Revisiting the Plasmodium falciparum druggable genome using predicted structures and data mining

Abstract

Identification of novel drug targets is a key component of modern drug discovery. While antimalarial targets are often identified through the mechanism of action studies on phenotypically derived inhibitors, this method tends to be time- and resource-consuming. The discoverable target space is also constrained by existing compound libraries and phenotypic assay conditions. Leveraging recent advances in protein structure prediction, we systematically assessed the Plasmodium falciparum genome and identified 867 candidate protein targets with evidence of small-molecule binding and blood-stage essentiality. Of these, 540 proteins showed strong essentiality evidence and lack inhibitors that have progressed to clinical trials. Expert review and rubric-based scoring of this subset based on additional criteria such as selectivity, structural information, and assay developability yielded 27 high-priority antimalarial target candidates. This study also provides a genome-wide data resource for P. falciparum and implements a generalizable framework for systematically evaluating and prioritizing novel pathogenic disease targets.

Keywords: Target identification; Target validation.

Fig. 1. Identification of P. falciparum protein-coding genes that are potentially druggable (n = 1660) and genes that have evidence of blood-stage essentiality (n = 2992).

a Identification of potentially druggable genes. Diagram illustrating the four methods used to identify genes with evidence of small-molecule binding. Out of the 5318 protein-coding 3D7 falciparum genes, 226 were homologous to at least one of the 6202 validated drug targets in BindingDB based on BLAST, while 520 genes were found to be orthologous based on the phylogenomic databases OrthoMCL, OrthoDB, OMA, or HOGENOM. We found 1233 genes with confident AlphaFill hit transplant(s), and 927 falciparum genes were mapped to EC numbers with inhibitors in the BRENDA database. Mosquito graphic was generated with BioRender (https://BioRender.com/s17b944). b Binding evidence for validated targets vs. all genes. Distribution of binding evidence for 43 known targets (outer pie) compared to the distribution across all 5275 P. falciparum 3D7 genes (inner pie). c Workflow for the identification of 867 candidate targets and subsequent prioritization. Identification and filtering process to define P. falciparum candidate drug targets. The intersection of 1660 genes with evidence of small-molecule binding and 2992 genes with essentiality support yielded 867 candidates, after excluding hypervariable regions. Subsequent candidate prioritization resulted in 540 candidates subjected to initial scoring, 67 of which underwent a second round of scoring, culminating in 27 top-ranking targets that were discussed by a panel of experts. d Essentiality classifications for validated targets vs. all genes. Distribution of essentiality classifications for 43 known targets (outer pie) compared to the distribution for 5275 3D7 genes (inner pie). Essentiality classifications are described in “Methods”.

Fig. 2. Characteristics of 867 candidate targets compared to 4451 non-candidate genes.

a Characteristics of candidate targets vs. non-candidates. Comparison of characteristics between the 867 candidates (blue) and 4451 non-candidates (purple). Numbers of candidate targets versus non-candidates labeled essential only in the P. falciparum essentiality screen and not the P. berghei datasets (850 vs. 2421); labeled essential in both P. falciparum and berghei datasets (577 vs. 967); having at least one GO term (857 vs. 3624); having human ortholog(s) (650 vs. 1356); having > 2 associated literature references (570 vs. 1775); or having PDB structures (112 vs. 174) are shown. b Gene ontology (GO) term enrichment analysis for the 867 candidates. GO term enrichment analysis for the 867 candidate targets compared to all 5318 protein-coding genes in the P. falciparum 3D7 genome using GOATOOLS. Terms with ontology tree depth > 2 (n = 939) are displayed based on the number of candidates having the GO term (X axis) versus −log₁₀ Bonferroni corrected enrichment P value, with a maximum uncorrected P value of 0.05 (Y axis, Fisher’s exact test). Points corresponding to GO terms are colored by ontology type: red for biological process (BP), blue for cellular component (CC), and green for molecular function (MF). Highly enriched terms or groups of terms are labeled with shared descriptors. c Scientific literature references for candidates vs. non-candidates. Distributions of a number of unique scientific publications associated with the 867 candidate targets (blue) vs. 4451 non-candidate genes (purple). Median lines are shown for both groups, and the most highly referenced genes are labeled. d Gene expression in the asexual blood stage (ABS) compared to all protein-coding genes. Distribution of classifications for gene expression evidence in ABS (ring, trophozoite, or schizont) according to Le Roch et al.. Candidate targets with clear evidence of ABS expression (737, teal), unclear evidence (57, purple), no expression (37, orange) and no data (36, gray) are shown in the outer pie, in contrast to the distribution among all P. falciparum protein-coding genes in the inner pie.

Fig. 3. Rubric-based scoring of 540 candidate targets with strong evidence of essentiality.

a Gene scores across rubric categories. Plot showing first score distributions for 540 candidate targets with strong evidence of blood-stage essentiality across the ten categories of the scoring rubric (“Methods”, Supplementary Table 4). The average value per category is shown for each category. b Frequency distribution of first total scores. Histogram showing the frequency (X axis) of first total scores (Y axis) for the 540 scored candidates. Bin center was determined and plotted with Prism v.9.5. Examples of candidates falling in select total score bins are shown, including gene product description and notable characteristics when applicable. c Comparison of first and second scores for the top 67 scored candidates subject to secondary review. Identity line is marked by a dashed black line. Dark blue circles denote equal score in both rounds; light blue circles represent score differences of 1–19 points; and cyan circles represent a score difference of at least 20 points.

Fig. 4. Selection of five high-scoring targets and examples of two understudied but promising candidates.

a Scoring distribution across categories for top five candidate targets. Individual scores for the top five candidate targets across rubric categories, compared with the average scores across all 540 scored candidates. The average score is highlighted by light purple circles, and top five candidates are shown in blue (TopoI), green (BDP1), orange (GluPho), salmon (ATCase) and dark purple (GyrB). b, c AlphaFill models for advanced candidate targets. Predicted AlphaFill models for PfATCase (b) and PfGluPho (c) are shown. Red rectangles highlight the region where transplant hits were found, with a zoomed-in inset of hit transplant structure having the highest percentage of identity. d TopoI model. TopoI (PF3D7_0510500) model was constructed using UniProt ID Q8I3Z9 and ligand hits (Supplementary Table 2). For simplicity, five ligands (BDBM-50249684, −50033788, −50259215, −50249691, and −50092821) associated to the UniProt ID were randomly selected from BindingDB hits. Ligands were docked onto the model using openbabel 3.1.1 and smina 2020.12.10. The model was visualized using PyMol version 2.5.5. e, f AlphaFill models for understudied but promising candidate targets. Predicted AlphaFill models for PfPGM1 (e) and PfARF1 (f) candidate targets. Red rectangles highlight the region of some transplant hits, and a zoomed-in inset including hit transplant structure with the highest percentage of identity is shown.