Design and synthesis of a cross-species hydroxamic acid-based inhibitor selective for Plasmodium M1 alanyl aminopeptidase
We previously reported a series of PfA-M1 and M17 aminopeptidase inhibitors that possessed a hydroxamic acid zinc binding group to coordinate catalytic zinc ion(s) and a variety of hydrophobic groups to probe the S1′ binding pocket of these enzymes.21 Whilst a number of these modifications successfully improved inhibitory potency, their incorporation reduced polarity and aqueous solubility. One such example is the pivaloyl group present in compound 1 (Fig. 1A).20,21 In order to balance these opposing factors, one of the methyl groups of the pivaloyl was replaced by an alcohol to increase polarity whilst maintaining the capacity to engage the S1’-pocket via hydrophobic interactions (Fig. 1A). This modification resulted in MIPS2673 (compound 4), which has a significantly lower cLogP (1.45 vs 2.23 for 1) and favourable solubility (i.e. the kinetic solubility of MIPS2673 was estimated to be in the 50–100 µg/mL range using nephelometry vs 12.5–25 µg/mL for 1). MIPS2673 (4) was synthesized from methyl 2-amino-2-(4-bromophenyl)acetate (2) in three synthetic steps as shown in Fig. 1. Amide formation was achieved via an EDCI-mediated coupling and the trifluorophenyl moiety was installed using a Suzuki-Miyaura coupling to form the synthetic intermediate 3. Finally, the methyl ester was converted to the corresponding hydroxamic acid using 5 M methanolic KOH and NH2OH•HCl to afford MIPS2673 (4).
We determined the inhibitory activity of MIPS2673 against M1 and M17 enzymes from both P. falciparum and P. vivax (Fig. 1B). The binding affinities (Ki) of MIPS2673 toward purified, recombinant PfA-M1 shows the compound to be a potent inhibitor (Ki = 211 ± 11 nM), with > 4-fold selectivity over PfA-M17 (Ki = 921 ± 69 nM). MIPS2673 was significantly more potent against Pv-M1 (Ki = 6.4 ± 0.5 nM) with > 155-fold selectivity over Pv-M17 (Ki > 1000 nM). We confirmed that MIPS2673 is non-inhibitory against the PfA-M18 (Ki > 500 µM) and PfAPP (Ki > 40 µM) recombinant metalloaminopeptidases (data not shown).
In contrast to the M1 selectivity demonstrated by MIPS2673, the inhibitor 1 was 2-fold more selective for PfA-M17 vs PfA-M1.21 When bound to PfA-M1, the position of the biaryl of MIPS2673 compares with that of our previous inhibitors.20,21 The moiety makes similar interactions with Met1034 (hydrophobic) and Glu319 (carbonyl-pi) (Fig. 1B). The fluoro-substituents sit deep in the S1 pocket and form the same intricate network of water-mediated hydrogen-bonds as 1, in which the fluorine atoms act as H-bond acceptors. The substitution on the tert-butyl, replacing a methyl with an alcohol, is difficult to definitively position in the structure due to the rotation around the C-03 bond. It is possible that this alcohol could form a long H-bond to Arg-489, which should provide some improvement in position and potency. The Arg-489 side-chain shows two orientations in the electron density, suggesting that some rotation of the alcohol is occurring (Fig. 1B). Inspection of the structure of PfA-M17 with 1 (PDB ID 4ZY2)20 does not reveal why the compound is selective. The tert-butyl of 1 is sitting in an open pocket and the introduction of an alcohol is unlikely to induce any clashes, steric or electrostatic. This suggests that the reason for the selectivity between the two enzymes results from the ability (or inability) to access the interior cavity of the PfA-M17 hexamer that houses the active sites.
To investigate the potential for off-target effects, we examined the selectivity of MIPS2673 over several human M1 homologs, including leukotriene A-4 hydrolase (LTA4H), endoplasmic reticulum aminopeptidase 1 (ERAP1) and endoplasmic reticulum aminopeptidase 2 (ERAP2). MIPS2673 showed minimal inhibition of ERAP1 and ERAP2 at concentrations below 500 µM (Table S1), whereas MIPS2673 caused > 50% inhibition of LTA4H at concentrations above 10 µM (Table S1). We also tested MIPS2673 against the HEK293 cell line to further predict potential human cytotoxicity. No cytotoxicity to HEK293 cells was observed up to a concentration of 40 µM. As a comparison, the current antimalarials chloroquine and dihydroartemisinin inhibit HEK293 proliferation by 67% and 29%, respectively, at 40 µM. At 120 µM, MIPS2673 inhibited cellular proliferation by 89% compared to the untreated control (Table S2).
In vitro antiparasitic activity of MIPS2673
Having synthesised a Plasmodium specific M1 aminopeptidase inhibitor, we next explored the activity of MIPS2673 against in vitro blood-stage cultures of P. falciparum. We found that MIPS2673 has sub micromolar activity against the laboratory reference strain, 3D7 (72 h EC50 = 324 nM (250–470 CI)) (Fig. 1C), as well as against a panel of drug resistant P. falciparum lines (Table S3). Stage specificity assays were performed to identify which stage of asexual development MIPS2673 is most active against. Synchronised ring or trophozoite cultures were exposed to MIPS2673 for 24 h or 48 h at 10x EC50, then stringently washed to remove the compound. Parasite growth was assessed after 48 h compared to a vehicle-treated control. This confirmed MIPS2673 is most potent against trophozoite stage parasites (Fig. 1D), corresponding to the period of peak PfA-M1 expression26 and activity.48 As expected, no stage-dependent difference in activity was observed in parasites exposed to MIPS2673 for 48 h, as this duration incorporates the full asexual erythrocytic cycle. The control compound, artesunate, was equipotent across all conditions, consistent with the known activity and kinetic profiles for artemisinins.49
To visually characterise the effects of MIPS2673 on parasites, we treated synchronised 3D7 cultures at the early ring stage and monitored their development by light microscopy evaluation of Giemsa-stained thin blood smears (Fig. 1E). We found that MIPS2673 caused significant growth retardation when compared to the untreated control and parasite growth stalled at early trophozoite (22–28 h trophozoites) and late ring (~ 22 h rings) stages at 5x and 10x EC50, respectively. The presence of haemozoin in both treatment groups indicated that the initial catabolism of haemoglobin is unaffected by MIPS2673.
Having confirmed activity against asexual stages of P. falciparum, we next determined whether MIPS2673 kills the transmissible sexual forms of the parasite, known as gametocytes. We found that MIPS2673 has transmission-blocking activity and is most potent against early gametocytes (stages I-III), compared to late (stages IV-V) and mature (stage V) stages (Table S4), but is less potent than the artemisinin derivative, artesunate.
Identification of molecular targets for MIPS2673 in P. falciparum by thermal stability proteomics
Our biochemical studies showing binding and inhibition of recombinant Plasmodium metalloaminopeptidases do not rule out the possibility that the anti-parasitic activity of MIPS2673 is due to non-specific binding to other parasite metalloproteins or off-target proteins. Genetic target validation was not possible as previous attempts to mutate PfA-M1 were not successful,12,50 and MIPS2673-resistant parasites were not available. Therefore, we developed orthogonal mass spectrometry-based chemoproteomics approaches to validate that MIPS2673 is on-target and M1-selective in the complex parasite environment (Fig. 2). We initially developed a streamlined thermal stability proteomics workflow that combined traditional thermal proteome profiling methods with an optimised data-independent acquisition (DIA)-LC-MS/MS approach.51,52
To identify the binding target/s of MIPS2673 in P. falciparum asexual blood stages in a proteome-wide manner, we used native parasite lysates, as direct drug-protein interactions are more selectively identified in cellular lysates, rather than live cells that are susceptible to downstream effects of drug action. The parasite lysates were exposed to 1 µM or 4 µM of MIPS2673 or vehicle (DMSO control) for 3 minutes prior to heating at 60°C, a temperature that should allow detection of most drug-induced protein stabilisation events in an untargeted manner with wide proteome coverage.45,53 After the thermal challenge, the soluble (non-denatured) protein fraction was isolated by ultracentrifugation, digested overnight with trypsin and the peptide mixture analysed directly using global DIA-LC-MS/MS. Proteins detected with significantly higher abundance in treated relative to control samples reflect thermal stabilisation of the target due to ligand binding. To minimise identification of false positive hits, proteins reproducibly stabilised (p < 0.05 and fold-change ≥ 1.2 compared to the untreated control) across multiple drug concentrations and experiments were considered drug interacting proteins. Among 1632 proteins reproducibly detected with a minimum of two peptides across two independent experiments (each with at least three independent incubations of MIPS2673 or vehicle with protein lysate), five proteins were consistently stabilised at both drug concentrations compared to DMSO (Fig. 3A, Data S1 and Data S2). Of these, PfA-M1 (PF3D7_1311800) was one of the most significantly stabilised proteins (p < 0.01 at 1 µM). PfA-M1 exhibited concentration-dependent stabilisation in the presence of MIPS2673, with an average stabilisation of 1.8-fold and 2.8-fold at 1 µM and 4 µM, respectively, relative to the untreated control (Fig. 3B). The four other proteins consistently stabilised were not metalloproteins and included two conserved Plasmodium proteins with unknown functions (PF3D7_1026000 and PF3D7_0604300), a putative AP2 domain transcription factor (PF3D7_1239200) and a human protein, Ras-related Rab-39A (Q14964) (Fig. 3B). These proteins could represent possible off-target binding. The PfA-M17 protein was not stabilised by MIPS2673 at either drug concentration after a 60°C thermal challenge (Fig. S1). We previously showed PfA-M17 is stabilised at this temperature in parasite lysates treated with a selective PfA-M17 inhibitor.11 Overall, our unbiased thermal stabilisation proteomics approach confirmed MIPS2673 selectively targets the M1 aminopeptidase over M17 and does not bind indiscriminately to parasite metalloproteins.
Identification of MIPS2673 targets in P. falciparum by limited proteolysis coupled mass spectrometry
We also developed an efficient protocol for limited proteolysis-based studies of P. falciparum and applied it for the unbiased identification of targets of MIPS2673 (Fig. 2). Native P. falciparum lysates were treated for 10 minutes with different concentrations of MIPS2673 (1 µM or 10 µM in experiment one and 0.1 µM, 1 µM or 10 µM in experiment two) or vehicle in at least four independent incubations per experiment. Proteome extracts were then subjected to double protease digestion. An initial limited proteolysis with proteinase K for 4 minutes captured local structural alterations of proteins that become differentially susceptible to protease cleavage upon drug binding. Secondly, samples undergo complete digestion under reducing conditions overnight with trypsin/LysC to make peptides amenable for global proteomics analysis. Differentially abundant proteolytic peptides between MIPS2673 and vehicle-treated samples were then identified on a global scale with DIA-LC-MS/MS. The proteolytic peptide patterns of drug targets should be altered in the treated samples as compound binding prevents protein cleavage by proteinase K, resulting in decreased abundance of peptides with non-tryptic ends or an increase in concentration of the associated fully tryptic peptide (Fig. 4A). We quantified 26,611 peptides from 2,153 proteins in experiment one and 16,662 peptides from 1,989 proteins in experiment two (Data S3). Each dataset was initially analysed for differentially abundant peptides between each MIPS2673 concentration and vehicle by filtering based upon relative peptide abundance (absolute fold-change > 1.5), statistical significance (q < 0.01) and proteolytic peptide pattern (i.e. increased fully tryptic or decreased half tryptic). In experiment one and experiment two respectively, approximately 3% of peptides from 379 proteins and 0.6% of peptides from 69 proteins, met these thresholds at each drug concentration and were considered significant (Fig. 4B). After prioritising targets based upon the number of significant peptides detected per protein (Fisher exact test, Bonferroni corrected p < 0.05), we identified seven putative drug targets in experiment one and only one putative target in experiment two (Fig. 4C and Data S4). Interestingly, among the seven possible target proteins identified by limited proteolysis in experiment one, five were metalloproteins. These included PfA-M1, three other aminopeptidases - M17 leucyl aminopeptidase, aminopeptidase P (PfAPP) and M18 aspartyl aminopeptidase (PfA-M18) - and adenosine deaminase (PfADA) (Fig. 4C). The expected target, PfA-M1 aminopeptidase, was the only protein reproducibly identified as the MIPS2673 target across both independent limited proteolysis experiments. Indeed, the consistent identification of PfA-M1 as the only significant protein in both LiP-MS and thermal stability proteomics methods strongly supports PfA-M1 being the primary target of MIPS2673.
Features of structurally significant Pf A-M1 LiP-MS peptides
The identification of drug targets with LiP-MS typically identifies structurally perturbed peptides located in very close proximity to the ligand binding site.46,54 Among the 108 PfA-M1 peptides reproducibly detected across both LiP-MS experiments we identified nine structurally significant LiP-MS peptides (q < 0.01 and absolute fold-change > 1.5 at all drug concentrations in both experiments) (Fig. 5A) and mapped these to our PfA-M1 crystal structure with MIPS2673 bound. We measured the minimum distance between atoms of the significant peptides and those of MIPS2673 and found that structurally significant peptides identified with LiP-MS were frequently located in very close proximity to MIPS2673. The median minimum distance between the significant LiP-MS peptides and bound MIPS2673 was 6.5 Å, significantly less than the 13.9 Å median distance for all other detected PfA-M1 peptides (Fig. S2). Among the nine structurally significant LiP-MS peptides, four contained atoms within Van der Waals distance (< 4 Å) of MIPS2673. Based on these observations, we hypothesised that significantly dysregulated LiP-MS peptides could estimate the known MIPS2673 binding site on PfA-M1. A distance of 6.44 Å from MIPS2673 was used to define the binding cleft boundary54 and included the PfA-M1 residues known to interact with bound MIPS2673. Six of the nine significant LiP peptides overlapped with, or were within 4 Å of this binding cleft. The median distance between the atoms of significant LiP peptides and of cleft residues was 2.6 Å, compared to a median distance of 8.2 Å for all other PfA-M1 peptides (Fig. 5B). Of the three significant LiP peptides located > 4 Å from the operational binding site, two are located at the C-terminal opening in domain IV that forms a channel leading towards the active site.30 To understand if this type of data could approximate a ligand binding site correctly, we calculated the centre of mass of the atoms of the nine structurally significant LiP peptides and represented this as a geometric point within the PfA-M1 structure and then compared this to the known binding site (Fig. 5C). The minimum distance between this point and MIPS2673 was 5.2 Å and the MIPS2673 centre of mass neighbourhood (residues within 6.44 Å of the centre of mass) overlapped with the binding site (Fig. 5C). This indicated that our LiP-MS approach provides a good approximation of the MIPS2673 binding site with its target, PfA-M1.
Untargeted metabolomics analysis of P. falciparum infected red blood cells treated with MIPS2673
To further determine the specificity of MIPS2673 for PfA-M1 and whether it induces off-target effects, we performed untargeted metabolomics on infected red blood cells treated with 1 µM of MIPS2673 (3x EC50 value) for 1 h and compared the profile to vehicle (DMSO) control (4–9 biological replicates). Heatmap analysis of relative abundances of all putative metabolites revealed that treatment with MIPS2673 disproportionally impacted peptide metabolism (Fig. S3). Of the 201 putative peptides identified, 97 were significantly dysregulated (p < 0.05). The majority of dysregulated peptides were significantly increased (fold-change > 1.5 and p < 0.05) in abundance in MIPS2673 treated cultures compared to DMSO control, indicative of aminopeptidase inhibition. Targeted analysis of the 97 increased peptides in drug treated cultures revealed that the majority were likely derived from haemoglobin (Hb) chains (α and β) (Fig. 6A). Mapping to Hb sequences was possible for 24 peptides using MS/MS spectra to confirm peptide sequences (Fig. 6B; green dots). For the remaining putative peptides identified by accurate mass, but for which MS/MS spectra could not be obtained, we assessed whether any peptide isomeric to the putative peptide could be mapped to Hb (Fig. 6B; orange dots for peptides mapping to Hb chains and blue dots for peptides that could not be mapped). Overall, ~ 80% of significantly dysregulated peptides could be mapped to one of the Hb chains, with nearly all increasing in abundance following treatment with MIPS2673 compared to DMSO control. To further validate that the majority of elevated peptides are likely to be Hb-derived, we repeated this same analysis for each of the ~ 4700 proteins identified in our recent comprehensive proteomic analysis of P. falciparum-infected red blood cells.51 We quantified the number of peptide matches to each protein and then divided by protein length to yield a normalized estimate of the similarity of each protein to our significantly dysregulated peptides. By this measure, Hb chains α, and β were the most highly matched protein compared to the remaining infected red blood cell proteome (Fig. 6C; red bars), indicating that MIPS2673 predominantly, but not exclusively, disrupts Hb digestion.
We have recently reported that an inhibitor of PfA-M17 (MIPS2571) also predominantly disrupted metabolism of short haemoglobin-derived peptides, which was a metabolic signature consistent with genetic knockdown of PfA-M17.11 However, a direct comparison of peptide perturbations induced by MIPS2673 and MIPS2571 (data from Edgar et al, 2022) showed clearly distinct peptide profiles. Whilst many peptides accumulated with both inhibitors, the extent of peptide accumulation differed, and a subset of short basic peptides (containing Lys or Arg) were elevated following exposure to MIPS2673, but not the PfA-M17 inhibitor (Fig. 6A). Furthermore, the peptide changes observed following treatment with MIPS2673 differ substantially from the peptide changes observed following treatment of P. falciparum-infected red blood cells with other potent antimalarials such as artemisinins55,56 and mefloquine.57 Those antimalarials induce significant depletion of haemoglobin-derived peptides, in contrast to the accumulation observed with the aminopeptidase inhibitors. Overall, the metabolic profile induced by MIPS2673 is unique, and consistent with specific inhibition of PfA-M1.