Target Validation and Identification of Novel Boronate Inhibitors of the Plasmodium falciparum Proteasome

The Plasmodium proteasome represents a potential antimalarial drug target for compounds with activity against multiple life cycle stages. We screened a library of human proteasome inhibitors (peptidyl boronic acids) and compared activities against purified P. falciparum and human 20S proteasomes. We chose four hits that potently inhibit parasite growth and show a range of selectivities for inhibition of the growth of P. falciparum compared with human cell lines. P. falciparum was selected for resistance in vitro to the clinically used proteasome inhibitor, bortezomib, and whole genome sequencing was applied to identify mutations in the proteasome β5 subunit. Active site profiling revealed inhibitor features that enable retention of potent activity against the bortezomib-resistant line. Substrate profiling reveals P. falciparum 20S proteasome active site preferences that will inform attempts to design more selective inhibitors. This work provides a starting point for the identification of antimalarial drug leads that selectively target the P. falciparum proteasome.


■ INTRODUCTION
Malaria remains a major health problem, threatening hundreds of millions of people and causing ∼440 000 deaths each year. 1 Current antimalarial control is highly dependent on artemisinin-based combination therapies (ACTs), which makes the emergence of artemisinin partial resistance extremely concerning. 2−4 Decreased ACT sensitivity delays the clearance of parasites from patients and leads to clinical failure, resulting in ∼50% treatment failure in regions where resistance is entrenched, compared with ∼2% failure in regions where resistance is rare. 5,6 Replacement antimalarials are therefore urgently needed.
The proteasome is a multisubunit enzyme complex that is responsible for proteostasis and for regulating key processes such as the cell cycle. It has a 20S catalytic core that includes two heptameric rings of β subunits. The β1 subunit provides caspase-like activity (cleaves after acidic residues), the β2 subunit trypsin-like activity (cleaves after basic residues), and the β5 subunit chymotrypsin-like activity (cleaves after nonpolar residues). 7 In cells exposed to oxidative stress or inflammatory cytokines, three of the active constitutive proteasome subunits are replaced by "immuno" subunits to form immunoproteasomes. 8 Proteasome inhibitors show potential for the treatment of malaria, exhibiting parasiticidal activity against asexual blood stages, including young (ring stage) intraerythrocytic parasites, as well as sexual stage gametocytes and liver stage parasites, 9−11 life stages that are resistant to most other chemotherapeutic agents. Moreover, inhibitors of the proteasome are active against both artemisinin-sensitive and -resistant parasites 12,13 and, indeed, strongly synergize artemisinin-mediated killing of P. falciparum in culture and P. berghei in vivo. 11,12 To date, efforts to identify proteasome inhibitors as potential antimalarial compounds have concentrated on inhibitors with epoxyketone and vinyl sulfone warheads that bind irreversibly to the proteasome active site. A recent report examined noncovalent asparagine ethylenedi-amine (AsnEDA) inhibitors and revealed good cellular selectivity, but this class of inhibitors is not stable in vivo (half-life of ∼30 min) and thus is not active (when used alone) in a mouse model of malaria. 11 If a potent, specific, and "druglike" proteasome inhibitor could be identified, it would be a Trophozoites were re-exposed to Shield-1 or exposed to indicated compounds for 3 h. Results are representative of four independent experiments. (i) Cell extracts were analyzed by Western blot using anti-GFP or anti-Pf BiP as a loading control. (ii) GFP fluorescence measured by flow cytometry. Dotted line represents background (fluorescence of sample after washout of Shield-1). Data are the mean ± range/2 for the readings in technical duplicate from one experiment. Curves are fitted with the sigmoidal four-parameter model. promising antimalarial compound in its own right and would be particularly effective in combination with artemisinins.
The reversible, covalent peptide boronate proteasome inhibitors bortezomib 14 and ixazomib 15 are used clinically to treat multiple myeloma. Ixazomib is administered orally and offers a favorable efficacy/safety profile with weekly dosing delivered as a fixed dose. 16,17 Bortezomib has been shown to have activity against P. falciparum, 12,18,19 although it has not been formally demonstrated that this results from inhibition of the P. falciparum proteasome. In this work, we used in vitro directed evolution to generate bortezomib-resistant parasites, thereby validating the β5 subunit of the proteasome as the target. We screened a boronate peptide library to identify inhibitors of the growth of cultures of P. falciparum and selected compounds for further characterization. We show that inhibition of β5 activity is needed for potent antiplasmodial activity but that compounds that also inhibit β2 activity remain active against bortezomib-resistant parasites. Substrate profiling points to substrate preferences that can underpin the development of inhibitors with high specificity against P. falciparum.

■ RESULTS AND DISCUSSION
Establishment of an Assay of P. falciparum 20S Proteasome Activity. Purified Pf 20S proteasome (see characterization in Figure S1A,B) was activated using human proteasome activator complex (PA28αβ), and the activity was assessed using fluorogenic AMC-based substrates. Initial testing of a range of substrates ( Figure S2A) suggested the use of Ac-nLPnLD-AMC for caspase-like activity (β1 subunit), Ac-WLR-AMC for trypsin-like activity (β2 subunit), and Ac-WLA-AMC for chymotrypsin-like activity (β5 subunit). The activity of β2 and β5 subunits was completely abrogated by 0.8 μM carfilzomib ( Figure S2B), consistent with a previous report. 9 By contrast, carfilzomib exhibited only weak inhibitory effect against Pf 20S proteasome β1 activity ( Figure S2B).
To assess the level of proteasome activity contributed by the estimated human 20S proteasome contaminant (1 in 240), we compared the activity of 1 nM Pf 20S with that of 0.005 nM human constitutive proteasome ( Figure S2C). Human β1c and β2c activities were not detected, and the fluorescence signal generated from β5c activity was only 2% of that from Pf 20S β5.
Analysis of the Activities of Selected Compounds against P. falciparum and Mammalian Cell Lines. Peptide boronates from the library of the Takeda Oncology Company (Cambridge, MA, USA), were screened for inhibition of the growth of cultures of P. falciparum (3D7). 20 Further characterization of the compounds for activity against Pf 20S and human 20S proteasome activity led to the selection of four compounds for further investigation, namely, malaria proteasome inhibitors (MPI-1, MPI-2, MPI-3, MPI-4; compounds 1−4), as well as bortezomib itself ( Figure 1A). The pinane esters of MPI-1 to -4 (compounds 7−10) were resynthesized and characterized in-house (see Scheme 1 and Experimental Methods). 21 Analysis of MPI-1, MPI-2, MPI-3, MPI-4 as inhibitors of the activity of purified Pf 20S and human 20S proteasome, in fluorogenic peptide assays, revealed that they exhibit good inhibition of both β2 and β5 activities of Pf 20S (MPI-3 and MPI-4) or more selective inhibition of β5 activity (MPI-1, MPI-2 and bortezomib) (as summarized in Table 1). MPI-3 and -4 are more active against P. falciparum β2 than human β2c or β2i. However, all four compounds were more active against the β1 and β5 activities of the human 20S isoforms (Table 1). While the levels of selectivity are less dramatic than in a recent study of asparagine ethylenediamine (AsnEDA) inhibitors, 11 those assays were performed in the presence of 0.5 μM WLW-vinyl sulfone, which acts synergistically with inhibitors of Pf 20S β5. 11,13 MPI-4 was chosen for a detailed kinetic analysis because it exhibited some selectivity for P. falciparum β2 compared with human constitutive β2 (β2c) ( Table 1A,B). It shows very rapid and tight (K i = 0.13 nM) binding to the human β5c but weaker affinity for P. falciparum β5 (K i = 4.3 nM) ( Figure S3B,D). Binding to the human β2c is very weak (K i = 487 nM), while binding to P. falciparum β2 is relatively stronger (K i = 6.4 nM) ( Figure S3A,C).
Peptide boronates are covalent, slowly reversible inhibitors, 22 with bortezomib exhibiting a t 1/2 value of ∼110 min for dissociation from the β5 site. 23 We found that MPI-4 exhibited a similar t 1/2 at the human β5c site (106 min, K i = 0.1 nM) but only 2 min (K i = 490 nM) at the β2c site ( Figure S3A), indicating it is only tightly bound at the human β5c site. In contrast, MPI-4 bound to both the β2 and β5 sites of Pf 20S with high affinity (∼5 nM) and extended t 1/2 (∼65 min). This indicates a key difference between the P. falciparum β2 and human β2c sites, potentially arising from the open con-

Journal of Medicinal Chemistry
Article formation of the P. falciparum β2-binding pocket compared to the human proteasome, 13 which might be exploited in the development of Pf 20S selective inhibitors.
To investigate the ability of the compounds to inhibit P. falciparum proteasome activity in cells, we made use of transfectants expressing GFP fused to a destabilization domain (DD), 24 which, when destabilized, is targeted for degradation in a proteasome-dependent manner. Addition of a protective ligand, Shield-1, stabilizes the fusion protein and prevents GFP-DD degradation, leading to an accumulation of full-length protein (as assessed by Western blotting) as well as an increase in the fluorescence signal from the GFP reporter ( Figure  1Bi,ii). Inhibition of the proteasome, using bortezomib, prevented GFP-DD degradation in the absence of Shield-1, as indicated by the accumulation of the full-length protein ( Figure 1Bi) and an increase in the fluorescence signal ( Figure  1Bii). Treatment with MPI-4 had a similar effect, while a higher concentration of MPI-1 was needed to produce the same level of inhibition of GFP-DD degradation ( Figure  1Bi,ii). These data are consistent with the novel inhibitors exerting their activity by disrupting proteasome-dependent degradation.
Mutations in a Kelch domain protein of unclear function (K13-propeller; PF3D7_1343700) are associated with decreased artemisinin sensitivity in vivo 25 and reduced sensitivity of very early ring stage parasites in a pulse assay format in vitro. 25 We examined the activity of the novel proteasome inhibitors in a 3 h pulsed exposure against a K13 mutant field strain from Cambodia (Cam3.II R539T ) and an isogenic line (Cam3.II rev ), in which the K13 wild-type genotype has been restored and sensitivity to dihydroartemisinin (DHA) regained. 26 All compounds exhibited similar activity against K13 wild-type and mutant parasites (Table 2B). MPI-1 and MPI-2 showed relatively weaker activity than MPI-3 and MPI-

Journal of Medicinal Chemistry
Article 4 in this 3 h pulse assay, suggesting a slower mode of action or a faster off-rate. We next examined the activities (72 h exposure) of the selected compounds against a colorectal carcinoma cell line, HCT116, that is particularly susceptible to proteasome inhibitors 27 and against the human hepatoma cell line, HepG2, that has been used frequently to assess selectivity in the screening of antimalarial compounds. 11,28,29 While bortezomib exhibited 2-fold higher potency against HepG2 compared with P. falciparum, MPI-1 was 56 times more effective against P. falciparum cultures compared to the HepG2 (Table 2A). This represents a 112-fold improvement in selectivity relative to bortezomib. These data indicate that compounds that exhibit weaker binding to the human β5 subunit (see Table 1) permit better tolerability while retaining reasonable activity against P. falciparum.
We have previously shown that the clinically used proteasome inhibitor, bortezomib, is a potent inhibitor of the growth of P. falciparum in culture and exhibits strong synergism with the clinically relevant artemisinin, DHA. 12 Similarly, both MPI-1 and MPI-4 exerted a pronounced synergistic interaction with DHA against the early ring stages of the K13 mutant isolate, Cam3.II R539T ( Figure 1C).
In Vitro Evolution of Bortezomib-Resistant Parasites and Validation of the Inhibitor Target. In an effort to confirm that the peptide boronates exert their antiplasmodial activity via direct binding to one or more proteasome subunits and to determine the residues that are important for the inhibitory activity, we selected P. falciparum for resistance to bortezomib, followed by whole genome sequencing. This approach has been used successfully to identify and characterize the targets of a number of antiparasitic compounds. 30,31 Multiple cloned lines of 3D7 were subjected to gradually increasing concentrations of bortezomib over a period of about 30 weeks. Two independent resistant lines (B1, B2) were obtained ( Figure S4) and cloned by limiting dilution. The selected clones (B1a, B2a) exhibited an approximately 20-fold increase in the 50% inhibitory concentration (Table 3A). The resistance phenotype was stable to freezing and thawing of the parasite clones. Whole genome sequencing was performed on the parental P. falciparum 3D7 clone and three independent clones (B1a,b,c; B2a,b,c) from each of the resistant lines selected with bortezomib. These six genomes yielded on average 26 million (26 191 102) reads with an average read length of 100 bp. Of these, 99.1% mapped to the P. falciparum reference genome (PlasmoDB version 13.0). An average coverage of 87.4% was obtained with an average of 97.7% of the genome covered by five or more reads (Table S1). Aligning each sequence as well as the drug-sensitive parent clone sequence to the reference 3D7 line revealed that each of the B1 and B2 clones had acquired a single C → A/T nucleotide variant (at Pf3D7_10, position 441,641), while the B1 line had an additional T→ A variant (at Pf3D7_10, position 441,617). These changes are predicted to result in mutations in Pf 20S β5 (line B1, Met45Ile and Leu53Phe; line B2, Met45Ile) ( Figure  2A). The data also indicate a high degree of specificity: only two other newly emerged high-quality missense variants were detected among the 138 million bases scanned. These included a Gly425Asp change in in eukaryotic translation initiation factor 2 γ subunit (PF3D7_1410600, all B2 clones) and a Gln60His variant in a predicted dicarboxylate/tricarboxylate carrier protein (PF3D7_0823900) in one B2 clone (Table S2). No obvious copy number variants (CNVs) were observed in any of the selected lines.
In the yeast and human β5 subunit, the boron atom covalently interacts with the active site Thr1 and the interaction is further stabilized by the hydrogen bonding network between bortezomib and several closely located residues (i.e., Thr21, Gly47, Ala49, and Ala50). 22,32 The proteasome β5 subunit in P. falciparum has the identical Gly47, Ala49 and Ala50 residues (Figure 2A,C). The structure of Pf 20S has been solved using cryoEM (PDB code 5fmg). 13 As illustrated in Figure 2B and Figure 2C, the mutations are in residues that lie close to the expected bortezomib-binding pocket. 33 Met45 in the yeast and human β5 subunits has been shown to undergo a conformational change to drive induced fit binding of bortezomib into the S1 pocket in the β5 active site. 22,32 Of interest, a Met45Val mutation is readily selected when human cells evolve resistance to bortezomib, 33 and the equivalent Met120Ile mutant was observed in some patients under prolonged treatment with bortezomib 34 and was selected when yeast was rendered resistant to a peptide epoxyketone inhibitor with good activity against P. falciparum. 35 Overlay of the Pf 20S β5 structure and the crystal structure of the human 20S with bound bortezomib (PDB code 5lf3) reveals close conservation of the residues lining the active site ( Figure S5A). In the overlaid Pf 20S structure, Met45 is positioned unfavorably with respect to the bortezomib P1 leucine (predicted distance 1.8 Å; Figure S5B). This indicates that, as is the case for the yeast active site, 22 Met45 would need to be repositioned for bortezomib binding. PyMol 36 was used to model the Met45Ile mutation in silico (without energy minimization). The distance between the bortezomib P1 substituent (Leu) and Ile45 in Pf 20S β5 is predicted to increase to 3.5 Å ( Figure S5C). The increased distance and the inflexibility of the branched β-carbon of Ile45 may alter the local conformation at this site and/or inhibit the conformational changes in the S1 pocket that are needed for induced fit bortezomib binding.
Interestingly, clone B1a has an additional mutation (Leu53) in Pf 20S β5 and exhibits a moderately higher level of resistance

Journal of Medicinal Chemistry
Article than clone B2a. This additional mutation is located close to Met45 in the proteasome structure. A mutation in Cys52 (which lies adjacent to the Leu53 residue in P. falciparum) has been observed in human cell lines selected for bortezomib. The Leu53Phe mutation places a bulky hydrophobic group close to the P1 pocket of Pf 20S β5 ( Figure S5D) and may cause additional alterations in the local conformation of the active site that inhibit bortezomib binding. Taken together, these data confirm that the β5 activity of the P. falciparum proteasome is the main target of bortezomib.
We examined the abilities of the novel compounds to inhibit the growth of the bortezomib-resistant parasite line (B1). The resistant parasites showed very strong cross-resistance to MPI-2 and weaker (2-fold) cross-resistance to MPI-1 (Tables 2A,  3B). Interestingly, the bortezomib-resistant parasite line retained sensitivity to compounds MPI-3 and MPI-4, suggesting that co-inhibition of β2 activity can overcome the resistance phenotype. This is consistent with a recent study 11 that revealed marked synergism of inhibitors of Pf 20S β2 and β5. The resource of bortezomib resistant parasites will provide a valuable underpinning for further efforts to design and validate inhibitors.
Active Site Probe Analysis of Proteasome Subunit Specificity. A fluorescent activity probe (BMV037) contains an epoxyketone peptidic scaffold based on carfilzomib. 37, 38 We synthesized BMV037 using a procedure modified from the published method 38 (see Supporting Information). We treated purified Pf 20S ( Figure 3A) or intact P. falciparum ( Figure 3B) with different concentrations of the novel compounds. Residual proteasome activity in the treated purified Pf 20S or in lysates generated from the treated intact P. falciparuminfected erythrocytes was detected using BMV037. In the

Journal of Medicinal Chemistry
Article absence of inhibitors, the β5 and β2 subunits are labeled most efficiently, with label also incorporated into the β1 subunit ( Figure 3A,B, left lanes), consistent with previous reports. 37,38 Upon treatment of infected erythrocytes with bortezomib, preferential inhibition of the β5 activity is observed (as indicated by an increase in the ratio of β2 to β5 labeling), with some inhibition of β1 activity ( Figure 3A). This is consistent with activity assay results for the purified proteasome (Table  1) and similar to the profile of inhibition in human cells. 21 Similarly, MPI-1 and MPI-2 preferentially inhibited binding of the probe to the β5 subunit. By contrast, MPI-3 and MPI-4 inhibited both β2 and β5 activities ( Figure 3A,B).
We next analyzed the specificity of the compounds against the active sites in purified human 20S proteasome preparations ( Figure S6). BMV037 was incorporated into all the catalytic sites of human constitutive proteasome and immunoproteasome ( Figure S6, left lanes), with a labeling profile similar to that observed in previous studies. 39,40 In agreement with the inhibitory activities assessed using the fluorogenic assay (Table 1B,C), bortezomib preferentially inhibits the β1 and β5 subunits of the human constitutive proteasome and immunoproteasome ( Figure S6). Similarly, MPI-3 and MPI-4 exhibited preferential inhibition of the β1 and β5 subunits. Interestingly, despite MPI-3 exhibiting a low IC 50 value (9 nM) against β5 activity (Table 1B), labeling of the human β1 and β5 subunits was still evident after the treatment with 1 μM MPI-3, especially for the constitutive proteasome. This could be caused by the higher reversibility of MPI-3. MPI-2 preferentially inhibited β1c and β5i and was the only compound that displayed differential inhibition of the constitutive proteasome and immunoproteasome. This differential specificity was not observed in the fluorogenic assay, probably due to the small differences in IC 50 values (<5-fold) between β1 and β5 sites. MPI-1 preferentially targeted the β5 subunits of both the constitutive proteasome and immunoproteasome but did not completely ablate binding of the probe to β5c, even at a concentration of 10 μM, again suggesting reversibility of binding.
We examined the activity probe profile in bortezomibresistant parasites. In the absence of inhibitors, BMV037 labeled all three subunits ( Figure 3C,D), indicating that the carfilzomib-based probe is still able to bind, despite the changes to the β5 active site. As anticipated, bortezomib was unable to prevent binding of the probe in the resistant parasites, consistent with decreased binding of bortezomib to β5. The competition profile for MPI-3 was very similar in the bortezomib-resistant and wild-type parasites (compare Figure  3A and Figure 3C), consistent with the maintenance of activity of this compound against bortezomib-resistant parasites. Similarly, MPI-1 was still able to prevent BMV037 labeling of the β5 active site, consistent with the maintenance of substantive killing activity, while MPI-2 was no longer able to prevent BMV037 labeling of the β5 active site, consistent with the loss of parasite killing activity. Of particular interest, MPI-4 was able to completely inhibit BMV037 labeling of the β2 active site but did not prevent labeling of the β5 active site. This suggests that MPI-4 occupancy of the β5 active site is decreased. The maintenance of activity of this compound against bortezomib-resistant parasites suggests that strong inhibition of the β2 active site (plus potentially weaker, reversible inhibition of β5 activity) is sufficient for parasite killing. This is consistent with a previous study 41 showing that inhibitor occupancy of β2 can potentiate proteasome inhibition by weaker inhibitor binding to β5. This insight will guide efforts to generate inhibitors that are more refractory to the development of resistance.
Interestingly, a recent study examined the activity of AsnEDA β5 inhibitor 11 and proposed the extension of the inhibitor to the S4 pocket which is solvent-exposed and formed by amino acid residues of β6. This points to different modes of binding that can be exploited to further enhance specificity. For example, extending the peptidyl boronic acid structures so that they access this pocket could result in improved activity.
Proteasome Substrate Profiling. To determine differences in substrate preference that could be exploited in further development of inhibitors that specifically target the P. falciparum proteasome, we screened a large library of AMCbased discrete tripeptide substrates (i.e., 5920 of 8800 theoretical Ac-P3-P2-P1-AMC substrates). 42 Positions adjacent to the cleavage site are referred to as the P3, P2, and P1 positions ( Figures 4A−C and S7A,B). The majority of the substrates are preferentially cleaved by the human constitutive proteasome and immunoproteasome (compare Figure 4A, Figure 4B, and Figure S7A). The profile is generally less active and narrower for Pf 20S; however, residues in some positions do confer some specificity toward Pf 20S.
Interestingly, Pf 20S activity is enhanced for substrates with tryptophan (W) at both the P2 and P3 positions ( Figure 4A). By contrast, human 20S exhibited poor selectivity at P2 but also preferred tryptophan at P3 position ( Figures 4B and S7A). In human 20S, preference for tryptophan at P2 generally results from β2 or β5 activity, while it may select for β1 and β2  Figure 4A). The data are somewhat different from a previous study, 13 which suggested that Pf 20S prefers aromatic residues at P1 and P3 due to the more open conformation of the β2-binding pocket compared to the human proteasome. Our data suggest a broader preference profile at P1 and a preference for tryptophan at P3, but the preference for tyrosine or phenylalanine is not as pronounced. Substrate preferences for human constitutive proteasome and immunoproteasome were very similar (Figures 4B and S7A). The only apparent difference was the increased tolerance of large hydrophobic residues (i.e., F, W, Y) by the immunoproteasome at the P1 and P2 positions.
Because the proteasome harbors three different catalytic sites, it is difficult to pinpoint which β subunit is responsible for the cleavage of a particular substrate. Our Pf 20S substrate profiling method has the advantage that P2 and P3 preferences can be assigned to a particular P1 residue and has previously been used to profile several other mammalian and bacterial proteasomes. 43 It also can identify individual peptides with especially high activity or selectivity for one active site versus another, whereas these trends may be muted in pooled library approaches. Such substrates can be used directly in further biochemical assays. 43 Figure 4. Pf 20S and human 20S constitutive (c20S) proteasome substrate profiling and validation of active and selective substrates for P. falciparum proteasome. Human and P. falciparum proteasome activity was measured using a library of AMC-based tripeptide substrates to determine cleavage preferences for each P position. Substrate specificity at P1, P2, and P3 positions was examined for Pf 20S (A) and human constitutive (B) proteasomes. The Y-axis shows the average cleavage rate. Correlation between substrate sequence at each position and average cleavage activity by the proteasome is displayed. Cleavage rate for each substrate is normalized to Ac-WLR-AMC activity (×1000). B indicates either aspartic acid or asparagine. O is pyrrolysine. (C) Selectivity for Pf 20S proteasome compared with human constitutive proteasome. The ratio of Pf 20S to human constitutive proteasome cleavage activity is plotted based on the residues at P1, P2, and P3 positions. (D) Validation of selected substrates with Pf 20S and human constitutive proteasomes. The cleavage rate for each substrate is normalized to Ac-WLR-AMC activity (×1000). Data shown are the mean ± SEM from three independent experiments.

Journal of Medicinal Chemistry
Article Differences in the substrate profiles of Pf 20S and human constitutive 20S are illustrated by the correlation plot of the substrate library ( Figure S8), with many of the selective substrates associated with a high rate of cleavage by their respective targets. The Pf 20S P3 preference for tryptophan (for β2 and β5) favors substrates such as WLR and WLA, while Pf 20S P2 preference for tryptophan (for β1 and β2) favors substrates such as DWD and YWR. Pf 20S β5 (P1 hydrophobic) strongly prefers P3 tryptophan over similar aromatic residues (such as phenylalanine/tyrosine) at P3 (e.g., WQL vs FQL, YQL). This overlaps with human 20S β5 specificity, which is permissive of these alternative P3 residues. Pf 20S β2 (P1 basic) prefers P1 arginine over lysine with aromatic residues in P2 and P3 (e.g., FWR, YWR). In contrast, for human 20S, if P1 is basic (β2), a basic residue is also preferred at P3 with little selectivity in P2 (e.g., RQR, RLR), with KQY also favored, while if P1 is hydrophobic (β5), an aromatic residue is preferred at P3 (e.g., WLA). Pf 20S β1 (P1 acidic) appears to have selectivity for hydrophobic or aromatic P2 or P3 residues. Several P1 His substrates (e.g., KWH, WMH) cannot be definitively ascribed to a particular subunit but appear highly active and selective for Pf 20S. Taken together, these results suggest that significant differences exist between Pf 20S and human 20S, which might be exploited in the design of selective Pf 20S inhibitors.
Overall, the substrate profiling results are similar to a previous study 13 but with some important differences. The selectivity of Pf 20S for tryptophan in P3 is consistent across the studies, but the preferred residue at P2 (W, M, Y) does not match well with the previous report, 13 possibly due to poor selectivity in P2. The rank order of P1 preferred residues roughly corresponds to the previous report.
To validate these substrate preferences, we resynthesized a number of differentiating substrates based on their activity and selectivity for the Pf 20S proteasome and directly compared their relative activities as substrates for Pf 20S proteasome, human 20S constitutive proteasome, and immunoproteasome ( Figures 4D and S7C). The preference of Pf 20S for histidine and arginine in P1 was confirmed, as well as the preference for tryptophan at P2 and P3. It is important to consider residue preferences at different position as a combination rather than individual preferences. For example, Ac-RQR-AMC (arginine at P1) is inactive for Pf 20S but it is a good human proteasome β2 substrate. The preference for aspartate at P1 was not confirmed, possibly due to a problem with the quality of the substrates in the original library. The selectivity of a number of substrates for Pf 20S was confirmed. For example, Ac-KWH-AMC and Ac-WWK-AMC appear both active and selective for Pf 20S. Interestingly, both substrates have tryptophan at the P2 position. While the human immunoproteasome is more tolerant of large bulky side groups at P2 than the constitutive proteasome, Pf 20S retained evident selectivity for these substrates. Thus, the higher inhibitory activity of MPI-3 and MPI-4 against Pf 20S may reflect their bulky hydrophobic groups at P2 and P3, and the fact that the Pf 20S β2 site is more hydrophobic than human 20S. 9 While this manuscript was under review, another study was published that used substrate profiling to design optimized vinyl sulfone-based inhibitors. 44 This work suggests that choice of an optimal electrophilic warhead can further enhance selectivity.

■ CONCLUSIONS
We have identified potent inhibitors of the P. falciparum proteasome that have significant activity against both artemisinin-sensitive and -resistant parasites in culture. While preferential inhibition of P. falciparum β2 activity was readily achieved, we show that inhibition of P. falciparum β5 activity coupled with weaker or reversible activity against human β5 activity represents a signature that permits selective killing of P. falciparum compared with a mammalian line. We developed resistance to bortezomib in P. falciparum and show this is associated with mutations in the β5 active site. These mutations prevent binding of some but not other related peptide boronates, and we show that strong inhibition of the β2 active site permits retention of good antiplasmodial activity even in the context of weaker β5 inhibition. Deployment of proteasome inhibitors antimalarial drugs in combinations with drugs with a different mode of action will help avoid the development of resistance. We identified MPI-1 as a compound with good selectivity between P. falciparum and a mammalian cancer cell line and the ability to maintain reasonable activity against the bortezomib-resistant parasite line. We have identified amino acid residues at the P1, P2, and P3 that can confer selective binding to the P. falciparum proteasome, providing a path to further modification of the peptide boronate scaffold to generate more selective inhibitors.
■ EXPERIMENTAL METHODS General Chemistry. All reagents and solvents were used as obtained. 1 H NMR spectra were run on a 400 MHz Bruker spectrometer in the solvent indicated. Full assignment of the NMR peaks can be found in the Supporting Information ( Figure S9). LC/ MS spectra were recorded on an LC or UPLC system connected to a mass spectrometer using reverse phase C18 columns. Various gradients and run times were selected in order to best characterize the compounds. Mobile phases were based on ACN/water or MeOH/water gradients and contained either 0.1% formic acid or 10 mM ammonium acetate (typical gradients of 100% mobile phase A (mobile phase A = 99% water + 1% ACN + 0.1% formic acid) to 100% mobile phase B (mobile phase B = 95% ACN + 5% water + 0.1% formic acid) at a flow rate of 1 mL/min for a 16.5 min LC run or 0.5 mL/min for a 5 min UPLC run). All compounds were determined to be >95% pure by 1 H NMR or LC/MS. BMV037 Fluorescent Probe. BMV037 was prepared using a modification of the published procedure 38 (see Supporting Information for details). 1 H NMR spectra were run on a 400 MHz Varian Inova spectrometer. LC/MS spectra were recorded on an Agilent RP-HPLC system connected to a mass spectrometer using a reverse phase C18 column. Compound purity was assessed using a C18 150 mm × 4.6 mm 5 μm column in gradient mode with eluent (buffer) A, 0.1% aq TFA, and buffer B, 0.1% TFA in ACN.
HCT116 and HepG2 Cell Culture. HCT-116 and HepG2 cells were obtained from the American Type Culture Collection and maintained as recommended by the supplier. Cells were cultured in the ATCC-formulated McCoy's 5A medium or DMEM (Life Technologies, CA), supplemented with fetal bovine serum to a final concentration of 9−10%, and incubated at 37°C in an atmosphere of 95% air and 5% CO 2 . Cellular toxicity assays were performed using Promega's CellTiter-Glo assay system. Varying concentrations of the test compounds in 5% DMSO/95% PBS were dispensed into a 384well plate and cells added to each well. The plates were incubated at 37°C for 72 h, and the CellTiter-Glo assay to assess cell viability was performed as described by the manufacturer (Promega).
Assessment of Cellular Toxicity. For standard assays of antiplasmodial activity, sorbitol-treated ring stage parasites were incubated with the drugs for 72 h and viability was assessed in the second cycle. 12 Drug pulse assays were performed as described previously. 47 For short pulse assays, tightly synchronized early ring stage parasites (1−3 h after invasion) at 0.2% hematocrit and 1−2% parasitemia were subjected to 3 h drug pulse at 37°C and returned to culture until parasites reached trophozoite stage of the following cycle. Cells were then stained with 2 μM Syto-61 (Thermo Fisher Scientific), and their fluorescence was measured by flow cytometry (FACSCanto II cytometer; Becton Dickinson). Data were gated and

Journal of Medicinal Chemistry
Article analyzed using FCS Express software (version 3) to determine parasitemia of each sample. Viability represents the parasitemia normalized to untreated and "kill treated" controls, where "kill treated" refers to samples treated with 2 μM DHA for 48−72 h. Interactions between DHA and MPI-1 or MPI-4 against the K13 mutant (DHA resistant) Cam3.II P. falciparum strain, exposed to a 3 h pulsed treatment, were determined at 4 h after invasion as described previously. 12 For human cell line activity assays, cellular toxicity assays were performed as described previously. 23,48 P. falciparum Proteasome Preparation and Assay Optimization. P. falciparum 20S proteasome was enriched from infected RBCs using a two-step chromatographic procedure. 9 Mass spectrometry revealed the presence of all 14 Pf 20S proteasome subunits with multiple unique peptide fragments. To determine the level of human proteasome contamination in the purified samples, 7 μg of the purified Pf 20S proteasome was reduced with DTT and alkylated with iodoacetamide. The sample was then subjected to SDS−PAGE, and the gel fraction containing 14−38 kDa bends was excised. In-gel digestion with trypsin was performed, and peptide content was analyzed by LC−MS/MS. The data were processed using Proteome Discoverer 1.4 employing a user-defined protein database containing human and Pf 20S proteasome subunits.
Peptide pairs of human and Pf 20S subunits with similar amino acid sequences were selected and used to estimate the amount of P. falciparum proteasome relative to human proteasome. Signal intensity (peak area) from multiple peptide pairs was determined. The signal intensity ratio of Pf 20S to human 20S proteasome was 240:1 on average, indicating a level of contamination by human proteasome of less than 0.5%. Human proteasome and human proteasome activator complex, PA28αβ, were prepared as described previously. 49 Optimization experiments indicated that 1 nM Pf 20S was sufficient to produce a strong and reliable signal in the presence of 24 nM PA28αβ, which was found to activate Pf 20S maximally.
Labeling of Proteasome Catalytic Subunit by BMV037. P. falciparum culture, purified Pf 20S proteasomes, human constitutive proteasomes, or immunoproteasomes (Boston Biochem) were incubated with test compounds at 37°C for 1 h. Treated parasite lysates (10 μg), Pf 20S proteasomes (80 nM), human constitutive proteasomes (20 nM), or immuno-20S proteasomes (20 nM) were labeled with BMV037 (10 μM) at 37°C for 2 h. Samples were mixed with SDS loading buffer and heated at 95°C for 5 min. Samples for P. falciparum and human immuno-20S were run on 4−12% Bis-Tris acrylamide gels using MES SDS running buffer (Thermo Fisher Scientific). Human constitutive 20S samples were run on a 10% Bis-Tris acrylamide gel using MOPS SDS running buffer (Thermo Fisher Scientific). Gels were imaged at the Cy5 channel on a Gel Doc XR+ Documentation System (Bio-Rad) or an Amersham Typhoon Trio imager (GE Healthcare Life Sciences).
In Silico Analysis. All structure analysis was performed using PyMOL. 36 The structure of human 20S (PDB code 5lf3, chain K) was superimposed onto P. falciparum 20S (PDB code 5fmg, chain L). The distances between atoms was measured using the Measurement Wizard feature, with bortezomib remaining in the position modeled by the human 20S bound state. Mutagenesis was performed using the Mutagenesis Wizard feature, with rotamers showing minimal clashes being selected. No energy minimization modeling was performed.
In Vitro Evolution of Resistance to Bortezomib and Whole Genome Sequencing. Prior to selection, an aliquot of the parental line was stocked as a reference for subsequent whole genome sequencing analysis. Several independent clones of P. falciparum 3D7 parasite line were cultured in Petri dishes exposed to increasing drug pressure for ∼30 weeks. Parasites were cloned by limiting dilution. For parasite extraction and genomic DNA isolation, cultures were scaled up to at least 4−5% parasitaemia, lysed in saponin, and genomic DNA was isolated using ISOLATE II Genomic DNA kit (Bioline). DNA libraries for each gDNA sample were prepared for sequencing using the Nextera XT kit as described previously. 50 Libraries were clustered and run on an Illumina HiSeq 2500 using the RapidRun mode, sequencing 100 base pairs in depth on either end. Paired-end reads were either aligned to the P. falciparum 3D7 reference genome (PlasmoDB version 13.0).
Measurement of GFP-DD Signal Using Flow Cytometry or Western Blotting. Trophozoite-stage parasite cultures (GFP-DD parasites) at 5% hematocrit and 5% parasitemia were subjected to the drug treatment indicated in the figure legends at 37°C. For flow cytometry analyses, culture samples were stained with 2 μM Syto-61 and adjusted to 0.1% hematocrit in PBS. Syto-61 and GFP fluorescence was recorded using the FACSCanto II cytometer (Becton Dickinson). Analysis was performed in FlowJo (version10). Parasites were gated based on parasite GFP fluorescence, and values reported are mean fluorescence. Data were fit by sigmoidal, nonlinear analyses using GraphPad Prism software. For Western blotting, trophozoite-stage parasites were isolated with 0.05% w/v saponin and pellets were washed in PBS, supplemented with EDTA-free protease inhibitor cocktail (Roche). Parasite pellets were solubilized in reducing SDS−PAGE sample buffer, boiled at 95°C for 8 min, resolved by SDS−PAGE on a 4−12% Bis-Tris acrylamide gel (Life Technologies) using MOPS running buffer and transferred to nitrocellulose membrane (iBlot; Life Technologies). Membranes were blocked with 3.5% skim milk for 1 h at room temperature and probed with mouse anti-GFP (Roche; 1:1000) or mouse anti-Pf BiP (1:1000) overnight at 4°C, followed by goat anti-mouse IgGperoxidase (Chemicon-AP127P; 1:20 000) for 1 h at room temperature. Polyclonal mouse anti-Pf BiP was generated using recombinant Pf BiP at the WEHI Antibody Services. Blots were visualized using Clarity ECL substrate (BioRad).

* S Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jmedchem.8b01161.

Journal of Medicinal Chemistry
Article Molecular formula strings and some data (CSV) Tables S1 and S2 listing sequencing statistics and summary of single nucleotide variants and insertiondeletions for bortezomib resistant P. falciparum clones; Figure S1−S9 showing purification of Pf 20S proteasome; optimization of Pf 20S proteasome activities; kinetic analysis of MPI-4; in vitro evolution of resistance to bortezomib; structure analysis of β5 mutations associated with bortezomib resistance; active site probe analysis of the selected compounds; proteasome substrate profiling of human 20S immunoproteasome and P. falciparum proteasome; substrate preferences of human constitutive and Pf 20S proteasomes; NMR structural assignments of MPI compounds; synthesis of BMV037 (PDF)