In silico Screening and Evaluation of Plasmodium falciparum Protein Kinase 5 (PK5) Inhibitors

Abstract

An in silico screen of 350,000 commercially available compounds was conducted with an unbiased approach to identify potential malaria inhibitors that bind to the Plasmodium falciparum protein kinase 5 (PfPK5) ATP-binding site. PfPK5 is a cyclin-dependent kinase-like protein with high sequence similarity to human cyclin-dependent kinase 2 (HsCDK2), but its precise role in cell cycle regulation remains unclear. After two-dimensional fingerprinting of the top scoring compounds, 182 candidates were prioritized for biochemical testing based on their structural diversity. Evaluation of these compounds demonstrated that 135 bound to PfPK5 to a similar degree or better than known PfPK5 inhibitors, confirming that the library was enriched with PfPK5-binding compounds. A previously reported triazolodiamine HsCDK2 inhibitor and the screening hit 4-methylumbelliferone were each selected for an analog study. The results of this study highlight the difficult balance between optimization of PfPK5 affinity and binding selectivity for PfPK5 over its closest human homolog HsCDK2. Our approach enabled the discovery of several new PfPK5-binding compounds from a modest screening campaign and revealed the first scaffold to have improved PfPK5/HsCDK2 selectivity. These steps are critical for the development of PfPK5-targeting probes for functional studies and antimalarials with reduced risks of host toxicity.

Graphical Abstract

A virtual screen of 350,000 compounds identified several molecules with potential to inhibit Plasmodium falciparum protein kinase 5 (PfPK5). Biochemical testing of predicted PfPK5-binding compounds identified a 4-methylumbelliferone as able to bind the kinase and inhibit P. falciparum blood stage parasites. Compound 2b binds PfPK5 with a similar affinity as it binds to the structurally related human cyclin-dependent kinase 2, offering improved selectivity when compared to previously reported PfPK5 inhibitors.

Malaria is a life-threatening disease that affects millions of people annually and resulted in over 400,000 deaths in 2016.^[1] Protozoan parasites from the Plasmodium genus infect human hepatocytes and erythrocytes to cause malaria, where P. falciparum is the species responsible for the largest number of deaths. Unfortunately, artemisinin-based combination therapies (ACT), the recommended treatment for P. falciparum infections, are losing efficacy due to the development of parasite drug resistance. This resistance to the first-line malaria treatment threatens disease control efforts and necessitates the development of new drugs with novel targets.^[2–3]

Protein kinases are attractive drug targets for several conditions including cancer, inflammatory diseases, infectious diseases, and others.^[4–7] The P. falciparum genome encodes for 86 protein kinases, many of which are thought to be essential based on genetic^[8–10] and chemical inhibition studies.^[11–12] Thus, the Plasmodium kinome remains a rich yet largely untapped area for novel drug target development.^[13–15] In contrast to human kinases, the biological functions of many Plasmodium kinases are unresolved due to the lack of molecular and chemical tools to probe mechanisms. Genetic methods to produce knockouts are often lethal and small molecule probes lack the selectivity necessary for cell-based studies.^[16]

The Plasmodium parasite within the human host first infects hepatocytes during the asymptomatic liver stage, which is requisite for the development of parasites that can infect erythrocytes to cause disease during the blood stage. Multi-stage inhibition of Plasmodium is desirable for drug development to reduce the risk of drug resistance. Additionally, dual stage inhibitors can be used to both prevent and treat malaria. Due to the broad importance of many kinase families for cellular functions in other systems, kinases have the potential to be multi-stage drug targets for malaria.^{[6–7, 17]}

Protein structures can often be leveraged to facilitate inhibitor discovery, but only six P. falciparum protein kinases currently have elucidated crystal structures.^[18–19] Of these, P. falciparum protein kinase 5 (PfPK5), which has been crystallized at 1.9 Å resolution, is a putative cyclin-dependent kinase (CDK)-like protein^{[18, 20–21]} with the highest similarity in sequence and structure^[18] to Homo sapiens CDK2 (HsCdK2, 63% sequence identity and 74% similarity). Like HsCDK2, PfPK5 may regulate cell division, however this role has not been established in the parasite.^[22] The high degree of homology to HsCDK2 (Figure S1) makes studying PfPK5 function with small molecules during infection difficult and all known PfPK5 inhibitors exhibit poor selectivity.

Previous studies have leveraged the wealth of compounds with known HsCDK2 affinity to discover PfPK5-binding compounds.^[18] This approach has been successful in identifying compounds with high nM affinity for PfPK5, but they often exhibit >100-fold selectivity for HsCDK2. For example, CDK1/2 Inhibitor III (1a) is a known HsCDK2 inhibitor with an IC₅₀ of 0.5 nM.^[24] Previously, 1a was found to bind to PfPK5 with a K_d of 320 nM,^[11] but this affinity is offset by poor species selectivity. There have been limited efforts to modulate species selectivity of PfPK5 inhibitors through a structure-activity relationship (SAR) approach. Therefore, to explore whether the selectivity of 1a can be favorably adjusted towards PfPK5, we conducted a modest SAR study. Compound 1a along with a strategically selected set of its structural analogs (1b–1e, Table 1) were synthesized^[24] for testing (see the Supporting Information for details). These compounds were selected from a previous SAR campaign of 16 compounds based on their varying inhibition of HsCDK1, a closely related kinase to HsCDK2, where potency decreased for 1b (217-fold), 1c (13-fold), 1d (1,100-fold), and 1e (75-fold) relative to 1a. Of these compounds, we observed that the parent compound 1a has the highest binding affinity to both PfPK5 (K_d(app) 0.37 μM) and HsCDK2 (K_d(app) 0.00039 μM), but it is 950-fold selective for HsCDK2 (Figure 1A). This observation agrees with the reported activity of 1a against HsCDK1 as determined with a kinase inhibition assay.^[24] Among the analogs tested, changes in potency/affinity generally agreed between HsCDK1, HsCDK2, and PfPK5 where 1c was the next most active (Figure 1B) and 1d was the least active (Figure S2). Compounds 1d and 1e did not bind to PfPK5 up to 30 μM, therefore binding constants could not be determined. Due to a reduction in HsCDK2 affinity for 1c we observed increased selectivity for PfPK5 — 180-fold selective for the human protein versus 950-fold for 1a. However, while selectivity was improved, the analogs from this approach did not produce a compound selective for the parasite kinase or achieve desirable potency. These results demonstrate the challenges of designing selective PfPK5 inhibitors using previously characterized CDK1/CDK2 inhibitors as starting points.

Table 1.

Structures of the triazolo-diamine (1) and 4-methylumbelliferone (2) analogues examined for binding to PfPK5 and HsCDK2.^[a]

			HsCDK2	PfPK5
Compound	R1	R2	Kd(app) (μM)	Kd(app) (μM)
1a		–	0.00039	0.37
1b		–	0.013	23
1c		–	0.013	2.4
1d		–	2.1	>300
1e		–	0.13	>30
2a		Me	11	5
2b		Me	5	3.8
2c		Me	28	14
2d		Me	>30	>30
2e		Me	>30	>30
2f		Et	>30	>30

^[a]Binding affinities determined in duplicate using the KINOMEscan® platform.

Figure 1.

Binding of selected compounds to PfPK5 (red triangles) and HsCDK2 (black circles). Structures are shown above plots. A) 1a (CDK 1/2 Inhibitor III) and its most active analog B) 1c are triazolo-diamines that exhibit selectivity to HsCDK2. Data are shown as the average ± SD and were fit to a standard dose response equation.

A target-based screening campaign is a previously unexplored approach to identify PfPK5 inhibitors without bias for HsCDK2 activity. To reduce both the cost and time of evaluating a large compound library, we first employed an in silico screening method to evaluate 350,000 accessible compounds for the ability to bind the PfPK5 ATP-binding site. Such a strategy proved valuable for identifying selective inhibitors of PfMRK, a homolog of HsCDK7,^[23] and we predicted that it may also facilitate the discovery of PfPK5 inhibitors with improved species selectivity. Our initial docking studies (standard precision and extra precision) identified 400 structurally diverse potential inhibitors of PfPK5. The top 182 compounds were chosen for further analysis based on structural diversity (Figure S3) and were obtained through the ICCB-Longwood Screening Facility (Harvard Medical School) for in vitro studies. Compounds were screened for binding to PfPK5 (Figure 2, Table S2) at 12 μM using an active-site-directed competition assay (KINOMEscan^®). In this assay, protein bound to beads is released when test compounds bind to the ATP-binding site. After normalization to a DMSO control, the relative protein remaining on the beads is assessed. As controls, two known PfPK5 inhibitors, purvalanol A and indirubin-3´-monoxime, were included in the library as they were also predicted to bind PfPK5 based on the in silico analysis. The identification of known PfPK5-binding compounds as candidates in the virtual screen supported the promise of the strategy to identify novel PfPK5 ligands. Additionally, the selection of these compounds for testing provided internal controls for the protein binding assay. After evaluation of the positive and negative controls, screening results were grouped into three categories; no/low-binding (100–95%, 44 compounds), moderate-binding (95–50%, 135 compounds, 74% hit rate), and high-binding (<50%, 3 compounds, 1.6% hit rate) (Figure 2).

Figure 2.

Competition-based biochemical assay for PfPK5 binding. The ability of predicted compounds (circles) to bind to the PfPK5 ATP-binding site at 12 μM was determined. Purvalanol A (blue circle) and indirubin-3´-monoxime (green circle) are known inhibitors. Compounds were grouped as high-binding, moderate-binding and no/low-binding. Controls are included in the moderate binding group (50–95%, indicated by grey dashed lines). Three compounds (2a, 3a, and 4a, red circles) are included in high binding group (>50%).

Purvalanol A and indirubin-3´-monoxime are known to inhibit PfPK5 with IC₅₀s of 8 μM^[18] and 6 μM^[25], respectively. To the best of our knowledge these are among the most potent PfPK5 inhibitors reported; therefore, the identification of these controls within the moderate-binding group suggests many uncharacterized PfPK5-binding compounds exist within this group. The high hit rate of 76% indicates that our in silico approach provided a library enriched with PfPK5-binding compounds. While a wealth of potential PfPK5 inhibitors stemmed from the primary screen, we sought to prioritize our studies on a select number of compounds.

The 182 PfPK5-binding candidates were counter-screened for HsCDK2 binding at 12 μM using a similar active-site-directed competition assay (Table S2). HsCDK2 was selected for this screen due to its high sequence similarity to PfPK5. After normalization to the DMSO negative control, the percent bound to HsCDK2 versus PfPK5 was plotted to reveal compounds that preferentially bind to either protein (Figure 3). This analysis highlights the challenge of discovering selective kinase inhibitors, as many compounds bound both proteins to a similar degree. However, four compounds displayed selectivity for HsCDK2 and three, 2a, 3a, and 4a, displayed selectivity for PfPK5. Compounds 2a, 3a, and 4a were selected for subsequent validation studies due to our interest in the selective targeting of Plasmodium parasites.

Figure 3.

Relative binding to PfPK5 and HsCDK2 for top 182 scoring compounds (circles) from the in silico screen. The majority of the compounds display similar binding potency for PfPK5 and HsCDK2. The most potent PfPK5- binding compounds are shown in red (2a, 3a, and 4a).

Screening actives were purchased and tested at additional concentrations (0–30 μM) to generate binding curves and determine apparent K_d values. Through this analysis, 3a and 4a (Figure S4) did not bind PfPK5 up to 30 μM. To probe this conflicting result, 3a and 4a from the original compound library plate were retested and it was observed that the binding observed in the primary screen was reproduced. This result indicates that the discrepancy in activity between the primary screen and follow up study is due to varying sample sources rather than a false discovery rate. It is not uncommon for compounds within screening libraries to degrade over time in DMSO, making validation studies critical. Unfortunately, these experiments exhausted the original supply of 3a and 4a from the compound library plate, therefore no further evaluation of these compounds was pursued. Interestingly, 2a (4-methylumbelliferone scaffold) showed 2.2-fold selective binding to PfPK5 (K_d(app) 5 μM) over HsCDK2 (K_d(app) 11 μM), as shown in Figure 4A. Compound 2a binding to PfPK5 and HsCDK2 was evaluated with molecular docking (Figure S5), which suggested hydrogen bonds to the hydroxyl group on the 4-methylumbelliferone scaffold may stabilize binding. To further assess 2a, five structural analogs (2b–2f, Table 1) were purchased for biochemical testing. The selection of analogs was based on 1) predicted interactions of 2a to the PfPK5 ATP-binding site (Figure S4), 2) structural similarity of >80% to 2a, and 3) commercial availability. Among the analogs tested 2d–2f did not bind PfPK5 or HsCDK2 up to 30 μM, whereas 2b and 2c displayed dose-dependent binding to HsCDK2 and PfPK5 (Figure 4B). Compound 2b bound to PfPK5 (K_d(app) 3.8 μM) and HsCDK2 (K_d(app) 5.0 μM) with similar potency while 2c showed 2-fold selectivity for PfPK5 (K_d(app) 14 μM) over HsCDK2 (K_d(app) 28 μM). This difference in binding affinity is not considered parasite selective (>10-fold), similar to the parent compound 2a, but it is significantly improved over the tested triazolo-diamine analogs (Figure 5). To the best of our knowledge 2a and 2c are the most PfPK5 selective compounds reported to date.

Figure 4.

Binding of 4-methylumbelliferone compounds A) 2a and B) 2b to PfPK5 (red triangles) and HsCDK2 (black circles). Structures are shown above plots. Data are shown as the average ± SD and were fit to a standard dose response equation.

Figure 5.

Fold selectivity of studied compounds for PfPK5 and HsCDK2 binding. Triazolo-diamine analogs (1a–1c) all display significant selectivity toward HsCDK2 binding, while the 4-methylumbelliferone analogs (2a–2c) discovered via in silico screening displayed improved PfPK5 selectivity. Fold selectivity was calculated as a ratio of the apparent dissociation constants, K_d(app). Compounds with binding >30 μM are not shown. Dashed red lines indicate ratios of −1 and 1. Data for SNS-032 and flavopiridol are from previous reports.^{[11, 27]}

For biological assays, the imine on 2a is undesirable due to its known reactivity. In buffer, imines can hydrolyze to produce a ketone or aldehyde, which we observed by LC-MS after incubating 2a in buffer (pH 7.4) for 10 min and 60 min. Additionally, certain imines have been included as Pan Assay Interference Compounds (PAINS)^[26] and are removed from screening libraries. Compounds 2b and 2c were selected to evaluate the importance of the imine and examine if the hydrolysis product contributes to bioactivity. Compound 2b is the aldehyde product of 2a, while 2c retains the imine, but includes an iodo- group that should hinder protein binding due to steric bulkiness. While we anticipated no binding of 2c, only a 3.7-fold decrease in the PfPK5 K_d(app) was observed. Compounds 2d–2f, which maintain the sterics of 2a without the imine, did not bind to either kinase. These observations suggest the hydrolysis products of 2a and 2c bind to the ATP-binding site. In agreement with this prediction, molecular docking indicates that 2b binds to the PfPK5 and HsCDK2 ATP-binding sites with a similar docking score as 2a (Figure S5,S6). Unfortunately, a comparative computational study does not reveal a molecular basis for the varying selectivity observed between 1a and 2a, which highlights the need for future structural studies (Figure S7).

To assess species selectivity in a cellular context, the studied compounds were tested for activity against human hepatoma cell lines (HuH7 and HepG2) and P. falciparum Dd2 asexual blood stage parasites. None of the studied compounds significantly inhibited HepG2 cells at 20 μM, but 1a and 1c were found to reduce HuH7 viability by >25% at 20 μM (Figure S8). Further evaluation of this inhibition with a dose-response study revealed that 1a and 1c inhibited HuH7 viability with EC₅₀ values of 67 and 107 μM, respectively. The varying inhibition activity between the two cells lines is likely due to differences in replication rates, where HuH7 cells double at a faster rate than HepG2 cells under our culture conditions.

Compounds were also tested for anti-Plasmodium activity. From an initial screen at 20 μM, 2b, 2e, 2f, and 1a–1e were all found to inhibit blood stage parasites by >25% (Figure S9A). After a dose-response analysis, the parent compound 2a was not active, but 2b, 2e, and 2f had EC₅₀ values ranging between 10–30 μM (Figure S9B). While these compounds did not exhibit liver cell toxicity, they selectively inhibited Plasmodium parasite growth. Evaluation of the triazolo-diamine derivatives revealed that 1a was the most potent inhibitor (EC₅₀ 0.14 μM), but both 1c and 1e also significantly reduced parasite load with EC₅₀ values of 0.38 μM and 6.4 μM, respectively (Figure S9C). Therefore, these series are selective for inhibiting Plasmodium over hepatoma cells. This observed species selectivity in cells may be due to the varying replication rates of Plasmodium and hepatocytes in vitro, which would make the parasites more susceptible to cell cycle inhibitors over the course of the 48-hour assay. Alternatively, the compounds may be binding to other targets in cells to influence parasite viability. The possibility of off-target binding would also explain the poor correlation observed between PfPK5 binding and parasite inhibition. However, biophysical properties such as cell permeability, solubility, and pharmacodynamic/pharmacokinetic profiles would need to be optimized to further explore this trend.

PfPK5 has been implicated as a Plasmodium cell cycle regulator, but identifying PfPK5 inhibitors to probe its function has been challenging due to its homology to HsCDK1 and HsCDK2. Here we report an in silico strategy coupled with a high-throughput screen to identify PfPK5-binding compounds. After prioritization of compounds by potency and selectivity for the parasite protein over HsCDK2, we identified the most selective scaffold discovered to date. When compared to modification of known HsCDK2 inhibitors, target-based PfPK5 screening appears to be a more fruitful path towards developing selective compounds with reduced host toxicity. The continued optimization of scaffolds to obtain species selectivity will be critical to generate chemical probes for better elucidating the function of cyclin-dependent like kinases in Plasmodium parasites and possible drug leads.

Experimental Section

Compound 1a (CDK 1/2 Inhibitor III) was tested for binding affinity to both PfPK5 and HsCDK2 as an initial study on selectivity and potency. Subsequently, analogs of 1a (1b–1e) were synthesized according to a previously reported procedure.^[24] All compounds were confirmed to have >95% purity based on high-resolution mass spectrometry (Agilent LC-MS).

Molecular docking studies were conducted to predict new compounds that bind to PfPK5. To begin, 350,000 readily accessible compounds were docked onto the ATP-binding site of PfPK5 (PDB ID: 1V00) from a virtual library using Glide (Schrödinger),^[28–29] a high throughput virtual screening program. The library members were carefully selected to enable efficient biochemical testing of any screening active. Compounds were first docked using standard precision (SP) followed by extra precision (XP) mode to eliminate false positives. Finally, 2D fingerprint clustering was used to further eliminate false positives by estimating differences in sub-structural fragments of each compound. The top 182 predicted binders were selected based on docking scores and structural diversity. Screening compounds were then cherry picked from diverse libraries and obtained through the ICCB-Longwood Screening Facility (Harvard Medical School) for biochemical testing. A target-based assay was used to test the ability of the 182 compounds to bind to PfPK5 and HsCDK2 at 12 μM. The assay conducted by DiscoverX employs an active site-directed competition binding assay using KINOMEscan^® Technology.^[27] First, singleconcentration tests were completed with compounds from the ICCB-Longwood Screening Facility and then compounds that exhibited binding to PfPK5 at 12 μM were purchased from ChemBridge Corporation for K_d(app) determination with both PfPK5 and HsCDK2 (DiscoverX). All compounds were >95% pure. Results were plotted using Excel for the initial biochemical test and GraphPad Prism for K_d(app) determination where data were fit to a standard equation. Analogs of 4-methylumbelliferone (2b–2e, all >95% pure) were subsequently purchased from ChemDiv (2b) and ChemBridge Corporation (2c–2e) for further studies.

Purchased and synthesized compounds were tested for cytotoxicity in two different human liver cell lines (HuH7 and HepG2). Compounds (20 μM) in triplicate were incubated with HepG2 (15,000 cells/well) or HuH7 (7,000 cells/well) cells in 384-well plates for 48 hrs in a standard tissue culture incubator at 37 °C. The final concentration of DMSO was 1% and the assay volume was 30 μL. After 48 hrs, CellTiter-Glo (Promega) was added to each well to quantify ATP using an EnVision (Perkin Elmer). Dose-response curves were generated for any compound that exhibited significant cytotoxicity compared to the DMSO control. Compounds were also evaluated for P. falciparum Dd2 inhibition using a previously published protocol.^[30] Compounds (20 μM) in duplicate were incubated with P. falciparum Dd2 in 384-well plates for 72 hrs with O-positive human blood in an atmosphere of 93% N₂, 4% CO₂, 3% O₂ at 37 °C. The final concentration of DMSO was 0.2% and the assay volume was 50 μL. After 72 hrs, nuclei were stained with SYBR Green I (Invitrogen) and fluorescence was subsequently measured using an EnVision. The positive control was artesunate and the negative control was DMSO. Data were normalized to the DMSO control and any compound that inhibited P. falciparum viability ≥50% was selected for EC₅₀ determination (GraphPad Prism).