Screening a protein kinase inhibitor library against Plasmodium falciparum

Cultures of 3D7, a P. falciparum chloriquine sensitive reference strain, were maintained in a 5% suspension of A+ human red blood cells (obtained from East of Scotland Blood Transfusion Service, Ninewells Hospital, Dundee, UK) cultured in RPMI 1640 medium (pH 7.3) supplemented with 0.5% Albumax II (Gibco Life Technologies, San Diego, CA, USA), 12 mM sodium bicarbonate, 0.2 mM hypoxanthine, and 20 mg/l gentamicin at 37 °C, in a humidified atmosphere of 1% O₂, 3% CO₂ with a balance of nitrogen. Growth inhibition was quantified using a fluorescence assay utilizing the binding of SYBR Green I to double stranded DNA, which emits a fluorescent signal at 528 nm after excitation at 485 nm. The SYBR Green assay system was adapted to maximize robustness and to align with available automation systems. The Prestwick Library then used as a pilot screen to validate the performance of the optimized assay. Mefloquine (potency range 30–60 nM) [16] was used as a drug control to monitor the quality of the assay (Z’ = 0.6–0.8, Signal to background ≥ 3, where Z’ is a measure of the discrimination between the positive and negative controls on a screening plate). Single point screens were carried out, and the potency of compounds of interest was determined in duplicate on two independent occasions. A 96-well [³H]-Hypoxanthine incorporation assay [17] was also developed as a secondary assay in order to validate key compounds from each hit series in an orthogonal platform. Compound bioactivity from both assays was expressed as EC₅₀, the effective concentration of compound causing 50% parasite death.

A counter-screen against normal diploid human fibroblasts (MRC-5 pd30, ECACC 84101801) was carried out to exclude non-selective, and toxic compounds as previously published [18]. Cells were plated and incubated overnight to allow them to adhere as monolayers. A working stock of each test compound was transferred to an intermediate 384-well plate and pre-diluted with minimum essential media (MEM). The pre-diluted stocks were then transferred onto the cell monolayers, and the plates were incubated for 68 h. Resazurin, to a final concentration of 50 μM was added to each well, after which plates were incubated for a further 3 h and measured for fluorescence (λ _ex = 528 nm, λ _em = 590 nm). A standard reference compound, doxorubicin, was included on all counterscreen plates to monitor the quality of the assay (potency range 200–400 nM).

All data were processed using IDBS ActivityBase [19]. Raw data were converted into per cent inhibition through linear regression by setting the high inhibition control as 100% and the no inhibition control as 0%. Quality control criteria for passing plates were as follows: Z’ > 0.5, S:B > 3, %CV_{(no inhibition control)} < 15. The formula used to calculate Z’ is \( 1 - \frac{3 \times (StDev\_high + StDev\_low)}{ABS(Mean\_high - Mean\_low)} \).

All EC₅₀ Curve fitting was undertaken using XLFit version 4.2 using Model 205 with the following 4 parametric equation: \( y = A + \frac{B - A}{{1 + ({\raise0.7ex\hbox{$C$} \!\mathord{\left/ {\vphantom {C x}}\right.\kern-0pt} \!\lower0.7ex\hbox{$x$}})^{D} }}, \) where A = % inhibition at bottom, B = % inhibition at top, C = EC₅₀, D = slope, x = inhibitor concentration and y = % inhibition. If the curve did not reach 100% of inhibition, B was fixed to 100 when at least 50% of inhibition was reached.

Protein kinases have been suggested as potential drug targets in malaria [3, 7]. However, rather than screen each individual kinase, a more rapid approach would be to screen the malaria parasite phenotypically. This has a number of advantages: it is not necessary to clone and overexpress each kinase and develop an individual assay; the protein kinase is in its relevant state of activation with the relevant substrates present; this only selects cell penetrant compounds and compounds that have a lethal effect; it would detect compounds that were acting on multiple proteins. The disadvantages of this approach include the unknown identity of the target(s), which could make optimization of hits problematic, particularly if there were pharmacokinetic or toxicological issues.

The Dundee library of protein kinase scaffolds was assembled as described in Brenk et al. [12]. The aim of this library was to find chemical start points for drug discovery programmes, so the molecules were deliberately kept low molecular weight, giving room for optimisation. In brief, a review of the literature and patents was carried out by industry experienced chemists to identify scaffolds that bind in the adenine pocket of all protein kinases, to the so-called hinge region. Allosteric inhibitors were not considered. The hinge region is part of the backbone that forms a hydrogen bonding network with the adenine moiety of ATP. Decoration of these scaffolds can achieve varying degrees of selectivity for particular kinases. These scaffolds were then searched for using an in silico database of 2.6 million discrete compounds compiled from the catalogues of 26 suppliers. The initial selections were further screened to eliminate non-lead like compounds, using over 100 filters based on chemically reactive groups and toxicophores. This process identified in excess of 150 scaffolds with at least one commercially available example. To limit the total number of compounds in the screening collection, a limit of the most diverse 50 examples of each scaffold was set using Tanimoto Indices to assess similarities. At the time of the screen this library contained 4731 individual compounds.

The 4731 compounds from the DDU curated kinase library were screened against P. falciparum at a single concentration of 3 μM using a fluorescence-based assay. Compounds with a percentage inhibition (PI) > 30 were retested in duplicate and 120 hits were progressed to potency determination against both P. falciparum and human MRC5 cells to assess toxicity. Potency values were established using a 10-point, half-log serial dilution of each compound from a top concentration of 15 µM. Two independent assays were run. Screening data statistics were comparable with those obtained in the pilot screen.

To further establish all discovered hits, structural identity and purity of all compounds progressed to potency was confirmed by LC/MS. Fresh material was sourced by resupply or resynthesis to confirm activity in the original assay and in an orthogonal (hypoxanthine incorporation) assay.

The compounds were cherry picked from library material and tested at 10 µM. This was carried out at the International Centre for Kinase Profiling, University of Dundee [20], where the principal assay method utilized is a radioactive filter-binding assay using ³³P ATP. At the time of screening, there were 76 protein kinases in the panel. Whilst this represents a relatively small proportion (~ 15%) of the mammalian kinome (518 predicted protein kinases in humans [21]), it should give an indication as to whether the compound is selective or promiscuous. The profiled compounds were checked for purity and molecular identity by LCMS.

LCMS analysis of the 120 compounds progressed into EC₅₀ determination was conducted on an Agilent technologies 1200 series LC/Agilent technologies 6130 quadrapole LC/MS, eluting with 20–90% MeCN in 0.1% aqueous formic acid. 96 of the 120 compounds had > 85% purity by diode array and confirmation of the expected mass to give an 80% pass rate.

In order to validate the optimized assay, the commercially available Prestwick Library (1120 compounds) was screened in duplicate at single concentration. A number of compounds with known anti-malarial properties were identified from this pilot screen, and results compare favourably with published data [22] (Fig. 1, Table 1 and Additional file 1: Table S1 for full list of the hits), thus indicating that the DDU optimized assay was fit for purpose. The final assay was demonstrated to be robust (Z’ = 0.84 ± 0.01), and reproducible (mefloquine) EC₅₀ = 48 ± 4 nM (n = 24).

The frequency distribution plot of percentage inhibition for the primary screen of the focused kinase library against P. falciparum is shown in Fig. 2a, illustrating a classical normal distribution. A cut-off of 30% inhibition was chosen to define an initial hit (3 standard deviations away from the averaged high controls). Table 2 summarizes the numerical results from the primary screen.

The 210 putative hits identified from the kinase set primary screen were cherry picked and retested in a duplicate single point screen (3 μM). Figure 3 shows the excellent correlation between the two replicate retest data points. For those compounds returning > 50% inhibition in the primary assay the confirmation rate to retest was 68%. 120 compounds with a percentage inhibition of > 80% were selected for progression to potency testing. Ten-point dose response curves were generated and assayed in both P. falciparum and mammalian MRC5 (human lung fibroblasts) proliferation assays. A range of potencies (Table 3) were identified against P. falciparum, with the majority of compounds being inactive in the MRC5 counter-screen with an EC₅₀ > 15 μM.