Basis for drug selectivity of plasmepsin IX and X inhibition for Plasmodium falciparum and vivax

Plasmepsin IX (PMIX) and X (PMX) are aspartyl proteases of Plasmodium spp. that play essential roles in parasite egress, invasion and development. Consequently, they are important drug targets for Plasmodium falciparum and P. vivax. WM4 and WM382 are potent inhibitors of PMIX and PMX that block invasion of liver and blood stages and transmission to mosquitoes. WM4 specically inhibits PMX whilst WM382 is a dual inhibitor of PMIX and PMX. To understand the function of PMIX and PMX proteases we identied new protein substrates in P. falciparum and together with detailed kinetic analyses and structural analyses identied key molecular interactions in the active site responsible for the specicity of WM4 and WM382 inhibition. The crystal structures of PMX apo enzyme and the protease/drug complexes of PMX/WM382 and PMX/WM4 for P. falciparum and P. vivax have been solved. We show PMIX and PMX have similar substrate selectivity, however, there are distinct differences for both peptide and full-length protein substrates through differences in localised 3-dimensional structures for the enzyme substrate-binding cleft and substrate interface. The differences in anities of WM4 and WM382 binding for PMIX and PMX map to variations in surface interactions with each protease in the S' region of the active sites. Crystal structures of PMX reveal interactions and mechanistic detail on the selectivity of drug binding which will be important for further development of clinical candidates against these important molecular targets. PfPMX PMIX protease P. falciparum parasites 3 . WM382 PfPMX, PvPMX PfPMIX. These results also suggest that the high potency of PfPMX, PvPMX and PfPMIX was driven to a great extent by the very slow off rates for linearized µg transfected simultaneously into selected maintained on 2.5 nM


Introduction
Plasmepsin IX (PMIX) and X (PMX) are highly conserved aspartic proteases expressed by protozoan parasites of Plasmodium spp. and they play a pivotal role in processing of proteins involved in essential steps in its life cycle and consequently they are important drug targets 1,2,3 . Six species of Plasmodium spp. infect humans and Plasmodium falciparum is responsible for the most severe form of malaria and P. vivax causes the most widespread disease 4 . These parasites infect humans through the bite of an infected female Anopheles mosquito and the injected sporozoites rapidly migrate to the liver and infect hepatocytes in which they develop to thousands of liver merozoites. The liver merozoites are released WM4 and WM382 are small molecule compounds that inhibit the function of PMIX and PMX in P. falciparum, P. knowlesi and P. berghei and the conservation of these proteases suggest they are potent inhibitors across Plasmodium spp. 3

(insert Ruiz et al. ref for Med Chem). Oral administration of WM382
cures mice infected with P. berghei or P. falciparum, the latter using a humanised mouse model 3 . It also blocks transmission of both P. falciparum and P. berghei to mosquitoes and prevents blood infection from the liver stage. WM382 is a potent dual inhibitor of both PMX and PMIX and consequently, a promising lead candidate for development of new antimalarials with a novel mode of action for treatment and prevention of malaria 3 .
Using recombinantly-expressed P. falciparum PMX, both WM382 and WM4 have been shown to be potent inhibitors of protease activity with a K i of 0.014 and 0.45 nM respectively 3 . However, little is known with respect to the functional properties of PMIX. Additionally, there is little understanding of the molecular mechanism of binding of the dual inhibitor WM382 with PMX and PMIX or the binding of WM4 to PMX. In this work we have expressed and puri ed recombinant PMX and PMIX and determined 3-dimensional structures in complex with WM382 and WM4 by X-ray crystallography and homology modelling. This has allowed us to gain an understanding of the function of PMX and PMIX and unravel the molecular events involved in the inhibition of these proteases by WM382 and WM4.

Results
Identi cation of PfPMIX protein substrates in P. falciparum. WM4 and WM382 are inhibitors of P. falciparum growth because of potent inhibition of the proteolytic function of the aspartic proteases PMIX and PMX 3 . WM4 is a selective inhibitor of PMX whilst WM382 blocks both PMIX and PMX function in P.
falciparum. Protein substrates have been identi ed that are proteolytically processed by PfPMX, however, only a few have been identi ed that are speci cally cleaved by PMIX 1,2,3 . To discover potential protein substrates of PfPMIX in P. falciparum asexual stage proteins known to be processed that have potential PMIX and X substrate recognition sequences were investigated. This identi ed the P. falciparum rhoptry neck protein 6 (PfRON6) 11 , rhoptry-associated membrane antigen (PfRAMA) 12,13 and rhoptry-associated leucine zipper-like protein (PfRALP1) 14 as potential PMIX substrates.
To test if PMIX was involved in the proteolytic processing of PfRON6, PfRAMA and PfRALP1, P. falciparum parasites that expressed haemagglutinin (HA) epitope-tagged forms of each protein were generated. The PMX-speci c inhibitor WM4 and the PMIX/PMX dual inhibitor WM382 were used to determine if they inhibited processing of PfRON6, PfRAMA or PfRALP1 (Fig. 1). Processing of SERA5, which is mediated by SUB1, a protease activated by PMX, was used as a control (Fig. 1A) 1, 2, 3 . As expected, both WM4 and WM382 inhibit processing of SERA5 3 . In contrast, processing of PfRON6 from p114 to p106 was inhibited by WM382 but not WM4 (Fig. 1B) suggesting PMIX and not PMX was involved. Similarly, processing of PfRAMA (Fig. 1C) and PfRALP1 (Fig. 1D) were more e ciently inhibited by WM382 compared to that observed for WM4, again indicating that PMIX was required. These results are consistent with proteolytic processing of PfRON6, PfRAMA and PfRALP1 being mediated by PMIX or that this aspartic protease activates other proteases required for these cleavage events.
Substrate speci city of PMIX and PMX. To understand the substrate speci city of these aspartic proteases in Plasmodium spp. active recombinant P. falciparum PMX (PfPMX) 3 , P. falciparum PMIX (PfPMIX) and P. vivax PMX (PvPMX) were expressed and puri ed to enable detailed kinetic and structural analyses (Fig. S1). PfPMX expression and puri cation of the active mature enzyme domain has been described previously and this protease has been shown to be autocatalytic 3 . The recombinant expression products expressing the full-length zymogen form of PvPMX and PfPMIX were the active mature enzyme domains consistent with autocatalytic activity (Fig. S1) 3 .
To show puri ed PfPMIX, PfPMX and PvPMX proteases were enzymatically active and to explore their substrate speci city, a uorogenic assay 15 was employed to investigate their ability to cleave peptide substrates designed from known PMX and PMIX processed sites 1, 2, 3 , as well as putative cleavage sites we identi ed for PfRON6, PfRAMA and PfRALP1 (Fig. 1E-G). PfPMX was able to cleave the peptide substrates to differing extents, as has been shown previously 3 , and it was also able to cleave putative PMIX recognition peptides identi ed in PfRON6, PfRAMA and to some extent PfRALP1 (Fig. 1E). Similarly, PvPMX showed a near identical cleavage pattern as PfPMX in cleaving this broad range of peptide substrates indicating this recombinant protein was active and that these proteases had very similar substrate speci cities (Fig. 1F).
In contrast, PfPMIX showed e cient cleavage of peptide substrates PfRh2N, PfRON3, PvRON3 and PfRAMA only and was not able to cleave those corresponding to other known PMIX or PMX processed proteins (Fig. 1G). This suggested PfPMIX either has a more restricted substrate speci city compared to PfPMX and PvPMX, possibly due to subtle differences in the structure of its substrate binding cleft or was less e cient in its ability to cleave peptide substrates compared to full-length proteins, which may have additional localised structure to assist binding to the enzyme. PfRON3 in P. falciparum parasites is processed by PfPMIX 3 and the ability of this enzyme to cleave the PfRAMA and PvRAMA peptide sequence corresponding to the proposed processing site was consistent with these proteins being direct protein substrates of PMIX in the parasite (Fig. 1G). Cleavage of the PfPMX peptide substrate PfRh2N by PfPMIX suggests that this protease can process PMX protein substrates when they can be accessed by the active site. PfPMIX was not able to e ciently cleave the PfRON6 or PfRALP1 peptides or other peptides from known PMIX protein substrates such as PfASP and PfRAP1 (Fig. 1G) 1,2,3 . This suggests that either PfRON6 and PfRALP1 are processed by another protease that is activated by PMIX processing 1, 2, 3 , or it was much less e cient at cleaving peptide substrates compared to full-length protein substrates.
To determine if PfPMIX could process full-length protein substrates, we expressed and puri ed P.
falciparum Apical Sushi Protein (PfASP) 16 and PfRh5-interacting protein (PfRipr) 17 as recombinant proteins (Fig. 1H, I, J). PfASP is 83 kDa but migrates aberrantly in SDS/PAGE, likely because of the number of charged amino acids, and it has been shown to be a substrate of the protease PfPMIX in P. falciparum ( Fig. 1I) 1,2,3 . PfASP is processed by PfPMIX at a known site resulting in a C-terminal p43 proteolytic fragment (43 kDa) (Fig. 1H) 3 with the fate of the N-terminal p40 fragment remaining unknown 3 . Although PfASP is a substrate for PfPMIX in P. falciparum 1, 2, 3 , a peptide containing the cleavage motif that produces the p43 and p40 fragments was not processed by recombinant PfPMIX and ine ciently by PfPMX and PvPMX ( Fig. 1E-G). Digestion of recombinant full-length PfASP using recombinant PfPMIX resulted in the production of the p43 C-terminal polypeptide fragment, as seen in the parasite 3 , but unexpectedly the p40 fragment was not seen, instead a smaller fragment (p32) was observed in the SDS-PAGE band pro le of the digest indicating the existence of a possible second cleavage site (Fig. 1I). Inspection of the sequence con rmed a second cleavage site that would result in a 32 kDa fragment (p32) and a much smaller 8 kDa fragment (p8). Incubation of PvPMX with full length PfASP resulted in the formation of the p43 and p40 fragments (Fig. 1J). However, the digestion involved cleavage at a single motif and did not progress further to produce the p32 fragment, as seen in the PfPMIX digestion of PfASP (Fig. 1I). This indicates the production of the p32 fragment in PfASP was the result of a two-step cleavage event mediated by PfPMIX, which occurs via the p40 intermediary fragment in PfASP. Although PvPMX can promiscuously cleave the rst motif it did not cleave the second motif due to differences in substrate speci city for the substrate which can be imposed by substrate structure or a subtle difference in the structure of the enzyme binding cleft.
Following removal of its signal sequence, PfRipr is predicted to be a 123 kDa protein which is a substrate of PfPMX in P. falciparum 3 . PfRipr is processed by PfPMX at a single site to convert the 123 kDa protein into two polypeptides (p64 and 59 kDa) (Fig. 1H) 3 . Recombinant PfPMIX was unable to process the PfRipr cleavage motif in either a peptide (Fig. 1G) or in the full-length protein (p123) as there was no observable difference between the protein band pro les in the control lanes with the PfRipr only control and the PfRipr/PfPMIX digest (Fig. 1J). Recombinant PfPMX and PvPMX were not effective at processing a peptide substrate with the cleavage motif for PfRipr (Fig. 1E, F). However, recombinant PvPMX was able to cleave full-length PfRipr protein (p123) into two polypeptides corresponding to p64 and p59 in P. falciparum parasites 3 (Fig. 1J). These results indicate that the PfRipr cleavage motif interaction with PvPMX requires additional structural interactions provided at the enzyme/protein substrate interface to enable e cient processing. These digests also validate PfRipr as a substrate of PMX but not PMIX and importantly reveal evidence that different and unique speci cities can exist between these two enzymes due to protein substrate structure.
PfPMIX has subtle differences in substrate speci city for cleavage of the PfRON3 peptide. To further understand the substrate speci city of PfPMIX, PfPMX and PvPMX the ability of each protease to cleave the PfRON3 peptide sequence was tested and compared with peptides in which speci c amino acids were mutated to alanine ( Fig. 2A). Firstly, the cleavage site within the PfRON3 (and PfRh2N) peptides following digestion with PMIX and PvPMX were identi ed by mass spectrometric analysis (Fig. S2). The phenylalanine of each peptide was shown to be the P1 position with cleavage by PvPMX and PfPMIX between it and the Leucine/Isoleucine residue (P1'). This was the same cleavage position as was determined for PfPMX 3 . Both PfPMX and PvPMX gave very similar patterns of cleavage for the wild type and mutant RON3 peptides consistent with them having very similar substrate speci cities ( Fig. 2A). Surprisingly, the P1 phenylalanine was not essential as both PfPMX and PvPMX showed relatively e cient cleavage (RON3 m3 mutant). In contrast, the P1' and P3' isoleucine and the P4' aspartate residues were important as the ability of these proteases to process the corresponding mutant peptides was reduced (RON3 m4, m6 and m7b mutants). This data was consistent with both PfPMX and PvPMX having essentially identical substrate speci cities ( Fig. 1E and F).
In contrast, PfPMIX showed distinct differences with respect to its ability to cleave the RON3 mutant peptides compared to PfPMX and PvPMX suggesting subtle variance in substrate speci city ( Fig. 2A). Similar to the results observed with PfPMX and PvPMX, the P1 phenylalanine residue was not essential as PfPMIX was still able to cleave the peptide, albeit weakly (m3 mutant). However, the P1' isoleucine, P2' glutamine, P3' isoleucine and P4' aspartate appear to be essential for e cient cleavage by PfPMIX (RON3 m4 -m7 mutants). This suggests that there is variation in substrate speci city for PfPMIX compared to the PMX proteases likely due to amino acid differences in the active site of the proteases (Fig. S3). Despite the contrasting ability to cleave the mutant RON3 peptides a sequence logo representation of PfPMIX cleavage sequences (Fig. 2B), compared to that for PfPMX 3 , showed close similarity across the core substrate recognition sequence.
WM4 is a selective inhibitor of PMX and WM382 is a dual inhibitor of both PMIX and PMX. Previously, we have shown that WM4 is a PMX speci c inhibitor whilst WM382 is a dual inhibitor of both PMIX and PMX protease function in P. falciparum 3 . Whilst WM4 and WM382 are potent inhibitors of P. falciparum growth, with an EC 50 of 7.4 nM and 0.39 nM respectively (Fig. 2C), the molecular basis for the differential inhibition of these compounds with respect to their inhibition of PMIX and PMX function is not known. To gain an understanding of the molecular interactions of WM4 and WM382 with PfPMX, PvPMX and PfPMIX we analysed the kinetics of inhibition (K i ) for peptide cleavage. PvPMX and PfPMIX e ciently cleaved the PfRON3 and PfRh2N peptides respectively at optima pH 5 to pH 5.5 (Fig. S4). WM4 has a similar K i for PfPMX (0.446 nM) and PvPMX (0.408 nM) with PfPMIX having a signi cantly higher K i (263 nM) with a >590-fold increase in the inhibitory effect against PfPMX and PvPMX over PfPMIX (Fig. 2C). This was consistent with WM4 speci cally inhibiting PMX proteolysis but not cleavage of PMIX protein substrates in P. falciparum 3 . WM382 also had a similar K i for inhibition of PfPMX (0.014 nM) and PvPMX (0.007 nM) as did WM4 consistent with them both being potent inhibitors of PMX enzyme activity. However, WM382 was also a potent inhibitor of PfPMIX (Ki = 0.4 nM), a 29-fold increase in the inhibitory effect against PfPMX over PfPMIX.
The equilibrium dissociation constant (K D ) for the a nity of WM4 binding to PvPMX (0.48 ± 0.08 nM) and PfPMX (0.39 ± 0.07 nM) was very similar as were the k on (PfPMX 6.3 ±2.3 x 10 6 M −1 s −1 , PvPMX 7.6 ± 5.4 x 10 6 M −1 s −1 ) and k off (PfPMX 2.4 ± 0.8 x 10 −3 s −1 , PvPMX 3.4 ± 2 x 10 −3 s −1 ) rates ( Fig. 2D and F, Table S1). This was consistent with the conservation of the active site of these proteases (Fig. S2) and showed that this compound bound strongly to these enzymes. In contrast, the K D of WM4 for PfPMIX (77 ± 13 nM) showed a 160-fold lower a nity of binding with the k on (6.0 ± 3.1 x 10 5 M −1 s −1 ) and k off (4.7 ± 3.0 x 10 −2 s −1 ) rates also signi cantly lower (Fig. 2H, Table S1). These results are consistent with WM4 being a speci c inhibitor of PfPMX but not PMIX protease function in P. falciparum parasites 3 . WM382 bound the three enzymes with very high a nity, with a K D of <0.1 nM. The dissociation curves were at for all proteases and it was not possible to determine k off rates and therefore a K D as the response did not decrease below 5% during the 25 min dissociation phase (Fig. 2E, G and I). This data was consistent with the potent enzyme inhibition observed and showed there were no measurable differences in the K D for WM382 between PfPMX, PvPMX or PfPMIX. These results also suggest that the high potency of WM382 for inhibition of PfPMX, PvPMX and PfPMIX was driven to a great extent by the very slow off rates for binding of the compound.
Structure of P. falciparum PMX. To understand the structural basis for the speci city for WM4 inhibition of PMX compared to the dual speci city of WM382 to inhibit both PMIX and PMX crystal screens were performed with the apo enzymes as well as in complex with WM4 and WM382 for PfPMX, PvPMX and PfPMIX. Diffracting crystals were obtained for the PfPMX apo protease as well as the protein/drug complexes PfPMX/WM382, PvPMX/WM382 and PvPMX/WM4.
A crystal structure of the apo protease PfPMX was obtained at a resolution of 1.85 Å ( Fig. 3A and B). PfPMX has a canonical aspartyl protease fold with a crescent shape and a predominantly b-sheet core ( Fig. 3B) 18,19 . The N-and C-terminal subdomains are anchored via a six stranded interdomain b-sheet. The uppermost b-strand (PfPMX F237-P240) of this sheet arises from the cleavage and removal of the prodomain and the subsequent relocation of the N-terminus to enable association with the enzyme domain 20 . Subsequently, a mobile loop is formed in the processed N-terminal (PfPMX F237-L252) region and anchored to the mature enzyme structure by incorporation into the b-sheet (Fig. 3B). This loop forms the edge of a sub-pocket within the S3 substrate binding site and its importance in inhibitor binding is discussed below. A single hairpin loop is centrally located on the outside of the substrate binding cleft (Fig. 3. A and B). The loop between residue Asn345-Asp352 had poor electron density and could not be structurally determined in PfPMX. These missing residues are located at the edge of the active site cleft and thus presumably not involved in ligand binding.
Aligning the apo PfPMX enzyme structure with that of another member of the plasmepsin family, plasmepsin V (PMV) in complex with a peptidomimetic inhibitor WEHI-842 (not shown in schematic for simplicity), shows that the PfPMX loop (residues Asn345-Asp352) has a closed conformation without requiring the engagement of a substrate or inhibitor (Fig. 3C) 19 . PfPMX consists of the core aspartic protease structure and lacks other domains present in PvPMV such as the two b-barrels present at the bottom of the structure and the 'nepenthesin 1-type' aspartyl protease (NAP1) fold (Fig. 3C). These sequences in PvPMV are likely involved in protein-protein interactions and access to the protease active site and the lack of these shows that PfPMX does not require additional domains to perform the basic function of proteolysis for its target protein substrates.
A modeled structure for PfPMIX was generated using the PfPMX structure and comparison has revealed that the key structural and functional residues in the active site are located in similar positions in the PfPMIX model ( Fig. 3. D). PfPMIX has a 59 residue insert (I431-N490) that is not shown and as this feature and structure is unique to this protease it could not be accurately modeled (Fig. 3D, S2). This insert occurs within four residues on the N-terminal side of the active site amino acid residue D457 and its in uence on substrate binding has not been determined to date. Although the modeled structure for PfPMIX shows similarity to that found for PfPMX, its electrostatic surface was predicted to have a charge reversal around the location of its S3 pocket, which could generate subtle changes in the substrate speci cities between each of these enzymes (Fig. S5).
PMX/inhibitor structure and interactions that mediate inhibitor binding a nity and speci city. Structures were obtained for PfPMX in complex with WM382 (resolution 2.76 Å) as well as PvPMX in complex with WM382 (resolution 2.22 Å) and WM4 (resolution 3.35 Å) ( Fig. 4 and Table S2). The structure of PvPMX had similarity to that obtained for PfPMX including the basic architecture of the active sites and binding of WM382, although there are speci c regions in the catalytic cleft that show localised structural movements due to drug binding (discussed below) (Fig. 4). Both WM382 and WM4 inhibitors align with the active site residues and are positioned between the inner and outer (S2 ap) surfaces of the substrate binding pocket (Fig. 4A, D and G). The structures for the protease-inhibitor complexes for PfPMX/WM382, PvPMX/WM382 and PvPMX/WM4 showed that the models for the inhibitors matched the observed electron density found within the substrate binding pocket as shown in the OMIT diagrams (2Fo-Fc density contoured at 1.0 σ) (Fig. 4B, E and H).
WM4 and WM382 share a common template but they have different moieties projecting from them designed to promote enhanced interactions within the substrate binding pocket. These subtle structural variations have led to the observed differences in binding a nity and substrate speci city for each of these compounds with PfPMX and PvPMX (Fig. 4C, F, I) and subsequently PfPMIX. Hydrogen bonds are the strongest surface interactions occurring between the active site of the corresponding protease and the inhibitors within the structures of these complexes. Within the PvPMX-WM4 complex (Fig. 4I), hydrogen bonding occurs in the S1 pocket between the hydrogen atoms of an amine and imine group from the 'warhead' on WM4 and the active site aspartic acid residues in PvPMX (D231 and D421) (Fig. 4I). An additional hydrogen bond also occurs between the main chain carbonyl group of G423 and the amide hydrogen of the carbonyl amide moiety on WM4. Hence, hydrogen bonds tether WM4 to the wall of the S1 pocket at the back of the substrate binding cleft.
The hydrogen bonding pattern observed between WM382 with PfPMX and PvPMX was identical in both structures ( Fig. 4C and F). The pattern contains the same hydrogen bonds observed for the PvPMX/WM4 structure ( Fig. 4I) but has an additional bond occurring between the carbonyl of the carbonyl amide moiety of WM382 and the hydroxy hydrogen of a serine located in the S2 ap at the front of the substrate binding cleft (PfPMX S313 and PvPMX S278). This additional hydrogen bond enables tethering of WM382 to the front S2 pocket in addition to the back surface of the S1 pocket in the substrate binding cleft in PfPMX and PvPMX ( Fig. 4C and F). This same hydrogen bond does not seem to form with WM4 as the positioning of the phenyl moiety in the S'1 and S'2 pockets is such that it results in a more downward placement of the central aromatic ring, shared in both compound structures, than observed in WM382. This slight change in relative placement (1.7 Å) has a ow on effect, with the carbonyl of the adjacent carbonyl amide moiety now being too far away from S278 to enable the formation of a hydrogen bond similar to one observed in WM382. Interestingly, the modeled structure for PfPMIX in complex with WM382 has a Ser to Thr amino acid substitution in the same position in the ap of the S2 pocket suggesting this additional hydrogen bond can occur in the PfPMIX/WM382 complex (Fig. S6).
Amino acid residues positioned within 4 Å of WM4 and WM382 in the PvPMX/inhibitor structures were visualized by mapping onto a surface representation of the structures tabled with their location in the substrate binding cleft and color-coded as common to both inhibitors and unique to a speci c inhibitor (Fig. 5). The common interactions (orange) occur about the active site aspartic acid residues and the compound warhead in the S1 and S2 pockets (Fig. 5). The parent WM4 inhibitor (cyan) has speci c interactions (green) which occur predominantly in the S1'/S2' pockets via a unique phenyl group protruding into this area ( Fig. 5B and D). The binding a nity (K D ), as determined by SPR, for binding of WM4 to the PvPMX active site was 0.48 nM (Fig. 2D). Interestingly, the WM382-speci c interactions (magenta) are located towards the opposite end of the substrate binding cleft, in the S2, S3 and S3 sub pocket areas ( Fig. 5A and C). This inhibitor has a methylated chromane moiety that enables greater extension and interaction with residues in this region compared to the a-methyl benzyl moiety located in the same position in WM4. Furthermore, WM382 was positioned such that an additional hydrogen bond occurred between Ser278 (S2 pocket) and the carbonyl of the carbonyl amide moiety, which engages the ap and tethers WM382 between the front and back surfaces of the catalytic cleft and explains the stronger binding a nity of WM382 compared to WM4. This was re ected in the K D values determined by SPR for binding of WM382 to PvPMX and PfPMX (K D <0.1 nM).
Although WM4 and WM382 are both capable of inhibiting PfPMX function in P. falciparum only WM382 shows a dual speci city that includes potent inhibition of PfPMIX 3 . Based on the results above, these differences in speci city are due to differences in the structures for the substrate binding cleft for PfPMIX and PfPMX or PvPMX. Given that the K D for WM4 interacting with PfPMIX was 77 nM, 160-fold weaker than that for PfPMX (0.39 nM) and PvPMX (0.48 nM), this indicated that WM4 has decreased interaction with the surface in the S1'/S2' pockets of PMIX and is the basis of the unique PMX speci city of this inhibitor compound. Interestingly, a large 59 amino acid loop, with unknown structure, is inserted between the two sub domains of the mature PfPMIX enzyme (Fig. S3). This loop attaches, at the C-terminal end, within six residues of the active site D495 and may destabilize or be responsible for a change in surface topography in the S1'/S2'area.
The binding a nity of WM382 for PfPMIX was the same as that obtained for PvPMX and PfPMX (ie. K D <0.1 nM). In the PMX structures, WM382 interacts with the S1-S3 pockets and was tethered via four hydrogen bonds to the front and back of the substrate binding cleft. In a modeled structure for PfPMIX (based on the structure of PvPMX/WM382) the same hydrogen bonding pattern can be predicted with a T293 replacement for the PvPMX S278/PfPMX S313 in the ap at the front of the substrate binding pocket. The model predicts a similar distribution of weaker dispersion forces about the S1 to S3 pockets. Collectively this supports a similarity in this region of the substrate binding surface in PMX and PMIX and the basis for the dual speci city observed for WM382 inhibition of these two proteases.
The localized structural response of plasmepsin X to enable inhibitor binding. For the PfPMX and PvPMX structure the S2 ap and the loop of the S3SP showed signi cant changes in conformation for inhibitor binding (Fig. 6). In the PfPMX apo structure the S2 ap, particularly between residues F311 to G314, was found to be twisted almost perpendicular to the substrate binding cleft resulting in F311 being orientated into and S313 (not shown) orientated away from the empty cleft. Meanwhile the S3SP loop (H242 to F248) has a width from the Q247Ca to the L243Ca of 6.6 Å and residues within this loop have an outward projection (Fig. 6A). Entry of the inhibitor WM382 into the substrate binding cleft of PMX appears to have occurred from the S' side of the cleft because of the unilateral change in orientation in the most affected residues. As a result, the amino acid F311 has moved 4.2 Å from where it was positioned in the apo structure and instead of projecting inward toward the cavity space of the cleft, the side chain has been reorientated toward the S2 ap to provide su cient spatial separation from inhibitor WM382. Amino acid residue S313 was also repositioned via approximately a 90˚ rotation and a 4.8 Å directional shift in the hydroxy group of this residue. This structural change subsequently enables a hydrogen bond to occur between inhibitor WM382 and the hydroxy moiety of S313 as discussed previously (Fig. 4C).
The positioning of the WM382 warhead for interaction with the active site aspartic acid residues of PfPMX also requires the S3SP loop to be displaced further out of the cleft as shown by the 3 Å shift to the right and the width of the loop (Q247Ca to L243Ca) was compressed by 1.2 Å (Fig. 6A). Otherwise, the methylated chroman moiety of WM382 would collide with residues forming the S3SP loop. Furthermore, the outward projecting residues of the S3SP loop observed in the PfPMX apo structure are found to be orientated more upward and toward the inside of the loop in the PfPMX/WM382 structure. This reorientation of the S3SP residues enables the loop to be stabilized by an increased number of hydrogen bonds ( Fig. 6 and Table S3).
As discussed previously, the S3SP loop is believed to be obtained from the rearrangement of the unstructured residual pro-sequence remaining after the autolytic cleavage events required to activate PMX, via a similar process described for other aspartyl proteases (autolytic activation) 20 . In the PfPMX apo structure the S3SP loop was quite mobile and has limited rigidity due to the low number of observed hydrogen bonds between residues within the loop (Fig. 6D and Table S3). In comparison, the structure obtained for PfPMX/WM382 complex has eight hydrogen bonds within the loop structure. This has arisen due to inhibitor induced compression of the loop and the inward reorientation of loop residues (Fig. 6E). The hydrogen bond pattern is three dimensional and not only increases the rigidity within the loop but also tethers the loop to a region outside via a hydrogen bond between Q247 and D245 ( Fig. 6E and Table S3).
The S3SP loop structure for the PvPMX/WM382 complex has the same number of hydrogen bonds as the S3SP loop in PfPMX/WM382 complex but the hydrogen pattern was slightly different due to the in uence of D323 (PvPMX) rather than S359 (PfPMX) as the external tethering point for the S3SP loop ( Fig. 6F and Table S3). The loop structure in the PvPMX/WM4 complex has six hydrogen bonds and incorporates both internal and external hydrogen bond tethering points for increased loop rigidity. The amethyl benzyl moiety of WM4 does not protrude as far into the S3SP as the methylated chromane moiety of WM382 and likely in uences positioning of the D210 residue in the structure of the PvPMX/WM4, as two hydrogen bonds involving Q212 and D210 are not observed. Hence, the S3SP loop becomes stabilized by inhibitor displacement and loop compression with reorientation of loop residues that favor hydrogen bond formation leading to increased loop stability. This region was not obvious in the structure of the apo form of PfPMX. The S3SP is consequently, an important region for increasing a nity of binding for members of this inhibitor family to PMIX and PMX via hydrogen bond formation and increased Van der Waals interactions.

Discussion
Plasmepsin IX and X in Plasmodium spp. are essential aspartic proteases that proteolytically process many different proteins required for parasite egress, invasion and development of liver, blood and transmission stage parasites and consequently are important drug targets for development of novel antimalarials 1, 2, 3 . WM4 and WM382 are lead compounds for development of antimalarials with a mode of action that can target either PMIX and X together or more speci cally PMX alone 3 . The mechanism by which these compounds can differentially inhibit the function of these two aspartic proteases has been unknown. In this study we have obtained the crystal structures of PMX as an apo enzyme and in complex with WM4 and WM382 as well as performed detailed enzyme kinetic and binding analyses to identify the different molecular interactions of these compounds in the active site that explain the speci city of drug inhibition for the different proteases.
The proteins PfRON6, PfRAMA and PfRALP1 have been identi ed as substrates of PfPMIX in P. falciparum parasites increasing those known to be processed by this protease. The ability of WM382 (PMIX and PMX dual inhibitor) to speci cally inhibit processing of these proteins suggested they are direct protein substrates of PfPMIX. However, it was possible this aspartic protease activated another intermediate protease that was directly responsible for processing of nal substrates, as is the case with respect to the activation of SUB1 1, 2 and SUB2 3 . The identi cation of PMIX amino acid recognition sites in those proteins that once cleaved correspond to the observed molecular weight of the proteolytic processed polypeptides for each protein in the parasite, and the ability of these peptides to be cleaved by either PfPMIX, PfPMX or PvPMX supports them as true PfPMIX protein substrates. Conditional knockout of the gene encoding PMIX in P. falciparum blood stages has shown that the function of this protease was essential 1, 2 indicating that proteolytic processing of at least some of these protein substrates is required for their function. In P. falciparum the PfPMIX processed protein PfRAMA is essential for correct tra cking of a subset of rhoptry proteins and a conditional gene knockout results in dysmorphic rhoptries lacking a neck region and merozoites that are unable to invade erythrocytes 12 .
Our results show that PMIX and PMX have similar peptide substrate selectivity, however, there are clear differences in their ability to e ciently process different full-length protein and peptide substrates.
Recombinant PfPMIX can process full-length PfASP but not the peptide corresponding to the cleavage site suggesting that additional interactive structural elements are required for recognition and effective cleavage of these motifs, which are provided by the protein. PMIX was also not able to proteolytically process recombinant PfRipr, a known PfPMX substrate 3 . Interestingly PvPMX, in addition to cleaving fulllength PfRipr, could process the recombinant full-length PfASP but it did so less e ciently and incompletely compared to PfPMIX, with only one of the two cleavage motifs processed. It is likely the properties of the protein substrates and the surface structure around and in the active site of both PMIX and PMX are responsible for these subtle differences. These observations indicate that the distinct subcellular localisation of PMX in the micronemes and PMIX in the rhoptries 1, 2 is an important aspect for the apparent protein substrate selectivity of these aspartic proteases. This raises the question as to why Plasmodium spp. require two aspartic proteases expressed in merozoites with similar substrate selectivity?
The most likely answer is that since PMX and PMIX substrates are localised to both the micronemes and rhoptries, two aspartic proteases are required to be independently tra cked to these organelles to process those substrates.
WM4 and WM382 are a promising novel class of compounds being developed as antimalarials and the fact that WM382 has dual activity against both PMIX and PMX is an invaluable property, as it results in a high bar for the development of resistance by P. falciparum 3 . Additionally, the development of WM4 that is speci c for PMX has provided a useful chemical tool to interrogate the role of both PMIX and PMX in parasite biology. Analysis of the structures obtained for PMX apo and PMX/inhibitor complexes have shown how a PMX speci c (WM4) and a dual PMIX and PMX (WM382) inhibitor are positioned within the enzyme substrate binding pockets and the changes required in this localised environment to accommodate inhibitor binding. WM4 speci c inhibition of PMX is due to unique interactions occurring in the S1' and S2' substrate binding pockets. The decreased ability of WM4 to block PMIX activity indicates this protease is structurally different in the S' region of its substrate binding pocket. Comparatively, the structures for the PfPMX/WM382 and PvPMX/WM382 complexes reveal that this inhibitor does not extend signi cantly into the S' region, as per WM4, but has more surface interactions in the S2' and S3' pockets in addition to those that are common to both inhibitors.
The design of WM382 enables the formation of an additional hydrogen bond, which occurs between a serine in the S2 ap, at the front of the catalytic cleft, and carbonyl of a carbonyl amide group in WM382.
This engagement of the S2 ap results in tethering of WM382 between the inner and outer surfaces of the substrate binding cleft. As a result, a much stronger interaction occurs between PMX and WM382 which is re ected in the PMX/WM382 binding a nity in comparison to that obtained for PMX/WM4. A similar binding a nity was observed for the PMIX and WM382 interaction and implies this inhibitor is binding to a region of the substrate binding surface of PMIX similar in structure to the corresponding region in PMX.

Analysis of the structures obtained for PMX apo and PMX/inhibitor complexes have shown how a PMX
speci c (WM4) and a dual PMIX and PMX (WM382) inhibitor are positioned within the enzyme substrate binding pockets and the changes required in this localised environment to accommodate inhibitor binding. In the rst instance, the PMX speci c inhibition by WM4 occurred because of unique interactions occurring in the S'1 and S'2 substrate binding pockets. The decreased ability of WM4 to block PMIX activity indicates this protease is structurally different in the S' region of its substrate binding pocket.
Comparatively, the structures for the PfPMX/WM382 and PvPMX/WM382 complexes reveal that this inhibitor does not extend signi cantly into the S' region, as per WM4, but has more surface interactions in the S2 and S3 pockets in addition to those that are common to both inhibitors. Additionally, PMX makes similar localised changes to its conformation and stability to accommodate both WM4 and WM382. The S2 ap was pushed aside and a mobile loop, which has been reorientated after autolytic activation of PMX, becomes compressed and displaced. This enables increased hydrogen bonding within and external to the loop resulting in increased stability and formation of a new surface not present in the apo structure of PMX. This surface could be targeted/engaged by informed moiety design on the core structure of this family of inhibitors and used to increase the a nity of new generation of dual inhibitors binding to this region of the substrate binding pocket. Protease processing inhibition assays in P. falciparum parasites. Processing inhibition assays were performed essentially as previously described 3 . Protease inhibitors were added to synchronised late trophozoite/early schizont cultures, then passed over LD or LS magnetic columns (Miltenyi Biotech) to remove uninfected erythrocytes. WM4 and WM382 were added at 40 nM and 2.5 nM nal concentrations respectively. A control dish without protease inhibitors was included. Parasites were eluted from columns with complete RPMI 1640 culture medium with inhibitor at the same concentration. Eluted parasites were adjusted to 5x10 6 schizonts/ml and 150 ml added per well of a 96-well at-bottomed culture dish. The parasites were cultured for 16 hr and a representative well smeared for Giemsa staining, to ensure that rupture had occurred normally (control well) or that rupture had been blocked (WM4, WM382 conditions).

Methods
Parasites were centrifuged at 10000g / 10 min to collect merozoite and supernatant fractions. Proteins from both fractions were added to Reducing sample buffer and separated on 4-12% or 3-8% acrylamide gels (NuPAGE, Invitrogen) for subsequent immunoblots.
Expression, puri cation and crystallisation of apo PfPMX, PvPMX and PfPMIX. The apo form of PfPMX was expressed as previously described 3 . In short, PfPMX was expressed for 72 hr at 28°C in Spodoptera frugiperda (Sf21) cells transfected with Baculovirus. The PfPMX protein was harvested by centrifugation at 4°C, 12,000 x g for 15 min and puri ed by nickel a nity chromatography using a ow gravity column and eluted in 15 mM imidazole increments. PfPMX was subsequently dialysed against 4 L 20 mM Tris pH 7.5, 150 mM NaCl, and concentrated using a 30 kDa concentrator. The PfPMX autocatalysis was promoted by diluting PfPMX into 50 mM sodium acetate pH 4.0, 150 mM NaCl for 20 min. at 37°C. The autocatalysis was concluded by the addition of 1 M Tris pH 8 until pH 7.5-8 was reached. Lastly, the processed PfPMX was further puri ed by size exclusion in 20 mM Tris pH 7.5, 150 mM NaCl using a HiLoad® Superdex 16/600 200 pg (Cytiva) column on an Akta Pure chromatography system.
To express recombinant PfPMIX the gene sequence (212KCD…NNL647) was recodoned for insect cell expression and synthesised by GenScript and subcloned into the p1TF vector that enabled expression in insect cells. For expression in mammalian cells the PfPMIX gene was was synthesised by GenScript and subcloned into a pCDNA 3.4 expression vector containing a N-terminal IL-2 signal sequence followed by a consensus Kozak sequence. PfPMIX was expressed in CHO cells as described 22 . In summary, sterile 1 mg PfPMIX DNA per mL expression culture was transiently transfected into CHO cells using the ExpiCHO Expression System (Thermo Fischer Scienti c) and expressed at 37°C and rotated at 120 RPM for 8 days.
The suspension culture was harvested by centrifugation at 4000 x g for 30 min at 4°C and then dialysed against 20 mM Tris pH 7.5, 300 mM NaCl overnight at 4°C. PfPMIX was subsequently puri ed by gravity ow nickel a nity chromatography and eluted by 15 mM imidazole increments. This was followed by ion exchange chromatography on a HiTrap® SP HP column (Cytiva) using 20 mM Tris pH 7.5, 300 mM NaCl and 20 mM Tris pH 7.5, 1.2 M NaCl as binding and elution buffers, respectively. The PfPMIX was ash frozen and stored at -80°C until further use.
PvPMX (residues H27-E545) was recodoned for insect cell expression (Bioneer), then cloned into an insect cell expression vector bearing an N-terminal gp67 signal peptide and a C-terminal fusion tag comprising a tobacco etch virus (TEV) protease-cleavage site and a FLAG tag. This was expressed in Sf21 insect cells. An N-terminally processed form of the expected recombinant zymogen was puri ed initially from cell supernatant using anti-FLAG M2-agarose (Sigma). Pooled fractions were concentrated, and further puri ed using Gel-ltration chromatography (Superdex 75, GE Life Sciences) in 20 mM Tris, pH 7.2, 100 mM NaCl, which resulted in pure and stable protein that was concentrated for crystallization and other biochemical and biophysical characterisations.
Crystallisation trials and protein structure analysis. Puri ed PfPMX was used for crystallisation trials alone and in complex with WM382. Apo crystals grew in 0.01 M zinc chloride, 0.1 M sodium acetate pH 5.0, 20% w/v PEG6000. These crystals were used as a seed stock and the crystal conditions repeated.
Extensive soaking and co-crystallisation trials of apo PfPMX with WM382 failed to produce protein crystals. Diffracting crystals of PfPMX complexed with WM382 were obtained from PfPMX expressed in the presence of WM382 in insect cells, from which 70 kDa protein band was detected by SDS-PAGE. This corresponds to the inactive unprocessed PfPMX. Puri cation by ion exchange chromatography resulted in highly pure uncleaved PfPMX/WM382 complex. Size exclusion chromatography of this material resulted in a mixture of unprocessed and processed PfPMX preparation. Interestingly, only this mixed PfPMX preparation resulted in protein crystals after 30 days and these diffracted to 2.76 Å. The structure of PfPMX complexed with WM382 structure was the same sequence length as the apo PfPMX. Our unpublished data suggest that PfPMX gets cleaved at neutral pH after a lengthy incubation time. Thus, it is possible as PfPMX+L-312 got cleaved, the optimal protein concentration for crystal nucleation was eventually reached.
The crystals were shot with the MX2 beamline at the Australian Synchrotron (Melbourne, Australia). The protein structures were determined by molecular replacement with the Autorickshaw server 23  Imidazole was added to 10mM nal concentration and PfRipr was puri ed by NiNTA Super ow resin (Qiagen) and eluted in 20 mM Tris pH 8, 150 mM NaCl, 500 mM Imidazole. The sample was further puri ed via size eclusion chromatography, using a HiLoad Superdex 200 16/600 column (Cytiva). Fractions containing PfRipr were pooled and concentrated to 1 mg/mL for protease assays.
The gene for P. falciparum (3D7) Apical sushi protein (corresponding to residues E22 to S710) was expressed as protein with a C-terminal fusion tag comprised of a TEV cleavage motif and ag peptide.
Apical sushi protein (ASP) was expressed in HEK exp293F and the crude supernatant puri ed initially using M2 anti-ag a nity resin. Fractions containing the desired expression product, as determined by SDS_PAGE analysis, were pooled and dialysed against 20 mM Bis-Tris pH 6.5/20 mM NaCl. The protein pool was then further puri ed using Strong Anion Exchange chromatography (SAX). ASP was eluted from a Hitrap column using a gradient from 0 to 50%B. Where Buffer A was 20 mM Bis-Tris pH 6. 5  Enzymes and peptide substrates were prepared in 25 mM sodium acetate (pH 5.5) and 0.005% Tween-20 for PfPMX and PvPMX assays and in the same buffer supplemented with 100 mM sodium chloride for PfPMIX assay. Reactions were allowed to proceed at 37 o C for 4 hr. Samples were excited at 340 nm and uorescence emission measured at 480 nm using a PHERAstarFSX plate reader (BMG Labtech). followed by 30 s regeneration with 50 mM glycine pH 9.5. Experiments were performed independently three times using a freshly immobilised sensor surface. PvPMX data were acquired on a BIAcore S200 instrument with data analysis in Biacore S200 evaluation software version 1.1 (Cytiva) and PfPMX data collected on a BIAcore 8K+ instrument with data analysis in Biacore Insight evaluation software version 3.0.12.15655 (Cytiva). All sensorgrams were double reference using an activated the ethanolamine blocked sensor surface and an HBS-EP+ buffer blank. A nities were determined using a 1:1 binding model tting on and off rates for WM4, it was not possible to report a nities for WM382 as the off rate was too slow to determine. A nities are reported as the mean and one standard deviation for the three independent experiments.

Data Availability
Data Availability The data that support the ndings of this study are available within the paper and its Supplementary Information. Figure 1 PMIX is responsible for processing P. falciparum proteins and can be inhibited by WM382 but not WM4.

Declarations
A. SERA5 processing was blocked by WM4 and WM382. Shown are supernatants containing proteins released from merozoites. SERA5 was detected using a speci c antibody as shown by the Ab symbol.
RON6 was tagged with HA epitopes at the C-terminus and detected using anti-HA antibodies. Shown are  Table S1. Experiments D-E were performed independently three times using a freshly immobilised sensor surface. A nities are reported as the mean and one standard deviation for the three independent experiments.  Surface interactions occurring at the interface of the inhibitor -substrate binding cleft. A and C.
PvPMV/WM382 and B. and D. PvPMX/WM4 highlight the distribution of common and speci c inhibitor interactions within the substrate binding cleft. Although interacting with both inhibitors, the active site aspartic acid residues are shown in red to assist with orientation. WM382 is represented as a green organic compound while WM4 is coloured cyan. Left-hand side view is from the front of the substrate binding cleft (Fig. 5A and B). The RHS view ( g. 5C and D) is from a 900 anti-clockwise rotation about the vertical axis to give a side view into the substrate binding cleft. E. Table comparing the VDW surface interactions (<4 Å) occurring between PvPMX and inhibitors WM382 and WM4. The substrate pocket (S) location of each residue is shown on the right-hand side. Residues have been colour coded as shown in the Table key. * or ** indicates a residue is involved in hydrogen bonding with an inhibitor(s). The location of each residue has been transposed onto surface representations of the structures for PvPMX /WM382 complex ( Fig. 5 A and C) and PvPMX/WM4 complex (Fig. 5 B and D). Refer to Table S3

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryFiguresandTables.docx