Transcriptomics-Guided In Silico Drug Repurposing: Identifying New Candidates with Dual-Stage Antiplasmodial Activity

Malaria, caused by Plasmodium parasites and transmitted by Anopheles mosquitoes, remains a major global health problem with over 247 million cases and approximately 619,000 deaths in 2021, predominantly in sub-Saharan Africa. The parasite’s complex life cycle spans human liver and blood stages and sexual stages in the mosquito vector. Widespread and emerging resistance to existing antimalarials, including artemisinin-based combination therapies, creates an urgent need for new drugs with novel mechanisms of action. Drug repurposing, enhanced by chemogenomics and computational approaches such as data mining and machine learning, offers a faster, cost-effective pathway to identify new antimalarial candidates. This study set out to identify multistage therapeutic targets in P. falciparum using transcriptomics and to repurpose existing bioactive compounds with predicted activity across parasite stages, prioritizing candidates for experimental validation.

The paper situates its approach within the context of drug resistance in malaria and the increasing importance of drug repurposing. Prior work highlights the complexity and high cost of de novo drug discovery, the utility of chemogenomics to link chemical and biological data for target identification and lead discovery, and the role of computational methods (including QSAR and machine learning) in predicting activity and guiding selection. Transcriptomic resources for various Plasmodium life stages, including blood and gametocyte stages of P. falciparum and liver-stage data from related species (P. cynomolgi), have enabled multistage target identification. Tools such as DrugBank, TTD, BLASTp, and Chemical Checker expand target and compound discovery, while in vitro assays and nanoluciferase-based transmission-blocking assays provide experimental validation frameworks.

- Multistage transcriptomics mining: Compiled highly expressed genes (≥80% expression) across P. falciparum life stages using public datasets. Blood-stage transcriptomes: López-Barragán (rings, early trophozoites, schizonts; 1921 genes), Bunnik (rings, trophozoites, schizonts; 1840 genes), and Zanghi (rings; 1119 genes). Gametocyte-stage: López-Barragán (stage II/V; 1461 genes) and Lasonder (male and female; 1699 genes). Sexual stages: López-Barragán (ookinetes) and Zanghi (sporozoites, oocysts; total 1748 genes). For liver stage, due to lack of P. falciparum data, used P. cynomolgi Cubi dataset (hypnozoite and liver; 2012 genes) and mapped 1991 orthologs in P. falciparum. Identified 674 overlapping genes across stages; 409 deemed essential based on PlasmoDB. - Target and compound retrieval: Queried TTD and DrugBank with predicted amino acid sequences of multistage targets to find homologous therapeutic targets and associated bioactives/drugs. Applied BLASTp filtering (e-value < 1e-30; identity > 30%) to ensure homology stringency. Kept 70 homologous targets; removed 17 without linked compounds. Retained 53 targets linked to 75 bioactive compounds. - Similar compound expansion: Used Chemical Checker to find compounds with similar bioactivity profiles to the 75 bioactives, considering levels beyond chemical similarity (targets, interaction networks, cellular, clinical). Retrieved 1887 hits across levels; after deduplication obtained 1557 unique similar compounds; combined with 75 known bioactives to yield 1632 candidates. - Machine learning virtual screening: Applied previously developed QSAR-ML models (random forest) to predict activity across five parasite contexts: blood stages (P. falciparum 3D7, W2), ookinete, liver, and gametocyte stages. Descriptor sets: ECFP (3D7, W2), MACCS (ookinete), hybrid Mordred+ECFP (gametocyte), FCFP (liver). Selected compounds predicted active by all five models. Filtered out aggregators/PAINS and considered commercial availability. - Experimental validation: Purchased two prioritized compounds (NVP-HSP990 and silvestrol aglycone). Conducted in vitro growth inhibition assays against P. falciparum 3D7 (chloroquine-sensitive) and Dd2 (multidrug-resistant) strains (0.5% parasitemia, 2% hematocrit; 72 h; SYBR Green readout), with artesunate and chloroquine controls. Determined EC50 values by nonlinear regression. Assessed transmission-blocking potential via P. berghei ookinete conversion inhibition (PbOokluc line; 10 μM single point and dose–response EC50). Evaluated cytotoxicity using MTT assays in HepG2 and COS-7 cell lines to derive CC50 and selectivity indices.

- Multistage gene set: 674 overlapping highly expressed genes across parasite stages; 409 classified as essential. - Target and compound identification: After stringent BLASTp filters (e-value < 1e-30; identity > 30%), 70 homologous targets were retained; 17 without compounds were removed, leaving 53 targets linked to 75 bioactive compounds. - Chemical Checker expansion: 1887 similar-compound hits across bioactivity profile levels; deduplicated to 1557 unique similars. Total candidate set = 75 known bioactives + 1557 similars = 1632 compounds. - QSAR predictions for 75 initial drugs: 23 predicted active vs 3D7; 11 vs W2; 9 vs ookinetes; 9 vs liver; 8 vs gametocytes. Eight of the 75 predicted active across all five models/stages; includes artenimol (dihydroartemisinin). - QSAR predictions for 1557 similars: 529 predicted active vs 3D7; 94 vs W2; 78 vs ookinetes; 78 vs liver; 76 vs gametocytes. Seventy-six predicted active across all five models/stages (Table S3). After aggregator filtering, 20 remained; two compounds were selected for purchase: NVP-HSP990 and silvestrol aglycone. - In vitro antiplasmodial activity: Both compounds showed nanomolar EC50s against P. falciparum 3D7 and Dd2. Silvestrol aglycone EC50: 3D7 = 0.0094 μM ± 0.0007; Dd2 = 0.006 μM ± 0.003. NVP-HSP990 EC50: 3D7 = 0.54 μM ± 0.05; Dd2 = 0.66 μM ± 0.1. Chloroquine reference: 3D7 = 0.0035 μM ± 0.0004; Dd2 = 0.056 μM ± 0.01. - Cytotoxicity/selectivity: Silvestrol aglycone CC50: COS-7 = 0.32 μM ± 0.2; HepG2 = 0.24 μM ± 0.23; selectivity index reported ~34 in text (calculated vs 3D7). NVP-HSP990 CC50: COS-7 = 0.09 μM ± 0.1; HepG2 = 2.69 μM ± 3.7; lower selectivity. - Transmission-blocking: Silvestrol aglycone inhibited 99.65% of P. berghei ookinete conversion at 10 μM; ookinete conversion inhibition EC50 = 3.21 μM. NVP-HSP990 showed weaker transmission-blocking activity (approx. 57.35% reported).

The integrated, transcriptomics-guided chemogenomics and machine learning workflow effectively prioritized multistage essential targets and associated compounds for repurposing against malaria. By focusing on genes highly expressed across life stages and filtering for homologous therapeutic targets with stringent sequence similarity, the study reduced the search space to bioactives with plausible mechanisms. Expansion via Chemical Checker and ML screening identified compounds with predicted activity in both asexual blood and transmission stages. Experimental validation confirmed potent nanomolar activity for silvestrol aglycone and NVP-HSP990 against chloroquine-sensitive and multidrug-resistant strains, with silvestrol aglycone showing favorable selectivity and strong transmission-blocking potential. These findings directly address the need for dual-acting antimalarials capable of treating disease and interrupting transmission, complementing current therapies and potentially mitigating resistance spread.

This study demonstrates a transcriptomics-guided in silico repurposing strategy that integrates chemogenomics tools, similarity expansion, and ML-based QSAR to identify antimalarial candidates with multistage activity. From 674 multistage-expressed genes (409 essential), the pipeline yielded 53 prioritized targets linked to 75 bioactives and expanded to 1557 similar compounds. ML screening and subsequent filtering nominated candidates, of which silvestrol aglycone and NVP-HSP990 exhibited in vitro efficacy, with silvestrol aglycone showing nanomolar potency, low cytotoxicity, and strong transmission-blocking activity. Silvestrol aglycone emerges as a promising dual-acting antimalarial candidate. Future work should include target deconvolution/mechanistic studies, in vivo efficacy and pharmacokinetics, broader cytotoxicity profiling, and optimization to improve therapeutic windows.

- Liver-stage transcriptomics: Due to unavailable P. falciparum liver-stage transcriptomes, liver-stage analysis relied on P. cynomolgi data and orthology mapping, which may introduce species-specific biases. - Limited experimental set: Budget and availability constraints reduced experimental validation to two compounds from a larger predicted active set, limiting generalizability. - Transmission-blocking model: Transmission assays used P. berghei (murine model) rather than P. falciparum, which may not fully recapitulate human parasite biology. - In vitro focus: Findings are based on in vitro assays; in vivo efficacy, pharmacokinetics, and safety remain to be established. - Homology-based target mapping: Identification of homologous targets via sequence similarity may miss functional differences; mechanisms of action were not experimentally confirmed.