times [\frac{K_{4} - K_{s}}{K_{s}} (1 - E^{- k_{s} T}) - \frac{K_{4} - K_{F}}{K_{F}} (1 - E^{- k_{F} T})]

Here, k₄ is the inhibitor off-rate (s^–1), K_F and K_S are complex parameters (see Suppl. File S2) composed of the tracer and inhibitor on- and off-rates, t is the time (s), Q is a scaling parameter, and Y is the background-corrected FRET signal measured in the binding assay. Q, K_F, and K_S are dependent on the tracer and inhibitor concentrations; k₄ is not. All curves were individually fitted (e.g., see Fig. 4A ). From curves with sufficient fits (R² values of >0.80), the geometrically average k₄ value was taken. All experiments were performed twice in duplicate.

Results

Primary CSF1R activity assay

To find CSF1R hits that would be suitable starting points for a compound optimization program, we screened a collection of 600 000 compounds in the biochemical CSF1R kinase activity assay. This assay uses the activated cytoplasmic domain of CSF1R ( Fig. 1 ). All compounds were tested at 10 µM, and actives were dose–response tested, resulting in 826 hits (0.14% of the library) with EC₅₀s between 2 and 1000 nM. All reference compounds showed potent inhibition in this assay ( Table 2 ).

Computational clustering

To prioritize the 826 hits, we arranged them computationally in 40 clusters on the basis of structural similarity. Further analysis showed them to be highly diverse. For instance, molecular weight varied between 180 and 728 Da, ligand-binding efficiency varied between 0.009 and 0.035, and cLogP varied between −2.7 and 8.6. The hits were analyzed interactively using Spotfire (TIBCO, Somerville, MA) for further compound selection, and 121 compounds were selected on the basis of their availability in solid form, cLogP <5.9, and potency in the kinase activity assay of better than 500 nM. A further selection step on the basis of chemical diversity and scope for chemical optimization yielded a final selection of 21 compounds.

Despite the good results, we felt that the rigorous selection process ran too much of a chance of missing interesting chemical matter. For instance, no representatives were selected from 16 of 40 clusters because not a single compound from these clusters was present as a solid. In addition, selections based on chemical diversity treat all scaffolds equally, whereas some might have intrinsically favorable (but unknown) properties. For this reason, we decided to set up additional orthogonal assays that would allow us to stick to a larger set of hits and validate these with our reference inhibitor set.

Cellular activity in M-NFS-60 cells

A very important characteristic that we wanted to know of the screening hits was their cellular activity. Therefore, we set up a cell proliferation assay with M-NFS-60 cells, which is a mouse myeloid line that is dependent for its growth on CSF1. It is used frequently to measure CSF1R inhibitor activity.⁵ To make the assay more convenient, we adapted it to work with frozen cells.²¹ Inhibitors EC₅₀s and assay window are similar between cultured and frozen cells (not shown). The final assay is highly reproducible and robust (Z′ factor >0.7, signal-to-background [S/B] ratio >15; Fig. 2A ). As a control for CSF1R-independent compound effects, we used a NS0 mouse myeloid line that grows independently of CSF1.¹² Compounds that act through CSF1R inhibition should be active in the M-NFS-60 assay but not in the NS0 ( Fig. 2A ).

Although these assays are easy and robust, they have two inherent drawbacks. The first is sensitivity. For instance, imatinib and tandutinib are not active in this assay ( Table 2 ), despite being potent biochemical inhibitors of CSF1R and able to target Abl and Flt3 in an in vivo setting.^9,14 The second (possibly related drawback) is that the cell lines contain murine and not human CSF1R.

CSF1R target engagement in THP1 cells

To work with human CSF1R and directly measure the receptor’s intracellular inhibition (target engagement), we set up a CSF1R auto-phosphorylation assay in THP1 cells. THP1 is a human monocytic leukemia cell line that expresses functional CSF1R. CSF1R phosphorylation status after a short CSF1 stimulus was measured in a sandwich enzyme-linked immunosorbent assay (ELISA) format ( Fig. 2B , C ). Although this assay has relatively low throughput, it directly measures CSF1R target engagement. This assay is much more sensitive than the M-NFS-60 assay ( Table 2 ). For example, inhibitors such as Ki-20227 and J&J37 are equally potent in this assay as they are biochemically. The results show furthermore that all reference compounds are active on CSF1R in a cellular context ( Table 2 ).

ERK pathway engagement in THP1 cells

It is known that CSF1 stimulation of CSF1R leads to phosphorylation of ERK1/2.⁷ Conveniently, phosphorylated ERK can be read out in a homogeneous format (Surefire; Fig. 2B , C ), allowing a higher throughput assay. The reference inhibitor results show that the assay delivers potent EC₅₀s ( Table 2 ). To get a reasonable window (S/B ~7), it is important that cells are handled gently and the CSF1 stimulus is short. The signal and its inhibition by reference compound GW-2580 were confirmed on Western blot ( Fig. 2D ). These results confirm that CSF1 stimulation of THP1 leads to activation of ERK and that inhibition of CSF1R after CSF1 stimulation reduces ERK activation.

Conveniently, the CSF1R and ERK activation assays can be performed on the same cell lysates. Therefore, in one experiment, the effects of inhibitors can be probed both at the start (CSF1R activation) and further down (ERK activation) of the CSF1R pathway. For an inhibitor that is active on CSF1R, such as sorafenib, the inhibitory potency is similar in both assays ( Fig. 2B ). In contrast, a selective inhibitor of MEK that acts just upstream of ERK (AZD6244) predictably does well in the ERK assay and poorly in the CSF1R assay ( Fig. 2C ). Because ERK inhibition can be influenced by off-target inhibitory activities, combining the CSF1R and ERK readouts might be a measure of compound selectivity (see below).

Profiles in a small selectivity panel

A second important parameter we wanted know early on in hit triaging is selectivity ( Fig. 1 ). Because hit profiling in large kinase panels is prohibitive in cost and effort for large numbers of hits, we assembled a small panel incorporating PDGFRβ, Flt3, and cKit assays as close sequence homologs of CSF1R. On the basis of full-profiling results, we added Abl as an antitarget because it is associated with cardiotoxic effects, as well as Aurora A because it regulates the cell cycle.

Testing the reference inhibitors in the panel confirms the broad activity of dasatinib and the exquisite selectivity of GW-2580 ( Table 2 ). It also confirms that all studied CSF1R inhibitors, except GW-2580, also inhibit PDGFRβ and cKit. Enhanced selectivity can therefore be a differentiator for new therapeutic CSF1R inhibitors. However, the limited size of the panel makes it difficult to estimate general selectivity of compounds. For instance, tandutinib and dasatinib have similar profiles in Table 2 , but in broad profiling, tandutinib is a much more selective compound.^8,22

Conformational probing

In an attempt to predict hit selectivity more generally, we tested binding to inactive CSF1R. We were encouraged in this experiment after it appeared that GW-2580 binds to the inactive DFG-out conformation of CSF1R,^23,24 suggesting this is the basis of the exquisite selectivity of this inhibitor. Conveniently, X-ray structures have shown that a DFG-out conformation is naturally present in inactive, unphosphorylated CSF1R because in this state, the N-terminal juxtamembrane domain pushes against the kinase domain, enforcing it to adopt a DFG-out conformation.²⁵ Therefore, by testing compound affinity for inactive CSF1R, we can test a compound’s affinity for the DFG-out conformation.

We set up a homogeneous competitive binding assay ( Fig. 3 ) to measure in high-throughput compound binding to active and inactive CSF1R. The affinity of the ATP-competitive tracer for inactive CSF1R (346 nM) is 10 times less than for active CSF1R (32 nM), consistent with inactive CSF1R presenting a DFG-out conformation (for which ATP has less affinity). The differences in tracer affinities were used to correct inhibitor EC₅₀s into inhibitor K_ds. The ratios of compound affinities for active and inactive CSF1R are shown in Figure 3C . GW-2580 and sunitinib, which are known to bind to inactive conformations,^24,26 have the highest relative affinity for inactive CSF1R ( Fig. 3C ). Surprisingly, however, imatinib, which preferentially binds to inactive cKit²⁶ and Abl,¹⁴ in our experiments preferentially binds to active CSF1R even more potently than the known ATP competitor staurosporine.

To verify this, an independent binding experiment was performed on inactive CSF1R, using the ALIS (mass spectrometry–based) method, which in itself is also an HTS platform.¹⁹ This showed that GW-2580 has an 80-nM K_d for inactive CSF1R, whereas imatinib has much less affinity (K_d = 4.7 µM; Fig. 3D and Suppl. File S1). If this is compared to their similar EC₅₀ on active CSF1R in the biochemical assay, this confirms that imatinib has significantly less preference for inactive CSF1R than GW-2580.

Off-rate profiling

A third and final important parameter for selecting hits is their target residence time or target dissociation rates. To measure this in high throughput, we used the LanthaScreen competitive binding assay to measure competitive binding kinetics ( Fig. 4 ).²⁰ After quick simultaneous mixing of tracer and compound, the emerging FRET signal is followed in time ( Fig. 4A ). If the tracer binds slower than the compound, this results in a slow increase in FRET, of which the rate depends on the compound off-rate (e.g., the sunitinib profile in Fig. 4A ).²⁷ If the tracer binds faster than the compound, an optimum in FRET signal emerges, after which the compound starts to compete the tracer out of the active site (e.g., the Ki-20227 signal in Fig. 4A ).²⁷ Both curves in Figure 4A reach the same equilibrium with the tracer, indicating that the inhibitors differ only in binding kinetics. This shape of the competitive binding curve contains information on the compound off-rate.^20,27

Although the mathematics for fitting the competitive binding kinetic curve has been developed,²⁰ it was impractical in our hands. We succeeded in correctly fitting empirical data only in about 10% of cases. To make the method more amenable for automatic fitting, we rearranged the existing equations (see above and Suppl. File S2). The adjusted model is numerically easier to minimize and gives an acceptable fit in about 90% of cases. Combined with the homogeneous 384-well format for the competitive binding assay, the method presents a high-throughput and general way of measuring inhibitor dissociation rates, which is a great advantage over existing low-throughput methods.

The data for the reference inhibitors ( Fig. 4B ) show that the method can identify inhibitors with both slow and fast off-rates. Most reference compounds, including the known DFG-out binders such as sunitinib and GW-2580, show a fast off-rate on the order of minutes. Hits with slow off-rates were identified from the screening set ( Fig. 4 ). Moreover, Ki-20227 shows a very slow off-rate, on the order of hours ( Fig. 4B ), revealing it as a very interesting CSF1R inhibitor.

Discussion

In this work, we show how a screening campaign with a single assay can be followed up by a panel of orthogonal profiling assays. In the case of CSF1R, a library screen of 600 000 compounds led to the discovery of >800 validated hits, which we ultimately wanted to reduce to about five series for hit optimization chemistry ( Fig. 1 ). Usually this is done by applying an EC₅₀ cutoff, computational criteria based on diversity, the Lipinski rules, and internal experience with certain series. This works well for a few dozens of hits but leads to arbitariness with larger hit numbers. For instance, compounds with relatively poor EC₅₀ but good cellular penetration will be missed in the selection, whereas these are more suitable starting points for drug discovery than potent, noncellularly active compounds. This can be avoided by continued testing in a follow-on assay portfolio that has sufficient throughput to be run before manual hit triaging.

Because of the importance of cellular activity, a high-throughput cellular assay was developed as the first follow-on assay. Many chemical scaffolds have inherent problems in crossing the cell membrane, and these should be discarded as quickly as possible. We found the frozen cell–based M-NFS-60 proliferation assay, despite its drawbacks, most practical for this use, as the other formats are less robust and more labor intensive.

As a second important property, we focused on selectivity. In kinase drug discovery, many scaffolds will bind more than one kinase, and it is important to deprioritize the most promiscuous scaffolds early on.²⁸ To avoid testing hits on a full panel of biochemical kinase assays but still get an idea of general selectivity, we measured binding to inactive CSF1R. However, this readout appears to poorly predict selectivity—for instance, the reasonably selective ( Table 2 ) imatinib has less affinity for inactive CSF1R than the promiscuous staurosporine ( Fig. 3C ).

By itself, the finding that imatinib preferentially binds the active form of CSF1R ( Fig. 3 ) is surprising because imatinib is usually classified as binding to an inactive, DFG-out kinase conformation.^14,23 However, as shown earlier,²³ the data are not necessarily contradictive, as modeling in 3D structures predicts that imatinib clashes with the JM domain in inactive CSF1R.²⁵ In the activated CSF1R, this JM domain is phosphorylated and more loosely associated with the kinase domain. Using our results, imatinib can therefore be classified as preferentially binding a DFG-out and JM domain-out conformation. Apparently, it is easier to impose such a conformation on activated CSF1R.

Aside from imatinib, we wanted to more quantitatively assess the predictive value of the binding assay for selectivity. We downloaded the full published kinase profiles of all reference inhibitors.⁸ From those, single values can be calculated that are a measure for selectivity ( Table 2 ), such as the selectivity score, Gini coefficient, partition index, and the selectivity entropy ( Table 2 ; reviewed in Uitdehaag and Zaman²²). A comparison showed that the selectivity entropy, which calculates the Boltzmann distribution of inhibitor molecules in a theoretical mixture of kinase targets, is the most robust value, as it avoids arbitrary cutoffs and use of a reference kinase.²² The selectivity entropy value confirms that GW-2580 is by far the most selective CSF1R inhibitor ( Table 2 ). Next, as validation of the binding assay, we plotted the reference inhibitor selectivity entropies against the inhibitor K_d ratio for binding to active and inactive CSF1R ( Fig. 5A ). Unfortunately, the poor correlation confirms that binding to inactive CSF1R poorly predicts reference inhibitor selectivity ( Fig. 5A ).

Fig. 5.

Correlation of assay parameters with compound selectivity, as expressed by the selectivity entropy (values in Table 2 ). (A) Correlation with the ratio of binding to active and inactive colony-stimulating factor 1 receptor (CSF1R). The higher the ratio, the more affinity for inactive CSF1R. (B) Correlation with the ratio of inhibition of extracellular signal-regulated kinase (ERK) and CSF1R phosphorylation (as measured in THP1 cells). The higher the ratio, the more potent a compound inhibits CSF1R signaling relative to ERK signaling. (C) Correlation with target residence half-life.

To further look for predictive assay readouts, we next plotted the selectivity entropy to two other readouts: the ratio between pERK and pCSF1R inhibition in THP1 cells ( Fig. 5B ) and the target binding half life ( Fig. 5C ). Both pathway inhibition and residence time have been argued to predict selectivity.^16,29 Of the two, the pERK/pCSF1R window best correlates with the selectivity entropy ( Fig. 5B ). The correlation is positive, meaning the more promiscuous a compound is (the higher the selectivity entropy value), the stronger CSF1R is inhibited relative to ERK. This suggests that inhibition of off-target kinases stimulates CSF1R auto-phosphorylation and dampens the ERK response. Intuitively, the inverse would be expected. Nevertheless, the analysis suggests that the pERK/pCSF1R pathway readout is the better marker for hit selectivity in our profiling campaign.

Finally, because of the pharmacological advantages of compounds with slow dissociation rates, we wanted to prioritize such hits. To do this, the competitive binding kinetics assay in a 384-well format offers great advantages in throughput and generality over other methods, although it remains an indirect setup.¹⁶ Within the competitive format, other experiments are possible, such as diluting a preequilibrated enzyme–inhibitor complex in tracer solution (Life Technologies), but in our hands, the simultaneous adding of inhibitor and tracer works better and has the additional activity of being able to build on existing kinetic models.²⁰

Our experiments show that Ki-20227 has a surprisingly slow off-target dissociation rate ( Fig. 4B ). Interestingly, Ki-20227 contains a thiazole–urea motif ( Table 1 ), which is also the basis for the slow kinetics of the p38α inhibitor BIRB-796, as it induces conformational changes in p38α.¹⁸ Related thiazole–urea compounds found in screening of p38α showed slow binding kinetics and induction of conformational changes.³⁰ Therefore, this chemical substructure might be responsible for the slow dissociation rate of Ki-20227 from CSF1R.

In short, following up the results of a high-throughput screen by other high- and medium-throughput assays allows for a highly rigorous hit selection process. As illustrated by the new assay formats for selectivity and residence time, such an approach can reveal unexpected properties for compounds and chemical series and thereby help drug discovery to be more successful.

Footnotes

Acknowledgements

The authors thank Guido Zaman and Paul Vink for comments on the manuscript, as well as Steve Riddle and Roland Leathers of Life Technologies for assistance with the Lantha Screen assay format.

Supplementary material for this article is available on the Journal of Biomolecular Screening Web site at .

References

Hamilton

J. A.

Colony-Stimulating Factors in Inflammation and Autoimmunity. Nat. Rev. Immunol. 2008, 8, 533–544.

Chitu

Stanley

E. R.

Colony-Stimulating Factor-1 in Immunity and Inflammation. Curr. Opin. Immunol. 2006, 18, 39–48.

Patel

Player

M. R.

Colony-Stimulating Factor-1 Receptor Inhibitors for the Treatment of Cancer and Inflammatory Disease. Curr. Topics Med. Chem. 2009, 9, 599–610.

Manthey

C. L.

Player

M. R.

Advances in the Discovery of Anti-Inflammatory FMS Inhibitors. Ann. Med. Chem. 2009, 44, 211–225.

Conway

J. G.

McDonald

Parham

Keith

Rusnak

D. W.

Shaw

Jansen

Lin

Payne

Crosby

R. M.

. Inhibition of Colony-Stimulating-Factor-1 Signaling In Vivo with the Orally Bioavailable cFMS Kinase Inhibitor GW-2580. Proc. Natl. Acad. Sci. U. S. A. 2005, 102, 16078–16083.

Ohno

Kubo

Murooka

Kobayashi

Nishitoba

Shibuya

Yoneda

Isoe

A cFms Tyrosine Kinase Inhibitor, Ki-20227, Suppresses Osteoclast Differentiation and Osteolytic Bone Destruction in a Bone Metastasis Model. Mol. Cancer Ther. 2006, 5, 2634–2643.

Jack

G. D.

Zhang

Friedman

A. D.

M-CSF Elevates c-Fos and Phospho-C/EBPα (S21) via ERK Whereas G-CSF Stimulates SHP2 Phosphorylation in Marrow Progenitors to Contribute to Myeloid Lineage Specification. Blood 2009, 114, 2172–2180.

Karaman

M. W.

Herrgard

Treiber

D. K.

Gallant

Atteridge

C. E.

Campbell

B. T.

Chan

K. W.

Ciceri

Davis

M. I.

Edeen

P. T.

. A Quantitative Analysis of Kinase Inhibitor Selectivity. Nat. Biotech. 2008, 26, 127–132.

Brownlow

Vaid

Dibb

N. J.

Tandutinib Inhibits Fms Receptor Signalling, and Macrophage and Osteoclast Formation In Vitro. Leukemia 2008, 22, 1452–1453.

10.

Huang

Hutta

D. A.

Rinker

J. M.

Huaping

Parsons

W. H.

Schubert

DesJarlais

R. L.

Crysler

C. S.

Chaikin

M. A.

Donatelli

R. R.

. Pyrido[2,3-d]pyrimidin-5-ones: A Novel Class of Anti-Inflammatory Macrophage Colony-Stimulating Factor-1 Receptor Inhibitors. J. Med. Chem. 2009, 52, 1081–1099.

11.

Shun

T. Y.

Lazo

J. S.

Sharlow

E. R.

Johnston

P. A.

Identifying Actives from HTS Data Sets: Practical Approaches for the Selection of an Appropriate HTS Data-Processing Method and Quality Control Review. J. Biomol. Screen. 2011, 16, 1–14.

12.

Goldstein

D. M.

Gray

N. S.

Zarrinkar

P. P.

High Throughput Kinase Profiling as a Platform for Drug Discovery. Nat. Rev. Drug Discov. 2008, 7, 391–397.

13.

Zuccotto

Ardini

Casale

Angiolini

Through the ‘Gatekeeper Door’: Exploiting the Active Kinase Conformation. J. Med. Chem. 2010, 53, 2681–2694.

14.

Johnson

Protein Kinase Inhibitors: Contributions from Structure to Clinical Compounds. Q. Rev. Biophys. 2009, 42, 1–40.

15.

Simard

J. L.

Klüter

Grütter

Getlik

Rabiller

Rode

H. B.

Rauh

A New Screening Assay for Allosteric Inhibitors of cSrc. Nat. Chem. Biol. 2009, 5, 394–396.

16.

Copeland

R. A.

Pompliano

D. L.

Meek

T. D.

Drug Target Residence Time and Its Implications for Lead Optimization. Nat. Rev. Drug Discov. 2006, 5, 730–739.

17.

Tonge

P. J.

Drug-Target Residence Time: Critical Information for Lead Optimization. Curr. Opin. Chem. Biol. 2010, 14, 467–474.

18.

Pargellis

Tong

Churchill

Cirillo

P. F.

Gilmore

Graham

A. G.

Grob

P. M.

Hickey

E. R.

Moss

Pav

. Inhibition of p38 MAP Kinase by Utilizing a Novel Allosteric Binding Site. Nat. Struct. Biol. 2002, 9, 268–272.

19.

Annis

D. A.

Athanasopoulos

Curran

P. J.

Felsch

J. S.

Kalghatgi

Lee

W. H.

An Affinity Selection–Mass Spectrometry Method for the Identification of Small Molecule Ligands from Self-Encoded Combinatorial Libraries: Discovery of a Novel Antagonist of E. coli Dihydrofolate Reductase. Int. J. Mass Spectrometry. 2004, 238, 77–83.

20.

Motulsky

H. J.

Mahan

L. C.

The Kinetics of Competitive Radioligand Binding Predicted by the Law of Mass Action. Mol. Pharm. 1984, 25, 1–9.

21.

Zaman

G. J.

de Roos

J. A.

Blomenröhr

van Koppen

C. J.

Oosterom

Cryopreserved Cells Facilitate Cell-Based Drug Discovery. Drug Discov. Today 2007, 12, 521–526.

22.

Uitdehaag

J. C. M.

Zaman

G. J.

A Theoretical Entropy Score as a Single Value to Express Inhibitor Selectivity. BMC Bioinform. 2011, 12, 94.

23.

Wodicka

L. M.

Ciceri

Davis

M. I.

Hunt

J. P.

Floyd

Salerno

Hua

X. H.

Ford

J. M.

Armstrong

R. C.

Zarrinkar

P. P.;

. Activation State-Dependent Binding of Small Molecule Kinase Inhibitors: Structural Insights from Biochemistry. Chem. Biol. 2010, 17, 1241–1249.

24.

Meyers

M. J.

Pelc

Kamtekar

Day

Poda

G. I.

Hall

M. K.

Michener

M. L.

Reitz

B. A.

Mathis

K. J.

Pierce

B. S.

. Structure-Based Drug Design Enables Conversion of a DFG-in Binding CSF-1R Kinase Inhibitor to a DFG-out Binding Mode. Bioorg. Med. Chem. Lett. 2010, 20, 1543–1547.

25.

Walter

Lucet

I. S.

Patel

Broughton

S. E.

Bamert

Williams

N. K.

Fantino

Wilks

A. F.

Rossjohn

The 2.7 Å Crystal Structure of the Autoinhibited Human c-Fms Kinase Domain. J. Mol. Biol. 2007, 367, 839–847.

26.

Gajiwala

K. S.

J. C.

Christensen

Deshmukh

G. D.

Diehl

DiNitto

J. P.

English

J. M.

Greig

M. J.

Y. A.

Jacques

S. L.

. KIT Kinase Mutants Show Unique Mechanisms of Drug Resistance to Imatinib and Sunitinib in Gastrointestinal Stromal Tumor Patients. Proc. Natl. Acad. Sci. U. S. A. 2009, 106, 1542–1547.

27.

Arányi

Kinetics of the Hormone-Receptor Interaction: Competition Experiments with Slowly Equilibrating Ligands. Biochim. Biophys. Acta 1980, 628, 220–227.

28.

Brandt

Jenmalm Jensen

Nilsson

Small Assay Panels Can Provide a Measure of Selectivity. Bioorg. Med. Chem. Lett. 2009, 19, 5861–5863.

29.

Hancock

M. K.

Lebakken

C. S.

Wang

Multi-Pathway Cellular Analysis of Compound Selectivity. Mol. Biosyst. 2010, 6, 1834–1843.

30.

Simard

J. R.

Grütter

Pawar

Aust

Wolf

Rabiller

Wulfert

Robubi

Klüter

Ottmann

. High-Throughput Screening to Identify Inhibitors Which Stabilize Inactive Kinase Conformations in p38α. J. Am. Chem. Soc. 2009, 131, 18478–18488.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB

0.03 MB

Multidimensional Profiling of CSF1R Screening Hits and Inhibitors

Abstract

Keywords

Introduction

Materials and Methods

Compound synthesis

CSF1R kinase activity assay

Selectivity profiling

M-NFS-60 and NS0 proliferation assays

CSF1R and ERK target engagement assays

Sources of active and inactive CSF1R

Competitive binding assay

Affinity selection–mass spectrometry (ALIS) assay

Off-rate profiling through competitive binding kinetics

Results

Primary CSF1R activity assay

Computational clustering

Cellular activity in M-NFS-60 cells

CSF1R target engagement in THP1 cells

ERK pathway engagement in THP1 cells

Profiles in a small selectivity panel

Conformational probing

Off-rate profiling

Discussion

Footnotes

Acknowledgements

References

Supplementary Material