Structural characterization of the ACDC domain from ApiAP2 proteins, a potential molecular target against apicomplexan parasites

The apicomplexan AP2 (ApiAP2) proteins are the best characterized family of DNA-binding proteins in Plasmodium spp. malaria parasites. Apart from the AP2 DNA-binding domain, there is little sequence similarity between ApiAP2 proteins. However, a conserved AP2-coincident domain mostly at the C-terminus (ACDC domain) is observed in a subset of the ApiAP2 proteins. The structure and function of this domain remain unknown. We report two crystal structures of ACDC domains derived from distinct Plasmodium ApiAP2 proteins, revealing a conserved, unique, noncanonical, four-helix bundle architecture. We used these structures to perform in silico docking calculations against a library of known antimalarial compounds and identified potential small-molecule ligands that bind in a highly conserved hydrophobic pocket that is present in all apicomplexan ACDC domains. These ligands provide a new molecular basis for the future design of ACDC inhibitors.

1. Introduction

Malaria is an infectious disease caused by intracellular apicomplexan parasites from the genus Plasmodium. The two most predominant Plasmodium species are P. falciparum and P. vivax, which annually kill over half a million people worldwide (World Health Organization). The parasite life cycle involves developmental stages in both the human and the Anopheles mosquito hosts, and mosquitos also serve as the vector for transmission. Malaria parasite development relies on the expression of specific sets of parasite genes at precise times throughout the life cycle (Tebben et al., 2021). Transcriptional regulation is largely coordinated by the Apicomplexa Apetala 2 (ApiAP2) protein family, the largest and most well characterized family of DNA-binding proteins in Apicomplexa, including Plasmodium (Painter et al., 2011; Balaji et al., 2005; Jeninga et al., 2019). ApiAP2 proteins regulate transcription by binding to specific DNA motifs using their 60-amino-acid AP2 DNA-binding domains. These domains are only found in the ApiAP2 proteins and in the AP2/ERF family of plant transcription factors (Feng et al., 2020; Balaji et al., 2005). Due to their essentiality and the lack of mammalian orthologues, ApiAP2 proteins are considered to be putative drug targets (Russell et al., 2022).

ApiAP2 proteins vary greatly in size (from 200 to 4109 amino acids in P. falciparum) and may contain 1–3 AP2 DNA-binding domains (Jeninga et al., 2019). Most members of the ApiAP2 family contain no other recognizable functional domains and have relatively low homology between species, apart from the AP2 DNA-binding domains. The ApiAP2 proteins are predicted to be mostly disordered (AlphaFold Protein Structure Database; Varadi et al., 2022) and, so far, only the isolated AP2 domain of a P. falciparum ApiAP2 protein has been structurally characterized (PDB entry 3igm; Lindner et al., 2010). In this study, we focus on a ∼90-residue domain of unknown function called the AP2-coincident domain mostly at the C-terminus (ACDC), which has been identified in a subset of the family (Oehring et al., 2012). According to the InterPro classification (IPR028078), the ACDC domain is observed in ∼1000 protein sequences. The ACDC is exclusive to the Apicomplexa, and it is present only once in each protein, always in combination with an identified AP2 domain.

We solved the crystal structures of two ACDC domains from P. falciparum ApiAP2 proteins. The first is PfAP2-05, which is the smallest ACDC-containing ApiAP2 protein in P. falciparum, with only 715 residues. AP2-O5 is also found in other Plasmodium spp. and has been reported to be an essential protein for asexual blood-stage development in P. falciparum but not in P. yoelii (Zhang et al., 2017, 2018). In P. yoelii it has been shown to regulate the expression of genes encoding proteins implicated in the mobility of the ookinete, the mosquito midgut invasive stage of the parasite (Zhang et al., 2017). PfAP2-O5 contains a single N-terminal AP2 domain and a C-terminal ACDC domain (Jeninga et al., 2019). The second ACDC domain that we characterized was from PfAP2-I, which is the only P. falciparum ApiAP2 protein to contain an ACDC domain at the N-terminus (Oehring et al., 2012). PfAP2-I is 1597 residues in length, contains three AP2 domains and has been shown to regulate the expression of genes encoding proteins that play a critical role in the invasion of host erythrocytes in P. falciparum (Santos et al., 2017). AP2-O5 is essential for the survival of all tested Plasmodium species (Jeninga et al., 2019).

Both AP2-O5 and AP2-I contain large disordered regions as predicted by AlphaFold (https://alphafold.ebi.ac.uk/entry/Q8IKY0 and https://alphafold.ebi.ac.uk/entry/Q8IJW6), impairing structural analysis of the full-length proteins. We thus solved the crystal structure of their isolated ACDC domains. Our structural analysis reveals that the ACDC domain adopts a novel conserved fold characterized by an idiosyncratic left-handed orthogonal four-helix bundle. PfAP2-O5 and PfAP2-I are both interesting drug targets since they are essential to the asexual blood stage of P. falciparum. We performed in silico docking simulations, which suggested that compounds from the Tres Cantos Anti-Malarial Set (TCAMS; Gamo et al., 2010) may bind to the ACDC domain. Taken together, these results demonstrate the druggable potential of the ACDC domains not only from Plasmodium but also from other apicomplexan parasites, such as Toxoplasma gondii, which is responsible for toxoplasmosis (Tenter et al., 2000), or Cryptosporidium, the main global cause of parasite-derived diarrhea (Alsaady, 2024).

2. Experimental procedures

2.1. Protein production

2.1.1. Cloning

Synthetic genes for full-length AP2-I from P. falciparum (UniProt entry Q8IJW6; ORF name PF3D7_1007700; 1597 amino acids) and P. vivax (UniProt entry A0A564ZT73; ORF name PVP01_0807400; 1397 amino acids) were codon-optimized for expression in Escherichia coli (ThermoFisher). The gene coding for the AP2-O5 protein of P. falciparum (UniProt entry Q8IKY0; ORF name PF3D7_1449500; 715 amino acids) was amplified from genomic DNA prepared from P. falciparum parasites using a DNeasy Kit (Qiagen) according to the manufacturer's instructions. Gene fragments coding for residues 1–152 of PfAP2-I and residues 30–230 of PvAP2-I were amplified by PCR using primers 3 and 4 or primers 5 and 6 (Eurofins), respectively (Supplementary Table S1), and were then cloned into a pET-28 plasmid (Novagen) with T4 DNA ligase (Thermo Scientific) using the restriction-enzyme pair NcoI/XhoI (Thermo Scientific) to generate C-terminal Strep-tag II affinity tags. The gene fragment coding for residues 625–715 of PfAP2-O5 was amplified by PCR using primers 1 and 2 (Supplementary Table S1) and was then cloned into a pSL1045 plasmid (Addgene) using XhoI and NcoI (New England Biolabs) followed by T4 DNA ligase (New England Biolabs) to generate a C-terminal His-tag preceded by a TEV protease cleavage site. The three plasmids were transformed into ultracompetent E. coli DH5α (ThermoFisher). Correct assembly was verified by PCR using Taq 2X Master Mix (New England BioLabs) and Sanger sequencing (Genewiz).

2.1.2. Protein expression and purification

Expression of the ACDC domains of PfAP2-O5, PfAP2-I and PvAP2-I was carried out in E. coli (DE3)-Gold strain in 800 ml 2YT medium at 37°C for 3 h. Induction was performed by adding 0.5 mM isopropyl β-D-1-thiogalactopyranoside (Sigma). The cells were harvested by centrifugation (4500g for 20 min at 8°C), resuspended in 40 ml buffer A (20 mM Tris–HCl pH 7.5, 200 mM NaCl) and stored at −20°C. After thawing, a protease-inhibitor cocktail (Thermo Scientific) was added to the resuspended bacterial pellet and cell lysis was carried out by sonication (Branson probe-tip sonicator). After centrifugation (20 000g for 30 min at 8°C), the 40 ml soluble fractions were purified by affinity chromatography. Strep-tagged AP2-I ACDC domains were loaded onto a 5 ml StrepTrap HP column (Cytiva) equilibrated at 4°C in buffer A for FPLC chromatography. After extensive washing with buffer A, the Strep-tagged proteins were eluted with buffer A supplemented with 2.5 mM desthiobiotin. The His-tagged AP2-O5 ACDC domain was loaded onto 3 ml Ni–NTA resin on a benchtop column, washed extensively with cold buffer A and then with cold buffer A supplemented with 20 mM imidazole. Finally, the AP2-O5 ACDC domain was eluted with cold buffer A containing 400 mM imidazole. The purified fractions were then loaded onto a HiLoad 16/60 Superdex 75 prep-grade size-exclusion chromatography column equilibrated in buffer A to eliminate aggregates. Protein concentrations were determined by measurement of the absorbance at 280 nm using a NanoDrop spectrophotometer (Thermo Fisher). Purified proteins were concentrated using Vivaspin 5000 molecular-weight cutoff concentrators, aliquoted and stored at −80°C in buffer A.

2.1.3. Protein characterization

Purified samples were characterized by SEC-MALS using an OMNISEC REVEAL gel-permeation chromatography/size-exclusion chromatography (GPC/SEC) multidetector system with a RESOLVE module (Malvern Panalytical). The protein samples were injected at a concentration of 1 mg ml⁻¹ (50 µM) onto a Superdex 75 10/300 GL Increase column (GE Healthcare) equilibrated in 20 mM Tris–HCl pH 8, 200 mM NaCl. The OMNISEC software from the manufacturer was used for data acquisition and analysis.

2.2. Crystal structure determination and analysis

Crystallization assays were performed at 18°C in 96-well plates using commercial crystallization screening kits (Molecular Dimensions) in 200 nl droplets (100 nl protein + 100 nl reservoir solution) prepared using a Cartesian Microsys pipetting robot (Proteigene) at the I2BC crystallization facility. Prior to diffraction assays, crystals were cryoprotected by transferring them into droplets comprising their crystallization solution supplemented with 30% PEG 400 and were flash-cooled in liquid nitrogen. Diffraction data were collected on the PROXIMA-2 beamline at the SOLEIL synchrotron (Saint-Aubin, France) at a cryogenic temperature of 100 K. Data processing was performed using the XDS package (Kabsch, 2010). Phasing was performed by molecular replacement using Phaser (McCoy et al., 2007). Initial models were predicted using ColabFold v1.5.3: AlphaFold2 using MMseqs2 (Mirdita et al., 2022; Jumper et al., 2021). For the ACDC domain of PfAP2-O5, AlphaFold2 predicted five similar models (r.m.s.d. below 0.5 Å) with high confidence scores [predicted local distance difference test (pLDDT) > 90] and the rank-1 model was efficiently used for phasing by molecular replacement. For the ACDC domains of PfAP2-I and PvAP2-I, disordered regions of the AlphaFold2 models were deleted for phasing.

Protein structure refinement was performed using Phenix (Liebschner et al., 2019) and was further improved by iterative cycles of manual rebuilding using Coot (Emsley et al., 2010). The final structures were deposited in the Protein Data Bank (PDB; Berman et al., 2000). We used PyMOL (The PyMOL Molecular Graphics System, version 1.2r3pre; Schrödinger) for graphical analysis of the structures and preparation of figures. We used ESPript (Gouet et al., 1999) for graphical enhancement of sequence alignments. Protein assemblies were analyzed using the PDBePISA server at the European Bio­informatics Institute (EBI; https://www.ebi.ac.uk/pdbe/prot_int/pistart.html; Krissinel & Henrick, 2004). Protein structure comparison was performed using FoldSeek with the 3Di/AA mode (https://search.foldseek.com; van Kempen et al., 2024) against the ∼200 000 experimental structures in the Protein Data Bank (Berman et al., 2000) and the ∼200 million predicted structures in the AlphaFold2 database (Varadi et al., 2022).

2.3. In silico docking screen

The three crystal structures solved in this study were prepared for docking simulations with MGLTool (Morris et al., 2009) using the prepare_receptor.py script. A screening box surrounding a conserved region was used as the target for the docking calculation. We screened against the 13 533 small molecules from the Tres Cantos Anti-Malarial Set (TCAMS), which have all been shown to inhibit P. falciparum growth by at least 80% at 2 µM concentration (Gamo et al., 2010). The KNIME platform (Mazanetz et al., 2012) was employed to process the SMILES notation of each compound. The molecules were desalted using the Vernalis node and protonated prior to transformation into their 3D forms. The latter were minimized using the MMFF94 force field (Halgren, 1996) over 2000 iterations, and Gasteiger partial charges (Gasteiger & Marsili, 1980) were assigned via RDKit (https://www.rdkit.org/) nodes. The refined molecular library was then stored in a single SDF file. Torsional angles for each molecule were calculated using the Meeko preparation script (https://github.com/forlilab/Meeko), resulting in the assembly of individual ligands into distinct PDBQT files. During this process, macrocyclic components were treated as conformationally rigid. The docking protocol was executed using SMINA (Koes et al., 2013) with the Vinardo scoring function (Quiroga & Villarreal, 2016) set to explore 20 binding modes and an exhaustiveness parameter of 8 with eight cores per molecule. Screening results were parsed and analyzed using in-house Python notebooks.

2.4. Molecular-dynamics simulation protocol

Prioritized compounds were prepared for molecular-dynamics (MD) simulations. The parameterization of these compounds was conducted utilizing ACPYPE (Sousa Da Silva & Vranken, 2012), which serves as a bridge to Antechamber (Wang et al., 2004). The selection of initial conformations was guided by visual assessments aimed at optimizing interactions with the targeted pocket. Molecular-dynamics simulations were performed with GROMACS 2022 (Abraham et al., 2015) and the Amber 99SB-ILDN force field (Lindorff-Larsen et al., 2010) on GPUs. The complexes were solvated with a TIP3P water model (Jorgensen et al., 1983) within a cubic periodic box, with a minimum distance of 1.5 nm from the protein–ligand complex to the edge of the box. Charge neutrality was achieved by the addition of appropriate Na⁺ or Cl⁻ ions. Initial energy minimization was carried out over 1000 steps using a conjugate-gradient algorithm, integrating a steepest descent every 50 steps. This was followed by a heating phase over 500 ps, where the temperature was increased from 0 to 300 K under NVT conditions (constant temperature, constant volume). The Berendsen thermostat (Berendsen et al., 1984) was applied with a time constant of 0.1 ps, while positional restraints were maintained on protein and heavy atoms of the ligand, exerting a force constant of 1000 kJ mol⁻¹ nm⁻¹. Subsequently, the systems underwent pressure equilibration for 500 ps under NPT conditions (constant temperature and constant pressure) employing the Berendsen barostat, set with a time constant of 1 ps and a compressibility factor of 4.5 × 10⁻⁵ bar⁻¹, maintaining the aforementioned positional restraints. Restraints were removed for production and each system was simulated over 100 ns. The integration time step was fixed at 2 fs for both equilibration and production phases. The LINCS algorithm (Hess, 2008) was employed to constrain bonds between heavy atoms and hydrogens. The Verlet cutoff scheme (Páll & Hess, 2013) was adopted to manage long-range interactions, with a threshold distance of 1 nm. We used the Protein–Ligand Interaction Profiler (PLIP; Adasme et al., 2021) for the analysis of noncovalent interactions between the ACDC domains and the selected compounds.

3. Results

3.1. Crystal structure of the C-terminal ACDC domain of AP2-O5

The ACDC domain of PfAP2-O5 (residues 627–715, InterPro annotation) was efficiently expressed and purified from E. coli. The best crystals were obtained in 30% PEG 400, 0.1 M MES pH 6.5, 0.1 M sodium acetate and diffracted to 1.75 Å resolution in space group P2₁2₁2, with two molecules per asymmetric unit (Table 1). We solved the phase problem by molecular replacement using an AlphaFold2 model of the domain. We found that the PfAP2-O5 ACDC domain forms a left-handed orthogonal four-helix bundle with a break in the middle of helix α3 (Fig. 1a). The refinement procedure yielded a structure characterized by an R_work of 20.80% and an R_free of 23.97% (Table 1). The two polypeptide chains of the asymmetric unit displayed the same fold (r.m.s.d. of 0.61 Å over 87 aligned C^α atoms). Chain B is fully visible (residues 625–715). In chain A, seven amino acids from the TEV site introduced between the protein sequence and the C-terminal His-tag form two additional turns in the C-terminal helix α4, whereas residues 676–677 from loop α2–α3 are not visible in the electron-density map. Analysis of the protein interfaces in the crystal packing revealed that the two molecules of the asymmetric unit form a symmetrical dimer considered to be a stable assembly characterized by a complex-formation significance score of 1, a buried area of 2590 Ų and a solvation free-energy gain ΔG^int of −21.6 kcal mol⁻¹, suggesting that they may form a stable dimer in solution. The interface is mainly stabilized by salt bridges and hydrogen bonds between the α3 helices of the two chains on one face of the dimer and the α1 helices on the other face of the dimer (Fig. 1b).

Table 1
Data-collection and refinement statistics

Values in parentheses are for the highest resolution shell.

	PfAP2-I (PDB entry 8rxo)	PvAP2-I (PDB entry 8rwu)	PfAP2-O5 (PDB entry 8rxa)
Crystallization conditions	1.5 M ammonium sulfate, 0.1 M HEPES pH 7.5, 2% PEG 400	30% PEG 400, 0.1 M MES pH 6.5, 0.1 M sodium acetate	30% PEG 400, 0.1 M MES pH 6.5, 0.1 M sodium acetate
Data collection
Temperature (K)	100	100	100
Space group	P41212	P41212	P21212
a, b, c (Å)	78.08, 78.08, 127.84	78.49, 78.49, 123.88	58.93, 108.93, 34.01
α, β, γ (°)	90, 90, 90	90, 90, 90	90, 90, 90
Resolution range (Å)	50.0–3.00 (3.18–3.00)	50.0–3.15 (3.34–3.15)	50.0–1.75 (1.86–1.75)
No. of unique reflections	8421 (1303)	7133 (1106)	22718 (3510)
Completeness (%)	99.9 (99.6)	99.7 (98.7)	99.5 (96.8)
CC1/2	0.999 (0.710)	1.0 (0.495)	0.999 (0.941)
Mean I/σ(I)	12.81 (0.86)	21.53 (1.18)	22.24 (2.23)
Rmeas† (%)	19.8 (259.2)	11.8 (308.4)	5.9 (97.4)
Refinement statistics
Resolution range (Å)	49.46–3.00 (3.23–3.00)	48.62–3.15 (3.61–3.15)	40.0–1.75 (1.79–1.75)
No. of molecules per asymmetric unit	2	2	2
Rwork‡ (%)	23.38 (38.88)	22.62 (33.97)	20.80 (32.08)
Rfree§ (%)	23.69 (49.67)	27.37 (38.43)	23.97 (35.40)
Ramachandran statistics
Favored (%)	97.89	96.67	98.45
Outliers (%)	0.00	0.00	0.00
R.m.s.d.
Bond lengths (Å)	0.005	0.004	0.006
Bond angles (°)	0.738	0.727	0.787
Chirality	0.043	0.042	0.045
Planarity	0.005	0.005	0.007
Dihedral	16.820	5.249	20.041
Average B, all atoms (Å^2)	108.0	115.0	41.0

†R_meas = is the bias-corrected R_merge calculated by XDS (Kabsch, 2010), where N(hkl) is the multiplicity of a given reflection, I_i(hkl) is the ith integrated intensity of a given reflection and 〈I(hkl)〉 is the mean intensity of multiple corresponding symmetry-related reflections.
‡R_work = , where |F_obs| and |F_calc| are the observed and calculated structure-factor amplitudes, respectively.
§R_free is the same as R_work but calculated with a 20% subset of all reflections that was never used in refinement.

Figure 1 Crystal structure of the C-terminal ACDC domain of PfAP2-O5. (a) The polypeptide chain is shown as a cartoon colored in a spectrum from blue (N-terminus) to red (C-terminus). The position of the AP2 and ACDC domains of PfAP2-O5 is schematized below the structure with their residue ranges. (b) Two orthogonal views of the dimer observed in the asymmetric unit. Chain A is represented as a cartoon and chain B as a ribbon. In the lower panel, residues stabilizing the interface are shown as sticks colored by atom type (O in red and N in blue) and labeled.

Size-exclusion chromatography with multi-angle static light scattering (SEC-MALS) analysis of the isolated domain showed that the PfAP2-O5 ACDC domain is monomeric in solution with a molecular weight of about 13 kDa (Supplementary Fig. S1). Therefore, given that the first 625 residues of PfAP2-O5 are missing in the crystallized fragment, it is unclear whether or not the ACDC domain dimerizes in the context of the full-length protein.

3.2. Crystal structure of the N-terminal ACDC domain of AP2-I

To determine whether the 3D structure of the ACDC domain was conserved in another protein environment, we solved the crystal structure of the N-terminal fragment of PfAP2-I (residues 1–152) containing the ACDC domain of the protein (residues 62–140, InterPro annotation). The best crystals were obtained in 1.5 M ammonium sulfate, 0.1 M sodium HEPES pH 7.5, 2% PEG 400. They diffracted to 3.0 Å resolution in space group P4₁2₁2, with two molecules per asymmetric unit (Table 1). AlphaFold2 proposed confident models for the core ACDC domain of PfAP2-I (residues 60–150), but the N-terminal extension was predicted to be fully disordered. The proposed models were similar to those of the PfAP2-O5 ACDC domain, but loop α3–α4 (residues 120–130), which was predicted with low accuracy according to the pLDDT confidence score, needed to be deleted for efficient phasing by molecular replacement. The final structure of the PfAP2-I N-terminal fragment (Fig. 2a) was refined at 3.0 Å resolution with an R_work of 23.38% and an R_free of 23.69% (Table 1). Electron density could only be observed for the core ACDC domain, confirming that the N-terminal extension is disordered, at least in this truncated fragment of the protein. In turn, electron density was clearly visible to reconstruct loop α3–α4 that was omitted in the initial model (Supplementary Fig. S2). Surprisingly, it did not correspond to the conformation observed in the crystal structure of PfAP2-O5. Instead, it revealed an exchange of helix α4 between the two subunits of the asymmetric unit, resulting in two swapped orthogonal four-helix bundles similar to those observed in the PfAP2-O5 structure (Fig. 2a).

Figure 2 Crystal structure of the N-terminal ACDC domain of PfAP2-I. (a) The dimer observed in the asymmetric unit is shown with one subunit represented as a cartoon and the second as a ribbon, both colored in a spectrum. Swapping of helix α4 (red) between the two subunits of is highlighted by red arrows. The ACDC domain of PfAP2-O5, shown as a cartoon colored gray, superimposes on this swapped dimer with a r.m.s.d. of 1.065 Å over 56 aligned C^α atoms. The position of the ACDC and AP2 domains of PfAP2-I is schematized below the structure with their residue ranges. (b) Details of the PfAP2-I swapped dimer interface. Residues involved in stabilizing salt bridges and hydrogen bonds between the facing helices α4 and α2 of the two chains are shown as sticks colored by atom type and labeled.

This swapped dimer forms a stable assembly characterized by a ΔG^int of −24.5 kcal mol⁻¹ and a buried surface area of 2770 Ų. The interaction mode between the two subunits is completely distinct from that observed in the PfAP2-O5 structure. The interface formed by the helix swapping is stabilized by salt bridges and hydrogen bonds between the facing helices α4 and α2 of the two chains (Fig. 2b). Interestingly, PDBePISA analysis of the crystal packing suggested four other stable assemblies with strong interfaces forming dimers (data not shown) but, again, none of them correspond to that observed in the crystal of the PfAP2-O5 ACDC domain. This suggests that the protein assemblies observed in the two ACDC structures are likely to be artifacts due to crystal-packing constrains.

To test whether the observed helix swapping could be due to the truncation of the protein just after helix α4, we also determined the structure of an elongated fragment of AP2-I. Several constructs were tested for heterologous expression in E. coli. Because the ACDC domain of PfAP2-I is followed by a 40-residue polyasparagine repeat region, we used the homologous region from P. vivax, which does not contain this intrinsically disordered region (92% sequence identity between the two ACDC domains). In the end, the extended fragment covering residues 30–230 from the homologous PvAP2-I protein was efficiently purified and crystallized. The best crystals were obtained in 30% PEG 400, 0.1 M MES pH 6.5, 0.1 M sodium acetate. Despite the distinct crystallization conditions and protein fragment length, the crystal of the PvAP2-I ACDC domain belonged to the same space group P4₁2₁2 and had the same unit-cell parameters as that of the ACDC domain of PfAP2-I. The final structure of PvAP2-I was refined to 3.15 Å resolution with an R_work of 22.62% and an R_free of 27.37% (Table 1). The two subunits contained in the asymmetric unit displayed the same helix swapping as observed for the ACDC domain of PfAP2-I and the two dimers superimposed perfectly with a root-mean-square deviation (r.m.s.d.) of 0.981 Å over 176 superimposed C^α atoms. Again, no electron density was observed for the N- and C-terminal extensions (residues 30–60 and 150–230, respectively). It is therefore still not clear whether the helix swapping observed in the dimeric crystal structure of the isolated N-terminal ACDC domain of AP2-I is present in the context of the full-length protein when helix α4 is followed by more than 1000 amino acids.

3.3. Characterization of the idiosyncratic fold of ACDC

Because it is unclear whether the α4 helix swapping observed in AP2-I and/or dimerization are biologically relevant in the context of the full-length proteins, we used a single PfAP2-O5 ACDC subunit without helix swapping to search for structural homologues of the ACDC domain.

Using the FoldSeek server (van Kempen et al., 2024), no experimental structures with a probability of homology > 0.2 and an E-value < 1 could be found throughout the entire Protein Data Bank (Berman et al., 2000) or the CATH protein structure classification database (Orengo et al., 1997). When searching among predicted structures, significant hits with a probability of homology of 1 and an E-value < 1 were identified in the AlphaFold database (Varadi et al., 2022), but they all corresponded to either ApiAP2 proteins or uncharacterized apicomplexan proteins. More distant structural homologs (0.5 < probability < 1 and E-value > 1) were also predicted among uncharacterized bacterial proteins, but no significant structural homologues with identified functions could be found among predicted structures. Searching against the ESM Metagenomic Atlas (Lin et al., 2023) provided no further insight into the function of the ACDC domain.

A second search against the Protein Data Bank using the PDBeFold server (Krissinel & Henrick, 2004) allowed us to identify two structures displaying very weak similarity to the ACDC fold (Z-score ≃ 2, Q-score ≃ 0.2, r.m.s.d. ≃ 3 Å over ∼65 aligned C^α atoms; Supplementary Fig. S3). Interestingly, these two domains belong to proteins involved in transcription regulation: (i) the C-terminal domain of the yeast J-protein ZUO1, which is implicated in the activation of the transcription factor Pdr1 (Ducett et al., 2013; PDB entry 2lwx chain A), and (ii) the N-terminal four-helix bundle domain of the human death-domain associated protein DAXX acting as transcription co-repressor and histone H3.3 chaperone (Hoelper et al., 2017; PDB entry 5y6o chain A). According to the CATH classification, these two domains belong to the (1.10.8.840) and (1.10.8.810) superfamilies, respectively, with an orthogonal bundle architecture. However, the poor statistics of the superimpositions with the ACDC domain allow us to consider the ACDC left-handed orthogonal four-helix bundle to be a new topological superfamily characterized by a kink in helix α3.

A more in-depth analysis of the hydrophobic core of the ACDC domain revealed that the interactions that stabilize the four-helix bundle, and more particularly the kink of helix α3 around helix α1, rely on the most conserved residues found across 45 aligned sequences of ACDC domains from distant apicomplexan homologues (https://www.ncbi.nlm.nih.gov/Structure/cdd/cddsrv.cgi?uid=434166; Supplementary Fig. S4). These findings support the uniqueness of the ACDC structural domain characterized in this study and its potential as a drug target.

3.4. Docking analysis of plasmodial inhibitors

On the surface of the ACDC domain structures, in the characteristic kink region of helix α3, a patch of conserved residues form a hydrophobic pocket around which we defined a screening box to perform in silico molecular docking (Fig. 3 and Supplementary Table S2). The 13 533 small molecules from the Tres Cantos Antimalarial Set (TCAMS; Gamo et al., 2010) were tested against all three ACDC domain structures solved in this study. Protein side chains were fixed in a rigid conformation, except for three residues of the targeted pocket (Gln71, Leu119 and Asn122 in PfAP2-I), which were designated as flexible to accommodate ligand binding. The best hits were selected based on the lowest SMINA docking energy scores, lowest r.m.s.d. from the top-ranked poses (Fig. 4a) and proximity to the targeted conserved pocket. This proximity was quantified by the shortest distance from any ligand atom to the geometric center of specific residues (Gln71, Cys118 and Asn122 in PfAP2-I; Cys68, Gln71 and Leu119 in PvAP2-I; Leu712, Leu715 and Ile694 in PfAP2-O5). The typical affinity score hovered around −6.7 kcal mol⁻¹ (Supplementary Fig. S5). Since the targeted hydrophobic pocket is formed by conserved residues, we prioritized compounds demonstrating consistently low energy scores across the three ACDC domains. Out of the over 13 000 compounds, only seven met these stringent selection criteria: TCMDC-124223, TCMDC-124284, TCMDC-125666, TCMDC-134543, TCMDC-137354, TCMDC-139888 and TCMDC-140881 (Table 2). The compounds TCMDC-124223 and TCMDC-125666 are highly similar, with a single missing bond. TCMDC-124284 and TCMDC-137354 are also chemically related, with a methyl group replaced by an O atom.

Table 2
Selected compounds from the TCAMS

Our favorite compound TCMDC-134543 is listed first, followed by four other lead compounds, then the two compounds exhibiting a tendency to dissociate from the protein during MD simulations, TCMDC-124223 and TCMDC-125666, and finally the negative control TCMDC-142297. The optimal docking energy score of each compound is given for the ACDC domain of each of the three proteins. When available (nd, not determined), the inhibition characteristics of these selected TCAMS compounds are indicated (Gamo et al., 2010).

Compound	Structure	Optimal docking energy score (kcal mol^−1)		IC50 against P. falciparum
TCMDC-134543		PfAP2-I	−9.4	0.89 µM
		PvAP2-I	−8.8
		PfAP2-O5	−10.2
TCMDC-124284		PfAP2-I	−10.1	0.52 µM
		PvAP2-I	−9.0
		PfAP2-O5	−10.8
TCMDC-137354		PfAP2-I	−10.3	0.16 µM
		PvAP2-I	−9.0
		PfAP2-O5	−11.1
TCMDC-139888		PfAP2-I	−9.2	1.13 µM
		PvAP2-I	−8.2
		PfAP2-O5	−8.9
TCMDC-140881		PfAP2-I	−9.8	0.84 µM
		PvAP2-I	−8.2
		PfAP2-O5	−10.1
TCMDC-124223		PfAP2-I	−11.3	nd
		PvAP2-I	−8.9
		PfAP2-O5	−10.8
TCMDC-125666		PfAP2-I	−10.3	nd
		PvAP2-I	−9.3
		PfAP2-O5	−11.7
TCMDC-142297		PfAP2-I	−4.3	nd
		PvAP2-I	−4.2
		PfAP2-O5	−3.2

Figure 3 Targeted hydrophobic pocket. (a) Electrostatic surface of the ACDC domain of PfAP2-O5 shown with positively and negatively charged residues (from −67.9 to +67.9) colored blue and red, respectively. Hydrophobic regions are white. Conserved residues (Supplementary Fig. S4) are highlighted as sticks. (b) Screening box used for docking highlighted on the structure of the PfAP2-O5 ACDC domain shown in the same orientation as in (a).

Figure 4 Docking results for the ACDC domains of PfAP2-I, PvAP2-I and PfAP2-O5. (a) 3D scatter plot showing the docking energy scores (kcal mol^−1) for the three systems, highlighting the seven selected compounds. TCMDC-142297, which has an unfavorable affinity score but showed minimal distance to the pocket and low average r.m.s.d., was also selected as a negative control for MD simulations. The points are colored according to the r.m.s.d. (Å) observed in the PfAP2-O5 system from blue (low values) to yellow (high values), whereas the size of the points correlates with r.m.s.d. values in the PvAP2-I system. (b) Selected MD pose for the standout ligand TCMDC-134543 in the conserved pocket of PvAP2-O5. Dashed lines represent hydrophobic interactions with Tyr626, Asn629, Phe630, Leu633, Thr634 and Ile641 from helix α1 and Leu687, Ile694 and Leu699 from helix α3 and loop α3–α4.

These seven selected ligands plus a negative control (Fig. 4a and Table 2) were then submitted to molecular-dynamics (MD) simulations for 100 ns (Supplementary Videos S1, S2 and S3 and Fig. S6). Stability throughout the MD simulations was monitored by tracking the minimum distance between the ligands and the pocket, thereby assessing the ability of the ligand to remain stably associated with the protein surface and, consequently, its potential as an ACDC binder. We further evaluated each ligand through visual analysis of its dynamic behavior. As expected, the negative control, TCMDC-142297, disengaged from the pocket early in the simulations and remained detached across all three systems in most of the trajectories. TCMDC-124223 also exhibited a tendency to dissociate and not rebind, while the related molecule TCMDC-125666 only maintained its docked position with PfAP2-O5. However, five of the selected molecules, TCMDC-124284, TCMDC-134543, TCMDC-137354, TCMDC-139888 and TCMDC-140881, exhibited commendable stability throughout the MD simulations, signifying stable interaction with all ACDC domains tested. The standout performer was TCMDC-134543, which bound robustly to all three domains (Fig. 4b), underscoring its potential as a versatile ligand for several ACDC domains.

4. Discussion

The ACDC domain is exclusively found among Apicomplexa, mostly in transcription factors of the ApiAP2 family. In P. falciparum, nine of the 27 members of this family contain an ACDC domain (Singhal et al., 2024; Jeninga et al., 2019; Oehring et al., 2012). Here, we show that despite low sequence conservation and position within the proteins (N-terminal versus C-terminal), the ACDC domains from two distinct ApiAP2 proteins from P. falciparum adopt a similar structure. In the ACDC structure of PfAP2-O5 used as reference, the left-handed orthogonal four-helix bundle is formed by two αα-hairpins (α1–α2 and α3–α4) linked by an αα-corner (loop α2–α3) (Efimov, 1993), resulting in a X-shaped structure characterized by a crossing angle of about 70° and a topology never previously observed in other four-helix bundle types (Haddad et al., 2017; Harris et al., 1994; Kohn et al., 1997).

Four-helix bundles are found among a wide range of functionally diverse proteins and can bind to a variety of molecules, including nucleic acids, proteins, lipids and small molecules. This does not help to assign a function to the previously unobserved topology of the ACDC domain. However, the propensity to make strong contacts in crystal packing suggests that ACDC domains may be involved (i) in interactions between protein partners and the ApiAP2 proteins or (ii) in dimerization, in line with the tandem occurrence of the TGCA DNA-sequence motif recognized by PfAP2-I (Santos et al., 2017). We propose that within the context of the full-length protein, the ACDC domain is involved in protein–protein interactions. Interestingly, an ACDC domain has recently been identified in a Plasmodium protein (AP2R-2) lacking functional AP2 domains (Yuda et al., 2021). In P. berghei, PbAP2R-2 interacts with the ApiAP2 protein PbAP2-FG2, which also contains an ACDC domain, to form a transcriptional repressor complex (Nishi et al., 2023), supporting the hypothesis that the ACDC domain mediates protein–protein interactions. Future studies focused on the molecular function of the ACDC domain are required to shed light on the role of ACDC in ApiAP2 transcriptional regulators.

Given that the ACDC structure is highly conserved and is exclusively found in apicomplexan parasites, which are important pathogens of humans and domestic animals, this domain can be considered to be a new therapeutic target. We modeled the ability of antimalarial inhibitors from the TCAMS drug library to bind in silico to a conserved hydrophobic pocket that is found in all three ACDC domains studied, identifying five lead compounds. Future work will focus on elucidating the function and essentiality of the ACDC domain in the role of ApiAp2 DNA-binding proteins in regulating parasite development. This will include validation of the binding of the top ligands identified in this study (especially TCMDC-134543) to purified ACDC domains. Further chemical optimization of these compounds, aided by structure-guided studies, will help to rationally improve ligand specificity and selectivity. If our hypothesis that the ACDC domain acts as a protein–protein interaction domain is confirmed, these optimized ligands could disrupt these protein interfaces. Our work thus provides a new therapeutic opportunity for the development of pan-apicomplexan inhibitors targeting ACDC domain-containing ApiAP2 proteins, which are known to play essential roles in several stages of parasite development.