CSGID Solves Structures and Identifies Phenotypes for Five Enzymes in Toxoplasma gondii

Toxoplasma gondii, an Apicomplexan parasite, causes significant morbidity and mortality, including severe disease in immunocompromised hosts and devastating congenital disease, with no effective treatment for the bradyzoite stage. To address this, we used the Tropical Disease Research database, crystallography, molecular modeling, and antisense to identify and characterize a range of potential therapeutic targets for toxoplasmosis. Phosphoglycerate mutase II (PGMII), nucleoside diphosphate kinase (NDK), ribulose phosphate 3-epimerase (RPE), ribose-5-phosphate isomerase (RPI), and ornithine aminotransferase (OAT) were structurally characterized. Crystallography revealed insights into the overall structure, protein oligomeric states and molecular details of active sites important for ligand recognition. Literature and molecular modeling suggested potential inhibitors and druggability. The targets were further studied with vivoPMO to interrupt enzyme synthesis, identifying the targets as potentially important to parasitic replication and, therefore, of therapeutic interest. Targeted vivoPMO resulted in statistically significant perturbation of parasite replication without concomitant host cell toxicity, consistent with a previous CRISPR/Cas9 screen showing PGM, RPE, and RPI contribute to parasite fitness. PGM, RPE, and RPI have the greatest promise for affecting replication in tachyzoites. These targets are shared between other medically important parasites and may have wider therapeutic potential.


INTRODUCTION
Toxoplasma gondii is one of the most significant parasites that impacts human health, with estimates that as many as one third to one half of the human population are infected (Montoya and Liesenfeld, 2004;Furtado et al., 2011;Torgerson and Mastroiacovo, 2013;Flegr et al., 2014;McLeod et al., 2014;Lykins et al., 2016). A relative of the parasite that causes malaria, T. gondii is an intracellular parasite that has two major life stages in humans, tachyzoites and bradyzoites. Tachyzoites cause acute infection, while bradyzoites are the encysted, dormant life stage responsible for reactivation disease. While treatment is available for the acute infection, there is currently no effective medication for the bradyzoite stage . Additionally, parasites can be passed to a fetus in utero when a pregnant woman is acutely infected during gestation. This can cause chorioretinitis and neurological complications in the fetus (McLeod et al., 2012). Moreover, there is increasing understanding of the potential long-term sequelae of chronic infection with T. gondii on risk of neurodegenerative disease and malignancy . Treatment for active infection exists but is limited by toxicity and hypersensitivity. Thus, new therapeutic targets and medicines are needed, with several potential solutions in development (Zhou et al., 2014;McPhillie et al., 2016;Sidik et al., 2016).
At the Center for Structural Genomics of Infectious Diseases (CSGID), the first Toxoplasma Structural Genomics Pipeline was established. Subsequently, CSGID began selecting parasite proteins for structural characterization using established approaches capable of successful identification of potential drug targets, coupled with the Tropical Diseases Research (TDR) Database (Anderson, 2009;Crowther et al., 2010;Magariños et al., 2012). Herein, 5 soluble enzymes were selected for further study. This process was made possible due to the integration of large amounts of genomic, biochemical, and pharmacological data by the TDR Database, which provides evidence collectively generated by the scientific community concerning potential molecular targets and inhibitory compounds that have properties consistent with Lipinski's rules for orally available drugs (Lipinski, 2004). The targets studied herein were crystallized and their structures characterized, as structural studies have potential to inform molecular targeting and medicinal chemistry can facilitate development of novel anti-parasitic compounds.
We further hypothesized that using phosphorodiamidate morpholino oligomers linked to a cellular delivery moiety, such as either an octaguanidinium dendrimer [Vivo-Morpholinos (vivoPMOs)], or arginine-rich peptide, we would decrease expression of these enzymes, identified as potential drug targets by the Toxoplasma Structural Genomic Pipeline, in YFPexpressing T. gondii tachyzoites, and that down-regulation of these enzymes would result in decreased replication as quantified by fluorescent intensity. The approach of using morpholinos to target specific parasitic enzymes has been successful in previous studies (Lai et al., 2012;McPhillie et al., 2016). VivoPMOs are typically used to decrease gene expression by one of two different mechanisms, namely mechanical disruption of interactions between RNA and snRNP, thereby preventing splicing of introns, resulting in nonsense-mediated decay of the transcript and/or defective protein upon translation, and through direct prevention of translation by blocking interactions between mature mRNA and the ribosome. In preventing effective protein expression, we could determine whether a particular enzyme contributed to parasite replication, suggesting its potential as a therapeutic target.
Molecular transporters can deliver PMOs and small inhibitory molecules of therapeutic value. Transductive peptides or octaguanidinium dendrimer of a Vivo-Morpholino (Gene Tools, Philomath, Oregon) deliver PMOs or other molecules across cell membranes. Octaarginine can carry small molecules into the retina (McLeod et al., 2013). Similar arginine-rich cellpenetrating peptides can access other places where medication transport is problematic; for example, rabies virus glycoproteintagged small molecules are capable of passing through the bloodbrain barrier and octaarginine-conjugated small molecules, for example, cross into encysted bradyzoites (Samuel et al., 2003;Liu et al., 2009).
The enzymes selected from the TDR database as small and tractable for expression and crystallization included: phosphoglycerate mutase II (hereafter referred to as PGM), nucleotide diphosphate kinase (NDK), ribulose phosphate 3epimerase (RPE), ribose-5-phosphate isomerase (RPI), and ornithine aminotransferase (OAT). Information about candidate inhibitors of these apicomplexan enzymes is summarized in Table 1.

Plasmodium falciparum
Replication L-canaline, 5-fluoromethylornithine Kito et al., 1978;Storici et al., 1999;Müller et al., 2009;Sturm et al., 2009;Kronenberger et al., 2014 FIGURE 1 | The Toxoplasma Structural Genomics Pipeline identified promising targets with good predicted druggability (Phase III) and outstanding interest to the Toxoplasma research community (Phase I, II). Toxoplasma proteins were selected for Phase I from published work up to 2009 and communicated with the research laboratories. Phase II included proteins from published and unpublished studies suggested by international researchers at the end of 2011. Phase III candidates were selected from the Tropical Diseases Research Database (TDRT). Proteins judged very unlikely to crystallize based on analysis by XtalPred were eliminated (Slabinski et al., 2007). Several structures solved have already been described Tonkin et al., 2015;Dubey et al., 2017). This crystallography pipeline remains available for production and solution of structures of proteins for scientists in the Toxoplasma research community.
1.5 mM magnesium acetate, 1 mM CaCl 2 , 0.08% n-Dodecyl β-D-maltoside (DDM), 5 mM β-mercaptoethanol (BME)] pH 7.8 followed by sonication on ice. Crude sonication mixture was centrifuged at 19,000 rpm, 4 • C for 40 min to obtain soluble fraction containing target protein, which was applied onto a 5-ml Ni-NTA column (GE Healthcare, Piscataway, NJ) for purification. The column was washed with buffer containing 10 mM Tris-HCl pH 8.3, 500 mM NaCl, 25 mM imidazole and 5 mM BME to remove non-specifically bound E. coli proteins, followed by elution of target protein with 500 mM imidazole in the 10 mM Tris-HCl buffer pH 8.3 containing 500 mM NaCl and 5 mM BME (buffer A). A HiLoad TM 26/60 Superdex TM 200 column (GE Healthcare, Piscataway, NJ) was used to further purify target protein in the buffer A. Purity of all proteins was analyzed by SDS-PAGE. Pooled fractions were concentrated and stored at −80 • C for further use or used in screening crystallization conditions.

Crystallization, X-Ray Data Collection, and Structure Determination
Each target protein was crystallized by the vapor-diffusion sitting-drop method mixing 1 µL of the protein in buffer A and 1 µL of a crystallization screen solution at 22 • C. Single crystals were soaked in a crystallization condition for cryoprotection and flash frozen in liquid nitrogen for monochromatic X-ray data collection. Data were collected at 100 K from a single frozen crystal at the LS-CAT beamline 21-ID-F (λ = 0.97872 Å) at the Argonne National Laboratory (ANL), Advanced Photon Source (APS). Diffraction images were collected in oscillation mode and processed with HKL-3000 (Minor et al., 2006). Crystal structures of TgPGM, TgNDK, TgRPE, and TgOAT were determined by molecular replacement using Phaser (McCoy et al., 2007) from CCP4 suite (Winn et al., 2011). Initial molecular replacement solutions were rebuilt with ARP/wARP (Morris et al., 2003). Crystal structure of TgRPI was determined by single-wavelength Frontiers in Cellular and Infection Microbiology | www.frontiersin.org anomalous dispersion (SAD) method in HKL-3000 (Minor et al., 2006). Non-Crystallographic Symmetry (NCS) restrains and Translation-Libration-Screw (TLS) groups refinement in REFMAC v.5.7 (Murshudov et al., 2011) were used to improve the quality of the structures. Coot (Emsley and Cowtan, 2004;Emsley et al., 2010) was used to manually check structures after each cycle of refinement in REFMAC and correct for side chain rotamers and fitting. The final models were validated with the Protein Data Bank (PDB) validation server (ADIT validation server; https:// validate-rcsb-2.wwpdb.org/) and MolProbity (http://molprobity. biochem.duke.edu/) (Davis et al., 2007;Chen et al., 2010

Data Accessibility
Coordinates and structure factors of the determined crystal structures were deposited to Protein Data Bank (www.rcsb.org) (Berman et al., 2000). Diffraction images for the deposited crystal structures can be found at the CSGID website (http://www.csgid. org/csgid/pages/home).

VivoPMO Design
The vivoPMOs were designed using genomic DNA sequences obtained from ToxoDB (accession numbers for genomic DNA sequences: PGM-TGME49_297060, NDK-TGME49_295350, RPE-TGME49_047670, RPI-TGME49_039310, OAT-TGME49_069110) with exon/intron junctions identified. One of these junctions was identified in each target gene and a vivoPMO was designed to be complementary to nucleotides on both sides of said junction. A diagrammatic representation of vivoPMO structure and RNA binding is in Figure 7A. The sequences of these morpholinos can be found in Figure 7B.

VivoPMO Efficacy Assay
HFFs were grown in black, flat-bottomed 96-well microplates. HFFs were infected with 3,200 Type I RH parasites expressing yellow fluorescent protein (YFP). This allowed quantification of parasites in vitro post-treatment with vivoPMO. The parasites were incubated with the cells for 1 h, to allow sufficient time for invasion of HFFs, and were then treated with vivoPMO. Control triplicates with only fibroblasts and with pyrimethamine and sulfadiazine (the current standard of treatment for T. gondii infection) were also conducted. A concentration gradient of YFP parasites was also established, allowing quantification of knockdown. Several replicates of this efficacy assay were completed applying different concentrations of vivoPMO (2.5, 5, 10, and 20 µM). The cells and parasites were then incubated at 37 • C for 72 h. This timing was previously established in other work. Fluorescence was measured using a Bio-Tek Synergy TM H4 Hybrid Multi-Mode Microplate Reader. This methodology was consistent with previous work using morpholinos in T. gondii (Lai et al., 2012).

VivoPMO Toxicity Assay
HFFs were grown in 96-well microplates, as in the efficacy assay. A gradient of dimethyl sulfoxide (DMSO) was used to quantify the amount of cell death caused by the vivoPMO in vitro. Different concentrations of vivoPMO (3.5, 5, 10, and 20 µM) were used to identify the level at which toxicity occurred. Following 72 h of incubation at 37 • C, each well was treated with 10 µL WST-1, which reacts in metabolically active, viable cells through a complex set of chemical reactions dependent upon glycolytic NADPH production to form formazan dyes, which can be detected via a colorimeter. This method is also consistent with previous work (Lai et al., 2012).

Data Analysis of Knockdown With VivoPMO
Knockdown was analyzed statistically using student T-test comparing parasite fluorescence between enzyme-specific vivoPMO and off-target vivoPMO. Additionally, the enzymespecific vivoPMOs were compared to the levels of fluorescence at the standardized parasite load (2000 YFP-expressing parasites per well). Student T-test was also used to analyze toxicity data, comparing levels of optical density at 420 nm for untreated fibroblasts to treated cells. Statistical analysis was performed using STATA. Results were considered significant with p < 0.05.
Values in parentheses are for the highest resolution shell. † Statistics are based on PROCHECK (Laskowski et al., 1993).

Antibody Production
Antibodies were raised in mice at the University of Strathclyde (CR, SW). Briefly, mice were given two injections of formulated protein with NISV (non-ionic surfactant vesicle). The vesicles were made by melting mono-palmitoyl glycerol, cholesterol and dicetyl-phosphate (All from Sigma, UK) in a molar ratio of 5:4:1. 10 days after the final injection, serum was collected and tested by Western Blot using recombinant protein and Toxoplasma lysate.

Immunofluorescence Assay (IFA)
HFF cells were grown to confluence on sterilized coverslips in 24-well plates. Cells were fixed in 3% paraformaldehyde ∼20-24 h after infection with Type I RH-strain tachyzoites and permeabilized in 0.25% Triton X-100. Serum from immunized mice, coupled with another primary antibody, RPS13, was applied at 1:500 dilution in PBS 1x/3% BSA/Triton-X-100 and detected using either Texas Red-conjugated goat anti-mouse antibody, or Alexa-488-conjugated goat anti-rabbit antibody. DAPI was used to stain for DNA. Coverslips were mounted with Antifade (Molecular Probes, Eugene, OR), and images were analyzed by high-resolution fluorescence using deconvolution protocols. Microscopy was performed with an inverted microscope (IX81; Olympus).

Comparison to Predicted Essentiality via Literature CRISPR/Cas9 Screen
As an approach to identify whether target enzymes might be essential to parasite viability and fitness, a survey was done of phenotypic scores previously published in a genome-wide CRISPR/Cas9 screen (Sidik et al., 2016). Gene IDs were used and the average of the three available scores was taken. Negative scores were considered likely to be significant contributors to parasite fitness. Data of the genome-wide CRISPR/Cas9 screen are subsequently annotated in ToxoDB (http://toxodb. org/toxo/), including these gene IDs: TgPGM: TGME49_297060, TgNDK: TGME49_295350, TgRPE: TGME49_247670, TgRPI: TGME49_239310, TgOAT: TGME49_269110.

Structural Genomic Pipeline for Toxoplasma
The NIAID Structural Genomics Centers selected 265 proteins for the Toxoplasma Structural Genomic Pipeline in three phases (Figure 1). Phase I and II were a collaborative effort between CSGID and the Toxoplasma research community. Proteins were selected from published and unpublished work up to 2011. Selection was based on the mechanisms and pathways that are important for parasite infection and survival in human and FIGURE 3 | Crystal structure of TgNDK (A). Ribbon representation of TgNDK monomer (left side) colored blue (N-terminus) to red (C-terminus) and trimeric and hexameric assemblies of TgNDK (middle and right side) (B). Pairwise structural alignment of TgPGM (gray; SO4 is shown in sticks) and human NM23-H2 transcription factor [cyan (chain E with bound 2 ′ -deoxyguanosine-5 ′ -monophosphate (DG) and 2 ′ -deoxyadenosine-5 ′ -monophosphate (DA)] and magenta [chain B with bound 2 ′ -deoxyguanosine-5 ′ -monophosphate (DG)] showing active site with residues of TgNDK shown in sticks and labeled in one-letter code. Equivalent residues of human NM23-H2 transcription factor are displayed.
animal hosts. Protein sequences were analyzed by XtalPred and final selection was also based on their crystallization feasibility (Slabinski et al., 2007). Phase III utilized the TDR Database (Anderson, 2009;Crowther et al., 2010;Magariños et al., 2012). TDR integrated and weighed candidate drug targets based on extensive genetic, biochemical, pharmacologic, compound desirability and computationally-predicted druggability characteristics. Herein, 5 soluble enzymes were selected for further study. The genes encoding these enzymes were cloned, proteins expressed and purified, crystallized, and crystal structures were determined (Figures 2-6). The crystal structures have been deposited in the PDB database in accordance with CSGID and NIH policies. PDB codes can be found in Figure 1 and Table 2.
The structure of the Plasmodium falciparum PGM (Pf PGM; PDB 3kkk; DOI: 10.2210/pdb3KKK/pdb) is the closest homolog of TgPGM with estimated 75% sequence identity and ∼0.6 Å r.m.s.d. over 229 Cα atoms. The structure of the homologous bacterial enzyme from Burkholderia pseudomallei (PDB 3fdz; 67% sequence identity; r.m.s.d. ∼1.0 Å over 230 Cα atoms) determined in the complex with (2R)-2,3-diphosphoglyceric reveals that TgPGM possesses identical active site residues with an essential phospho-acceptor residue His26 (Davies et al., 2011). Preliminary validation of this active site in Schrödinger SiteMap gave a favorable score of 0.98. Structural alignment of BpPGM and TgPGM demonstrates significant main-and side-chain conformational changes at the protein active site that would most likely take place in TgPGM during the interconversion of its substrate 3-phosphoglycerate to 2phosphoglycerate ( Figure 2C). Conformational differences were also observed for residues located on the tip of the scorpion-like tail.

Nucleoside Diphosphate Kinase (NDK)
Crystal structures of a putative TgNDK were determined and refined to 2.4 Å (PDB 4o0n) and 1.7 Å (PDB 5bxi) resolution ( Figure 3A and  Sequences for all enzymes were designed to be complementary to regions overlapping splice sites so as to prevent appropriate interaction between mRNA and the spliceosome predicted to lead to the production of a non-functional protein product. atoms) and monomers in 5bxi align with ∼0.2-0.6 Å r.m.s.d (153 Cα atoms). In both structures, TgNDK forms a hexamer with an average BSA of ∼18,900 Å 2 ( Figure 3A). Comparison of TgNDK with known crystal structures of homologs proteins suggests that the TgNDK hexamer is likely a biologically functional assembly (Min et al., 2002;Vieira et al., 2015).
Residues 156-160 (GENLY) of the C-terminal tag in chains D and G only of the 5bxi structure were modeled, while they are absent in all chains of the 4o0n structure. These residues protrude to the solvent and make contacts with symmetryrelated hexamer. The side chain of Tyr160 is spatially positioned in a nucleotide base-binding pocket of the active site of a symmetry-related molecule (not shown). In addition, residues 56-64 (DLKGKPFFP; chain C of 5bxi) that belong to the Cterminus of helix α3, 3 10 helix η1, η1-α4 loop, and the Nterminus of helix α4 are disordered. This peptide stretch may constitute a portion of a nucleotide base-binding pocket of the active site. Superposition of all twelve chains in 5bxi and 4o0n structures revealed that this segment has similar secondary structure conformations. Presumably, lack of favorable crystal contacts resulted in its disorder in chain C of the 5bxi structure.
TgNDK adopts an α/β/α sandwich fold with four β strands and eight surrounding α helices (Figure 3A). The putative active site of TgNDK was identified based on a pairwise structural alignment with the crystal structure of human NM23-H2 transcription factor in complex with the dinucleotide d(AG) (PDB 3bbb; Dexheimer et al., 2009). The active site comprises residues Lys14, Tyr54, Leu57, Lys60, Phe62, Leu66, Thr96, Val114, and Gly115 (Figure 3B), where analysis by Schrödinger SiteMap revealed a score of 0.99 (using PDB 4o0n), indicating a "druggable" pocket. Crystals of TgNDK (PDB 4o0n) grew in the presence of ammonium sulfate and, thus, several sulfate ions were identified and modeled during refinement. Superposition of TgNDK with the structure of human NM23-H2 transcription factor (∼67% homology; ∼0.6 Å r.m.s.d. over 148 Cα atoms) revealed that a sulfate ion (e.g., SO4 201/A) binds close to a phosphate-binding pocket of the active site ( Figure 3B). A bicarbonate ion (e.g., BCT 201/A) was modeled at a similar location in the 5bxi structure. Similar main-and side-chain rearrangements of residues of the active site as seen in NM23-H2 are expected in TgNDK upon binding its substrate.
A Zn 2+ (e.g., ZN 303/A) ion was modeled per each chain of TgRPE, with His37, Asp39, His71, Asp182 and a water molecule coordinating the metal (Figure 4B). Similar coordination has been reported for D-xylulose 5-phosphate co-crystallized with human RPE enzyme (PDB: 3ovr) (Liang et al., 2011) (Figure 4B). PDBeFold identified human RPE as the closest homolog of TgRPE (∼0.9 Å r.m.s.d. over 215 Cα atoms; 52% sequence identity). The CheckMyMetal analysis identified that Co 2+ and Cu 2+ may also bind in a similar position as zinc in TgRPE. It is known that the RPE enzymes can utilize Fe 2+ , Co 2+ , and Mn 2+ for catalysis (Jelakovic et al., 2003;Wise et al., 2004;Caruthers et al., 2005;Akana et al., 2006;Liang et al., 2011). Residues of the metal binding site are strictly conserved in the RPE enzymes. Structural superposition revealed that Fe 2+ in ferrous-, substrateand product-bound human RPE structures is ∼1.0 Å away from the position of Zn 2+ in TgRPE. We also compared TgRPE structure with two other RPEs structures (PDBs 1tqx and 1h1z) (Jelakovic et al., 2003;Caruthers et al., 2005) that have zinc and SO 2− 4 bound in similar locations as TgRPE. Positions of zinc and side chains of the Zn 2+ -coordinating residues are similar in the three compared structures (not shown). Other RPE structures with bound Zn 2+ and other than SO 2− 4 ligands (e.g., PDBs 2fli, 3qc3, and 5umf) have similar positions of Zn 2+ and side chains of the metal-binding residues (Akana et al., 2006;Joint Center for Structural Genomics, 2011;Dranow et al., 2017). Additional experiments are needed to confirm the biological importance of zinc in the TgRPE-based catalysis. This metal-binding region was analyzed for "druggability" and scored 1.04 (Schrödinger SiteMap).
Crystals of TgRPE grew in the presence of 500 mM NaCl and 2 M ammonium sulfate and, thus multiple sulfate and chloride ions were modeled to interpret additional electron density. Structure comparison analysis revealed that a sulfate ion (e.g., SO4 304/A) in TgRPE occupies the binding site of a phosphate moiety of substrate or product observed in the structure of the human RPE enzyme (PDBs 3ovq and 3ovr) ( Figure 4B). In all ligand-bound human RPE structures, a loop (e.g., the β6-α5 loop in TgRPE) supports position of a ligand at the active site, i.e., moves from its ligand-free conformation to cap a ligand bound to the enzyme. Thus, similar loop movement is expected for TgRPE upon binding of the substrate, i.e., a ligand. Apparently, binding of SO 2− 4 sufficed the loop repositioning in TgRPE ( Figure 4B).

Ribose 5-Phosphate Isomerase (RPI)
Crystal structure of a putative TgRPI was determined and refined to 2.6 Å resolution with one protein molecule per asymmetric unit (PDB 4nml; Figure 5A and Table 2). TgRPI has a larger α/β/α sandwich catalytic domain (residues 1-127 and residues 225-259) with the Rossmann topology and a smaller α/β sandwich oligomerization domain (residues 128-224) (Figure 5A). Residues 180-186 between strand β8 and a 3 10helix (η3), Arg258, and Lys259 are absent in the structure due to their disorder. BME was used in buffers during purification and crystallization and is covalently linked to Cys82 in the structure. One D-malate molecule and two chloride ions were modeled to interpret additional electron density (Figures 5A,C). Applying crystal symmetry operations PISA predicted that TgRPI might exist as unstable dimer (BSA of ∼2,430 Å 2 ) or stable dimer (BSA of ∼3,340 Å 2 ) in the crystal environment ( Figure 5B).
Visualization of both dimers revealed that the first assembly is held in place primarily by hydrogen-bonded interactions between main-chain atoms of strand β13 of each monomer's catalytic domain ( Figure 5B). The second dimer is stabilized by multiple intramolecular contacts between residues from both TgRPI domains. Homologous RPIs form similar stable dimers and, thus stable TgRPI dimer may be considered biologically relevant. PDBeFold identified several homologs of TgRPI, and we used the crystal structure of RPI from Thermus thermophilus (TtRPI; PDB 1uj6) (Hamada et al., 2003) determined in complex with arabinose 5-phosphate to identify the active site of TgRPI ( Figure 5C). Both proteins share 45% sequence homology and are superimposed with ∼1.0 Å r.m.s.d. over 218 Cα atoms. This structural alignment revealed that D-malate and chloride ion (e.g., CL 301/A) in TgRPI are bound in the protein's active site, mimicking parts of arabinose 5-phosphate in 1uj6 ( Figure 5C). D-malate makes hydrogen bonds with side chains of Lys8, Thr32, and Thr33, and several van der Waals contacts with residues of the phosphate-binding pocket of the active site (Schrödinger SiteMap score 0.97). CL 301/A is bound in the oxyanion hole of the active site and coordinated by main-chain nitrogen atoms of Gly105 and Ala107, side chain carboxyl group of Glu111, and two water molecules, HOH 415 and HOH 433 ( Figure 5C). The second chloride ion (e.g., CL 302/A) in TgRPI binds near disordered residues 179-185 and makes bonds with main-chain nitrogen atoms of Phe172 and Ile190, and the ε-amino group of Lys154 (not shown).
The catalytic domain of homologous RPIs align well, while the oligomerization domain has some distinct structural differences among various species. For example, a peptide stretch between residues Leu160 and Arg169 of the oligomerization domain of TgRPI is longer than the equivalent region in TtRPI. In the TgRPI, structure this segment is helix α6. We have found that a similar helical element is present in E. coli RPI (Zhang et al., 2003) (PDB 1o8b; 45% sequence homology; ∼1.4 Å r.m.s.d. over 164 Cα atoms) and P. falciparum RPI (Holmes et al., 2006) (PDB 2f8m; 45% sequence homology; ∼1.1 Å r.m.s.d. over 233 Cα atoms). Another example, strands β8 and β9 in TgRPI are separated by a 3 10 -helix (η3) and disordered residues 179-185, while similar region in TtRPI, EcRPI, and Pf RPI is shorter, ordered, and extends toward the N-terminus of helix α4 of the superimposed TgRPI. Thus, η3 and residues 179-185 seem to be unique to TgRPI. Its length, proximity to the phosphate-binding pocket, and potential flexibility suggest that it may participate in the catalysis.

Ornithine Aminotransferase (OAT)
The crystal structure of TgOAT in complex with pyrodoxal-5 ′phosphate (PLP) was determined at 1.2 Å resolution (PDB 4nog; Figure 6A and Table 2) and contains two protein chains in the asymmetric unit (∼0.3 Å r.m.s.d. over 422 Cα atoms) that form a well-studied functional dimer with BSA of ∼5,770 Å 2 (Shah et al., 1997;Shen et al., 1998;Markova et al., 2005;Vedadi et al., 2007;Jortzik et al., 2010) (Figure 6B). Each monomer has a mixed α/β structural fold and consists of the cofactor PLP-binding domain (residues 87-336), an N-terminal domain (residues 17-86), and a C-terminal domain (residues 337-441) (Figure 6A). The PLP-binding domain comprises the eight-stranded central βsheet linked to 10 α-helices and 4 short α-helical segments. The N-and C-terminal domain comprises three α-helices bounded to a β-sheet of three and four β strands, respectively. The difference between TgOAT monomers is observed at the position of N-terminal α-helix and is likely affected by crystal packing. BME used in the protein buffer solutions forms a covalent bond to Cys353 in both TgOAT monomers. One 1,3,5-Tris(4carboxyphenyl)benzene (BTB), polyethylene glycol (PEG) and acetate (ACT) molecule were modeled to interpret additional electron density in chain B.

In vitro Studies of Targeted VivoPMO on T. gondii Replication
HFFs were infected with YFP-expressing T. gondii tachyzoites and treated with vivoPMOs targeted against all five enzymes of interest. When treated at 10 µM concentrations, each morpholino resulted in approximately 50% reduction in fluorescence in vitro (between 44% for PGM-targeted vivoPMO to 56% reduction for OAT-targeted vivoPMO). This was compared to the off-target, control morpholino, which resulted in <10% fluorescence reduction. When compared statistically using student T-test, levels of fluorescence in vitro were statistically decreased (p < 0.05) when treating with targeted vivoPMO relative to untreated fibroblasts infected with 2000 YFP-expressing tachyzoites as well as when compared to infected cells treated with off-target morpholino. A representative experiment showing these findings is in Figure 8A. Data from all replicate experiments of the morpholino efficacy assays, as well as statistical calculations and ratios reflecting efficacy of targeted morpholino to off-target can be found in Supplementary Table 1.
To confirm that these findings of decreased fluorescence in vitro was not secondary to host-cell toxicity from the vivoPMOs, a WST-1 assay was used. HFFs treated with 10 µM concentrations of enzyme-targeted vivoPMOs were statistically indistinguishable (p > 0.05) from untreated cells by Student T-test. Toxicity was demonstrated at higher vivoPMO concentrations of 20 µM. A representative experiment demonstrating the absence of host-cell toxicity with vivoPMO treatment at 10 µM is in Figure 8B.
The ratio of targeted to off-target vivoPMO are displayed in Figure 8C. This is shown for five replicate experiments (Supplementary Table 1). This demonstrates that, in cultures treated with on-target vivoPMO, parasitic replication was reduced.

Immunofluorescence Assay
Recombinant protein for the five targets was also used to produce antibodies in mice, which were used for immunostaining to determine expression patterns within the parasite. Immunolocalization was not successful with tachyzoites of Toxoplasma for four of the five enzymes. For NDK, enzyme was present in a granular pattern in the cytoplasm of tachyzoites, located around the perimeter of the parasite and the posterior part of the parasite. This can be seen in Figure 9.

DISCUSSION
The work characterized herein presents a detailed characterization of enzyme structure that can be used for modeling inhibitors of these targets and also presents approaches for studying target phenotype with vivoPMO and CRISPR/Cas9 which in combination develops a system to move forward candidate targets (Figure 10). The targeted vivoPMOs demonstrated statistically significant perturbation of parasitic replication when compared to off-target morpholinos, without concomitant host-cell toxicity, confirmed for some of these with CRISPR/CAS9 screen. Inhibition of target enzymatic function via small molecules or anti-sense might be a novel therapeutic modality, should such small molecules exist. Given the current limitations of anti-parasitic medicines for the treatment of toxoplasmosis, new pharmacotherapy is of significant interest. The multi-step approach, detailed here, of in vitro inhibition through anti-sense techniques accompanied by detailed structural characterization to identify possible exploitable differences between host and parasite enzymes. Confirmation by referring to a recently published CRISPR/Cas9based analysis, can also help to suggest importance of targets in T. gondii tachyzoites (Sidik et al., 2016). This provides another level of evidence that particular targets may be important for the parasites replication. This approach was performed for each of five enzymes identified as having potential biologic importance and with favorable predicted druggability using the TDR methodology. This favorable druggability was also confirmed with in silico modeling using Schrödinger SiteMap, which indicated each target demonstrated "druggable" pockets. Each enzyme will be considered sequentially hereafter.
Phosphoglycerate mutase II, an enzyme at the core of the glycolytic pathway, catalyzes the transition from 3phosphoglycerate to 2-phosphoglycerate, an important preparatory step upstream of enolase and pyruvate kinase. Although little work had been done on this Apicomplexan enzyme as a potential drug target, we inferred that interrupting glycolysis would prevent the production of pyruvate by glycolysis, and thereby reduce input into the Kreb's Cycle. This, in turn, will reduce the number of electron carrying compounds (NADH, FADH 2 ), which would markedly reduce ATP production. Interestingly, glycolysis has been targeted successfully in other parasites, and also has been demonstrated to have importance to host cell egress, as well as maintenance of energy reserves when the parasite is found outside of host cells (Ananvoranich et al., 2006;Fleige et al., 2007;Pomel et al., 2008;Lin et al., 2011;Singh et al., 2013). These observations, coupled with our own, are consistent with the observed effect on parasite replication. It appears that putative inhibitors could make this a robust tachyzoite molecular target with a compound easily available to test. With the present crystal structure, further structure-based molecular design approaches (such as virtual screening) is possible given the favorable SiteMap score (0.98).
Nucleotide diphosphate kinase catalyzes the movement of phosphate from a nucleoside triphosphate to a nucleoside diphosphate (e.g., GTP + ADP -> GDP + ATP). Naturally, disruption of this process could have an impact on the energy economy within the parasite, as ATP would not be available for important cellular tasks related to DNA replication and the production of more parasites. Additionally, this is a stress kinase, and thus its impact on the stressed organism (through a tetracycline-dependent gene expression construct, RPS-13) would be useful to explore. An immunofluorescence assay revealed peripheral and posterior concentration of staining. Interestingly, another intracellular pathogen, Leishmania amazonensis, has been shown to secrete NDK to prevent host-cell autolysis (Kolli et al., 2008). It would be of interest to determine whether this enzyme plays a similar role in Apicomplexans like T. gondii. Of note, NDK inhibition by candidate compounds has been demonstrated to have efficacy against certain species of Leishmania in vitro (Vieira et al., 2015;Mishra et al., 2017). Crystal structure of TgNDK suggests that the protein forms a hexamer with conserved nucleotide binding sites. Pairwise structural alignment revealed that active site of TgNDK may undergo similar conformational changes as its closest homolog, humanNM23-H2 transcription factor (Figure 3B). High structural homology to human transcription factor (NM23-H2) in the residues in the catalytic site indicate that selective inhibitors that do not act on human NM23-H2 will be needed. Possible strategies that will facilitate such selectivity include antisense, CRISPR, aptamer-based approaches where the DNA/RNA sequences are divergent, among others. When the dinucleotide is present in the active site, the residues in both Tg and Hs structures have similar side chain orientations FIGURE 9 | Immunofluorescence Assay. This series of images depict the localization of nucleoside diphosphate kinase, one of the molecular targets herein discussed. Note the concentration of red fluorescence at the periphery of the parasite. This is consistent with the localization of secreted proteins like Gra1, suggesting that NDK could be secreted by the parasite. This has precedents in other pathogens, including M. tuberculosis and Leishmania.
FIGURE 10 | Summary figure showing the T. gondii drug discovery pipeline as described herein. The TDR Database provided important initial insights into potentially important parasite proteins (Agüero et al., 2008;Magariños et al., 2012). Identification of proteins amenable to crystallographic analysis, solution of their structures, abrogation with vivoPMO and confirmation of essentiality by querying the CRISPR/Cas9 database completed the pathway, identifying enriched targets with improved therapeutic potential.
( Figure 3B), suggesting inhibitor selectivity between the species might be difficult to achieve. On the other hand, when the dinucleotide is absent, the active site loop comprising of residues G59-F62 (Tg numbering) is shifted significantly (3-4 Å) between species. Additionally, TgNDK residue K60 is residue R58 in the human ortholog, offering different hydrogen-bond capacities to putative ligands. These observations suggest opportunities for the selective design of TgNDK inhibitors using structure-based molecular modeling techniques. Wang et al. reported selective NM23-H2 (human ortholog) inhibitors based on an isaindigotone scaffold . These compounds were subjected to molecular docking studies and were predicted to bind to the dinucleotide pocket. More so, they offer an obvious starting point for biological evaluation in in vitro parasite models and the subsequent design of selective TgNDK chemical probes/ligands. The next two enzymes, due to their sequential placement within the pentose-phosphate pathway, should be considered together. Ribulose phosphate 3-epimerase functions in the conversion of ribulose-5-phosphate into xylulose-5-phosphate, which is a reaction in the Calvin cycle. It is just downstream of the next target enzyme, ribose-5-phosphate isomerase, and is important for the development of a pool of NADPH, as well as in the pentose phosphate pathway that can convert monosaccharides like glucose into nucleotide precursor pentose sugars. This pathway has been of interest in targeting various organisms, including P. falciparum and Trypanosoma cruzi (Barrett, 1997;Bozdech and Ginsburg, 2005;Igoillo-Esteve et al., 2007). The biology of RPE has proven to be of particular interest in T. cruzi (Gonzalez et al., 2017). Indeed, the structure of RPE had already been characterized in P. falciparum (Caruthers et al., 2005). Determined crystal structure of TgRPE revealed that Zn 2+ binds in a putative active site. In addition, the presence of SO 2− 4 in the active site induces the β6-α5 loop to move from its presumed apo conformation in the absence of ligand, as seen in the human RPE homolog, to holo conformation in the ligand bound state ( Figure 4B). Ribose-5-phosphate isomerase is the enzyme in the pentose phosphate pathway just upstream of ribulose phosphate 3-epimerase, and it has similar functions, though it catalyzes the transition from ribose-5-phosphate to ribulose-5-phosphate. It has been identified as a potential drug target in P. falciparum, one of the causative agents of malaria, due to its necessity in creating nucleotide precursors for DNA synthesis and for maintaining a large pool of NADPH for rapid replication and the metabolism necessary for the maintenance thereof (Holmes et al., 2006). It has also been of interest and targeted in Mycobacterium tuberculosis where inhibitors have been identified (Roos et al., 2005). Additionally, this pathway has been suggested as a drug target in trypanosomes, with identified compounds and multiple mechanisms of inhibition having been demonstrated, with current patents existing (de V. C. Sinatti et al., 2017). Perhaps the most interesting feature of the determined crystal structure of TgRPI is the disordered region between residues 179-185 that comprises a part of what may appear to be an active site loop. These residues and 3 10 -helix η3 are unique to TgRPI and might be important for catalysis ( Figure 5A).
Ornithine aminotransferase, the final enzyme considered herein, is an enzyme involved in the urea cycle, TCA cycle, polyamine synthesis, and other pathways. It catalyzes a reversible reaction allowing interconversion of intermediates from ornithine to amino acids. Other enzymes which act in this pathway have been suggested as potential drug targets, and it is critical for maintaining proper amounts of free amino acids, so it was a good candidate for further study. Herein we found a modest phenotype on replication but raised antibody did not immunostain tachyzoites. Ornithine aminotransferase will be studied further in the future to resolve some of these questions and attempt to develop effective inhibitors, were it to prove to be essential. Structural analysis reveals that TgOAT shares similar structural folds to known OAT enzymes (Figure 6). We have identified that TgOAT shares a conserved cofactor and substrate-binding site with its closest homolog human GABA-AT. To gain insights into function of TgOAT, we attempted to crystallize this enzyme with several different inhibitors and inactivators. As a result, three additional crystal structures of TgOAT in complex with gabaculine (a potent inhibitor of human GABA-AT) and (S)-4-amino-5-fluoropentanoic acid have been determined (PDBs 5DJ9, 5E5I, and 5E3K). Therefore, we have performed a full kinetic and structural analysis. The details of the structure, substrate binding site, kinetic mechanism and function of TgOAT will be described in our subsequent works.
Our approach described herein is a productive way to identify molecular targets and could potentially be useful for identifying small molecule inhibitors. In silico analysis of each enzyme's surface pockets/active sites suggest "druggable" areas for binding of putative small molecules. The recent discovery that CRISPR/Cas9 has future potential for treating HIV  also raises the possibility that expression of simply validated molecular targets can be eliminated by CRISPR/Cas9. SiRNA is another possible therapeutic modality, and is being studied in several infectious diseases, including Hepatitis C, Ebola, and viral encephalitis, among others (Kumar et al., 2006;Wan et al., 2014;Watanabe et al., 2014;Thi et al., 2015). As our studies suggest, vivoPMO inhibited T. gondii replication; antisense PMO is, therefore, another potential therapeutic modality that might effectively treat parasitic infection. The data presented in Figure 8 and CRISPR/Cas9 screen suggest a phenotype for TgPGM, TgRPE, and TgRPI, but do not yet prove the targets are essential. With confirmation in the future, vivoPMO-based therapy with a less toxic molecular transporter has promise. The safety of PMO is well-documented in several clinical trials in treating genetic, cardiovascular, and infectious diseases including Duchenne muscular dystrophy (DMD), restenosis, and Marburg and Ebola hemorrhagic fevers (Kipshidze et al., 2007;Kinali et al., 2009;Cirak et al., 2011;Heald et al., 2014). It has even shown efficacy against a relative of T. gondii, the causative agent of malaria, P. falciparum (Augagneur et al., 2012). Eteplirsen (ExonDys51) is an FDAapproved PMO drug for treatment of DMD in patients who have confirmed mutation of the DMD gene that is amenable to exon 51 skipping. A clinical trial is underway using a cell-penetrating peptide conjugated eteplirsen (the PPMO technology) to increase intracellular delivery of eteplirsen for greater efficacy, lower dose and less frequent dosing (clincialtrials.gov: NCT03375255). With the advent of novel modalities, including antisense and small molecule inhibition, for the treatment of both active and latent infection, it may be possible to eradicate human T. gondii infection, and the successes achieved by physicians and scientists in combatting smallpox and dracunculiasis might extend to toxoplasmosis as well.

ETHICS STATEMENT
This study was carried out in accordance with the recommendations of The Home Office of the UK Government under the Animals [Scientific Procedures] Act 1986. All work was covered by License PPL60/4568, Treatment and Prevention of Toxoplasmosis with approval by the University of Strathclyde ethical review board.