Discovery and Optimization of Selective Inhibitors of Meprin α (Part I)

Meprin α and β are zinc-dependent proteinases implicated in multiple diseases including cancers, fibrosis, and Alzheimer’s. However, until recently, only a few inhibitors of either meprin were reported and no inhibitors are in preclinical development. Moreover, inhibitors of other metzincins developed in previous years are not effective in inhibiting meprins suggesting the need for de novo discovery effort. To address the paucity of tractable meprin inhibitors we developed ultrahigh-throughput assays and conducted parallel screening of >650,000 compounds against each meprin. As a result of this effort, we identified five selective meprin α hits belonging to three different chemotypes (triazole-hydroxyacetamides, sulfonamide-hydroxypropanamides, and phenoxy-hydroxyacetamides). These hits demonstrated a nanomolar to micromolar inhibitory activity against meprin α with low cytotoxicity and >30-fold selectivity against meprin β and other related metzincincs. These selective inhibitors of meprin α provide a good starting point for further optimization.


Introduction
Meprin α and meprin β are zinc-dependent proteinases implicated in multiple diseases including cancers [1], fibrosis [2,3], and Alzheimer's [4,5]. Meprins cleave multiple cytokines and adhesion molecules thus contributing to inflammation and migration of inflammatory cells [6]. Chronic inflammation can lead to the excess deposition of collagen I resulting in fibrosis [7,8]. Meprins have been shown to cleave procollagen I leading to its maturation and deposition in skin and lung [3,9]. The roles of meprins in various processes are mediated via the cleavage of biological molecules. There are examples of common substrates that meprin α and meprin β share amongst themselves [10] and with other proteases [4]. This complicates the understanding of their respective roles in the specific disease scenarios and, as a consequence, their value as targets for drug discovery. To validate either meprin as a target in any particular disease, target modulation by combination of genomic (e.g., knockdown, overexpression) [11] and pharmacologic means (e.g., small molecules) [12] could be useful. However, due to the relatively recent discovery of meprins' involvement in pathologic conditions there are very few reports of small molecule inhibitor discovery efforts for these enzymes. Kruse et al. [13] reported several known metzincin meprins' involvement in pathologic conditions there are very few reports of small molecule inhibitor discovery efforts for these enzymes. Kruse et al. [13] reported several known metzincin inhibitors that are capable of inhibiting meprins with some degree of selectivity. However, these inhibitors were not selective for other metzincins, which made their utilization for studying the roles of meprins in various diseases difficult. Our group had reported the first low nanomolar meprin β inhibitors, NFF449 and PPNDS (Figure 1, Ki = 22 nM and 8 nM, respectively), with ~100-fold selectivity against meprin α and good selectivity against adamalysins and matrixins [14]. Ramsbeck et al. (2017) reported the low nanomolar selective meprin β inhibitor, 11 g, with 46-fold selectivity against meprin α (Figure 1, IC50 = 2735 nM and 60 nM for meprin α and β, respectively) with good selectivity against adamalysins and matrixins [15]. They also reported improved compounds based on the same scaffold [16] (Figure 1). The best compounds from this series, 8 h and 8i, are 27-fold and 15-fold selective against meprin α (IC50 = 23 nM and 626 nM for 8 h and 24 nM and 368 nM for 8i, for meprin β and α, respectively). A measure of 200 µM of either inhibitor had only limited effect on MMP and ADAM activity, but IC50 values were not reported. Tan et al. (2018) reported the first selective inhibitors of meprin α, 10d and 10e, with 18-and 19-fold selectivity against meprin β [17] (Figure 1). Herein we report the results of a large-scale parallel high-screening throughput effort to discover novel inhibitors of meprin α and meprin β.

Assay Miniaturization and Optimization in 1536 Well Plate Format
The meprin α and meprin β assays, which utilize the substrates (Mca)-YVADAPK-(K-ε-Dnp) and (Mca)-EDEDED-(K-ε-Dnp), respectively, have been described previously [14]. To enable an ultra-high-throughput screening (uHTS) campaign, we proceeded to miniaturize both assays to 1536 well plate format (wpf). First, we recapitulated the assays in 1536 well plate using reagents at the same concentrations as in 384 well plate format assays by scaling the volume down by the factor of 2.5. This resulted in the final volume of the assays of 4 µL. The meprin α assay in 1536 well plates demonstrated a lower signal-to-basal (S/B) ratio than in 384 well plates (1.85 vs. 2.3, respectively), but a better Z' value (0.76 vs. 0.6, respectively), suggesting that the assay is very suitable for large-scale HTS [18]. Actinonin's IC50 values were within 2-fold of each other (5.7 nM and 11 nM for 1536 and 384 well plate format, respectively) ( Figure 2A and Table 1).

Assay Miniaturization and Optimization in 1536 Well Plate Format
The meprin α and meprin β assays, which utilize the substrates (Mca)-YVADAPK-(K-ε-Dnp) and (Mca)-EDEDED-(K-ε-Dnp), respectively, have been described previously [14]. To enable an ultra-high-throughput screening (uHTS) campaign, we proceeded to miniaturize both assays to 1536 well plate format (wpf). First, we recapitulated the assays in 1536 well plate using reagents at the same concentrations as in 384 well plate format assays by scaling the volume down by the factor of 2.5. This resulted in the final volume of the assays of 4 µL. The meprin α assay in 1536 well plates demonstrated a lower signalto-basal (S/B) ratio than in 384 well plates (1.85 vs. 2.3, respectively), but a better Z' value (0.76 vs. 0.6, respectively), suggesting that the assay is very suitable for large-scale HTS [18]. Actinonin's IC50 values were within 2-fold of each other (5.7 nM and 11 nM for 1536 and 384 well plate format, respectively) ( Figure 2A and Table 1).  Meprin β assay exhibited greater S/B in 1536 wpf than in 384 wpf (6.9 vs. 4.4, respectively), while Z' factor values were identical at 0.9. NFF449 IC50 values were 48 nM and 53 nM for 1536 and 384 wpf, respectively ( Figure 2B and Table 1). Despite excellent Z' values in the 1536 wpf in both assays, we wanted to ensure an optimal balance between robustness and sensitivity; in particular with meprin α.
First, both assays were run for 180 min at three different enzyme concentrations including the concentrations at which the assays were recapitulated in 1536 wpf (1.3 nM and 0.05 nM for meprin α and meprin β, respectively). QC parameters (Z' and S/B) and IC50 values of pharmacological controls (actinonin and NFF449) were calculated at 30, 60, and 90 min of the reaction time. The meprin α assay displayed the best S/B values after 90 min of reaction time using 1.3 nM enzyme; however, the reaction progress curve was not linear at the 90 min time point ( Figure 3A). This suggested that while longer reaction times and higher than 1.3 nM enzyme concentration may lead to somewhat better S/B values, the assay sensitivity may suffer due to a nonlinear relationship between signal and proteolysis inhibition. Therefore, to ensure optimal assay sensitivity, we chose 60 min reaction end point and 1.3 nM meprin α as final assay conditions for the primary HTS campaign.  Meprin β assay exhibited greater S/B in 1536 wpf than in 384 wpf (6.9 vs. 4.4, respectively), while Z' factor values were identical at 0.9. NFF449 IC50 values were 48 nM and 53 nM for 1536 and 384 wpf, respectively ( Figure 2B and Table 1). Despite excellent Z' values in the 1536 wpf in both assays, we wanted to ensure an optimal balance between robustness and sensitivity; in particular with meprin α.
First, both assays were run for 180 min at three different enzyme concentrations including the concentrations at which the assays were recapitulated in 1536 wpf (1.3 nM and 0.05 nM for meprin α and meprin β, respectively). QC parameters (Z' and S/B) and IC50 values of pharmacological controls (actinonin and NFF449) were calculated at 30, 60, and 90 min of the reaction time. The meprin α assay displayed the best S/B values after 90 min of reaction time using 1.3 nM enzyme; however, the reaction progress curve was not linear at the 90 min time point ( Figure 3A). This suggested that while longer reaction times and higher than 1.3 nM enzyme concentration may lead to somewhat better S/B values, the assay sensitivity may suffer due to a nonlinear relationship between signal and proteolysis inhibition. Therefore, to ensure optimal assay sensitivity, we chose 60 min reaction end point and 1.3 nM meprin α as final assay conditions for the primary HTS campaign.  The meprin β assay progress curve was hyperbolic rather than linear at 0.05 nM and 0.025 nM enzyme; therefore, we chose 0.0125 nM enzyme concentration where assay linearity was demonstrated ( Figure 3B). Z' and S/B values were acceptable at 60 min reaction end point (0.86 and 2.6, respectively). IC50 values of NFF449 were not significantly affected by the variations of reaction length and meprin β concentrations.
Next, we performed substrate optimization to achieve balanced assay conditions defined as [S]/KM = 1 [19]. In order to do that, we first determined kinetic parameters of proteolysis of meprin α and meprin β substrates by the respective enzymes ( Figure 4A,B). Meprin α and meprin β proteolysis exhibited similar KM values (2.4 ± 0.3 µM and 2.7 ± 0.7 µM, respectively) suggesting the need for optimization of both assays' substrate concentration. Meprin β exhibited >20-fold faster turnover of its substrate than meprin α (6.4 ± 0.06 s -1 versus 0.29 ± 0.06 s-1, respectively) which is consistent with >100-fold difference in enzyme concentrations for meprin α and meprin β assays (1.3 nM versus 0.0125 nM, respectively). To optimize substrate concentrations, both assays were run for 90 min at three different substrate concentrations (10, 5, and 2.5 µM) which included the concentration at which the assays were recapitulated in 1536 wpf (10 µM for both meprin α and meprin β) and the concentration approximating [S]/KM = 1 condition (2.5 µM). Enzyme concentrations were fixed at 1.3 nM for meprin α and 0.0125 nM for meprin β. QC parameters (Z' and S/B) and IC50 values of pharmacological controls (actinonin and NFF449) were calculated at 40, 60, and 90 min of the reaction time ( Figure 4C,D). The 2.5 µM substrate condition resulted in increased apparent potency of pharmacological controls for both assays (2-fold for actinonin in the meprin α assay and 3-fold for NFF449 in the meprin β assay). This suggested that 2.5 µM substrate concentrations result in greater assay sensitivity. Assay QC parameters (S/B and Z') at 2.5 µM substrate concentrations did not differ significantly from assays run at 10 µM substrate concentrations; therefore, we chose 2.5 µM substrate concentrations as a final assay condition. The meprin β assay progress curve was hyperbolic rather than linear at 0.05 nM and 0.025 nM enzyme; therefore, we chose 0.0125 nM enzyme concentration where assay linearity was demonstrated ( Figure 3B). Z' and S/B values were acceptable at 60 min reaction end point (0.86 and 2.6, respectively). IC50 values of NFF449 were not significantly affected by the variations of reaction length and meprin β concentrations.
Next, we performed substrate optimization to achieve balanced assay conditions defined as [S]/KM = 1 [19]. In order to do that, we first determined kinetic parameters of proteolysis of meprin α and meprin β substrates by the respective enzymes ( Figure 4A,B). Meprin α and meprin β proteolysis exhibited similar KM values (2.4 ± 0.3 µM and 2.7 ± 0.7 µM, respectively) suggesting the need for optimization of both assays' substrate concentration. Meprin β exhibited >20-fold faster turnover of its substrate than meprin α (6.4 ± 0.06 s -1 versus 0.29 ± 0.06 s-1, respectively) which is consistent with >100-fold difference in enzyme concentrations for meprin α and meprin β assays (1.3 nM versus 0.0125 nM, respectively). To optimize substrate concentrations, both assays were run for 90 min at three different substrate concentrations (10, 5, and 2.5 µM) which included the concentration at which the assays were recapitulated in 1536 wpf (10 µM for both meprin α and meprin β) and the concentration approximating [S]/KM = 1 condition (2.5 µM). Enzyme concentrations were fixed at 1.3 nM for meprin α and 0.0125 nM for meprin β. QC parameters (Z' and S/B) and IC50 values of pharmacological controls (actinonin and NFF449) were calculated at 40, 60, and 90 min of the reaction time ( Figure 4C,D). The 2.5 µM substrate condition resulted in increased apparent potency of pharmacological controls for both assays (2-fold for actinonin in the meprin α assay and 3-fold for NFF449 in the meprin β assay). This suggested that 2.5 µM substrate concentrations result in greater assay sensitivity. Assay QC parameters (S/B and Z') at 2.5 µM substrate concentrations

Online Robotic Pilot Study
To ascertain the readiness of the assays for a large-scale screening effort, a small pilot screen was conducted using Kalypsys GNF integrated online robotic platform (San Diego, CA, USA) [20]. Overall, ~39,000 compounds were tested using 31 assay plates in both meprin α and meprin β assays. Both assays performed well on the Kalypsys robotic system, as the meprin α assay average Z' and S/B were 0.88 ± 0.03 and 2.9 ± 0.07, respectively, while the meprin β assay average Z' and S/B were 0.91 ± 0.03 and 4.5 ± 0.17, respectively. The number of hits identified in the meprin α and meprin β assays were 169 and 260, respectively, which constituted 0.43% and 0.67% hit rates, respectively. After removal of duplicates, Venn analysis showed that 37 compounds inhibited both meprins, while there were 129 compounds selectively inhibiting meprin α and 220 compounds selectively inhibiting meprin β, suggesting that selective probes for both enzymes could be discovered. This also suggested that both assays were ready for large scale effort.

Online Robotic Pilot Study
To ascertain the readiness of the assays for a large-scale screening effort, a small pilot screen was conducted using Kalypsys GNF integrated online robotic platform (San Diego, CA, USA) [20]. Overall,~39,000 compounds were tested using 31 assay plates in both meprin α and meprin β assays. Both assays performed well on the Kalypsys robotic system, as the meprin α assay average Z' and S/B were 0.88 ± 0.03 and 2.9 ± 0.07, respectively, while the meprin β assay average Z' and S/B were 0.91 ± 0.03 and 4.5 ± 0.17, respectively. The number of hits identified in the meprin α and meprin β assays were 169 and 260, respectively, which constituted 0.43% and 0.67% hit rates, respectively. After removal of duplicates, Venn analysis showed that 37 compounds inhibited both meprins, while there were 129 compounds selectively inhibiting meprin α and 220 compounds selectively inhibiting meprin β, suggesting that selective probes for both enzymes could be discovered. This also suggested that both assays were ready for large scale effort.

Primary HTS Campaign
Primary HTS campaigns were conducted using The Scripps Research Institute proprietary library of 649,570 compounds using both meprin α and meprin β assays [21]. Overall, 522 plates were used for each assay with excellent QC parameters (average Z' = 0.86 ± 0.04 and average S/B = 2.8 ± 0.09 for meprin α assay and average Z' = 0.88 ± 0.03 and average S/B = 4.4 ± 0.27 for meprin β assay). IC50 values of control compounds were reproducible with literature and our preliminary experiments (meprin α actinonin IC50 = 2.9 ± 0.12 nM, n = 11 plates; meprin β NF449 IC50 = 10.4 ± 0.85 nM, n = 11 plates). Using hit cutoffs derived from the average and 3 standard deviations of the activity of all samples tested which were 10.76% and 14.33% for the meprin α and meprin β assays, 5064 and 4929 hits were identified which constituted hit rates of 0.78% and 0.76%, respectively. It was noted that the majority of meprin α hits exhibited a percentage inhibition close to the hit cutoff, whereas meprin β hits were distributed evenly in the range of 20−100% inhibition ( Figure 5A,B).

Hit Confirmation and Prioritization
For the confirmation assays all compounds that inhibited either of the meprins with >20% inhibition were selected. Confirmation assays were done at a single concentration point in triplicate. Out of 2378 total compounds tested in confirmation assays, only 206 confirmed activity against meprin α and 1097 confirmed activity against meprin β constituting 8.7% and 46.1% confirmation rate for meprin α and meprin β, respectively. The low confirmation rate for meprin α was not unexpected due to the majority of meprin α hits from the primary campaign being close to the hit cutoff ( Figure 5A).
It was also noted that the majority of the most active hits for each enzyme were potential Zn-binders due to the presence of hydroxamate and reverse hydroxamate moieties. Compounds acting via Zn binding may be undesirable due to clinical trial failures observed previously based on a lack of selectivity, toxicity, and metabolic instability. To prioritize selectivity, we introduced additional assays to help with triaging the compounds to ascertain that we are not biasing for nonselective compounds. We utilized ADAM10, MMP-8, and MMP-14 as the most relevant counter targets. The counter screens were conducted in triplicate using the same 2378 compounds that were tested in confirmation assays. 2.9 ± 0.12 nM, n = 11 plates; meprin β NF449 IC50 = 10.4 ± 0.85 nM, n = 11 plates). Using hit cutoffs derived from the average and 3 standard deviations of the activity of all samples tested which were 10.76% and 14.33% for the meprin α and meprin β assays, 5064 and 4929 hits were identified which constituted hit rates of 0.78% and 0.76%, respectively. It was noted that the majority of meprin α hits exhibited a percentage inhibition close to the hit cutoff, whereas meprin β hits were distributed evenly in the range of 20−100% inhibition ( Figure 5A,B). inhibiting meprin α and meprin β, respectively, with ≥50% inhibition. Cheminformatics analysis of the Scripps HTS assay database containing hundreds of biological assay results showed that 660 out of 1237 confirmed hits were not promiscuous; meaning they hit in less than 5 other assays. Out of these 660 compounds 536 were meprin α active and 195 were meprin β active. Medicinal chemistry triage suggested that 289 compounds out of 536 meprin α actives were tractable, while out of 195 meprin β actives 180 were tractable, which constitutes 469 total tractable compounds. Removal of 62 duplicates left us with 407 unique compounds of which 404 were available for concentration response studies. Despite the majority of top actives from the 2378 primary HTS hits being potential Zn-binders, the hit rate in counter screens was < 2.0% ( Figure 5G,H) suggesting low metzincin promiscuity of meprin hits.
We conducted concentration response studies of 404 compounds in meprin α and β assays using 10-point 3:1 serial dilutions starting at the highest concentration of 17.4 µM in triplicate. Out of 404 tested compounds, 13 exhibited IC50 values < 1 µM and 47 < 5 µM in in both meprin α and meprin β assays.
To pick compounds for further characterization and probe development we used a cutoff of IC50 values < 10 µM against either meprin and 10-fold selectivity window for meprin α or meprin β. Additionally, we picked the top selective compounds with IC50 values < 10 µM that had no apparent Zn-binding moieties. More specifically, we prioritized selective compounds without apparent Zn-binding groups (hydroxamates, carboxylates, etc.). Using these criteria, we selected 46 compounds. Interestingly, the majority (42) were selective for meprin β and only 4 were selective for meprin α. These 46 potentially non-Zn-binding compounds were clustered in [21] distinct scaffolds. The most populated scaffold had 9 members suggesting its amenability to medicinal chemistry.
The second group of compounds was chosen based on selectivity between main target (either meprin α or β) and four other tested metzincins (either meprin α or meprin β, ADAM10, MMP-8, and MMP-14) and potency towards the main target (either meprin α or meprin β) regardless of the presence of Zn binders. These criteria yielded 41 compounds belonging to 17 distinct clusters. Interestingly, the majority (32) were selective for meprin α and only 9 were selective for meprin β, which is the opposite trend from non-Zn-binders.
We also tested representative compounds from each scaffold for effects on skin fibroblast and melanocyte viability to ascertain cytotoxicity towards various skin cell types.
Overall, hits showed either no or very little effect on cell viability ( Figure 8) suggesting a lack of general cytotoxicity and amenability of hit chemotypes for the development into in vitro probe for biological studies.  Figure 7. Results of concentration response studies of top potent and selective meprin β inhibitors.  Figure 7. Results of concentration response studies of top potent and selective meprin β inhibitors.  Figure 7. Results of concentration response studies of top potent and selective meprin β inhibitors.  Figure 7. Results of concentration response studies of top potent and selective meprin β inhibitors. We also tested representative compounds from each scaffold for effects on skin fibroblast and melanocyte viability to ascertain cytotoxicity towards various skin cell types. Overall, hits showed either no or very little effect on cell viability ( Figure 8) suggesting a lack of general cytotoxicity and amenability of hit chemotypes for the development into in vitro probe for biological studies.  We also tested representative compounds from each scaffold for effects on skin fibroblast and melanocyte viability to ascertain cytotoxicity towards various skin cell types. Overall, hits showed either no or very little effect on cell viability ( Figure 8) suggesting a lack of general cytotoxicity and amenability of hit chemotypes for the development into in vitro probe for biological studies.  We also tested representative compounds from each scaffold for effects on skin fibroblast and melanocyte viability to ascertain cytotoxicity towards various skin cell types. Overall, hits showed either no or very little effect on cell viability ( Figure 8) suggesting a

Discussion
As the result of the uHTS effort we discovered and characterized several novel scaffolds with activity against meprin α and meprin β. All top selective meprin α HTS hits contain a hydroxamate moiety, whereas meprin β hits lack one. Based on the presence of

Discussion
As the result of the uHTS effort we discovered and characterized several novel scaffolds with activity against meprin α and meprin β. All top selective meprin α HTS hits contain a hydroxamate moiety, whereas meprin β hits lack one. Based on the presence of the hydroxamate moiety in meprin α inhibitors it is likely that they act via binding of the active site zinc atom as was demonstrated for numerous other metzincins. Tan et al. [17] proposed the interaction model whereby the hydroxamic moiety of an analog of compounds 10d and 10e (Figure 1) binds zinc and carboxylate moieties interact with residues of the S1 and S1 subsites. Based on this model, the selectivity of 10d and 10e is derived from differences between meprin α and meprin β S1 and S1 subsites. Both 10d and 10e structures are symmetric with a central hydroxamate moiety connected via propyl linkers to either terminal benzodioxanes or benzodioxols. Our HTS hits are unlikely to interact with both subsites as the hydroxamate is terminal in all cases. Similar to 10d and 10e, most of the hits ( Table 2) have at least one other electronegative moiety in addition to the hydroxamate that could be interacting with positively charged residues in either subsite of meprin α. However, only five (SR19855, SR1596857, SR220670, SR162799, and SR162808) out of nine hits show selectivity for meprin α suggesting that additional interactions may be responsible for selectivity against meprin β.
The most selective and potent meprin α HTS hit, SR162808, exhibited more than 30-fold selectivity against meprin β and other metzincins ( Table 2) and no cytotoxicity ( Figure 8B). For comparison, 10d and 10e exhibit 18-fold and 19-fold selectivity, respectively (Table 4). Unfortunately, nothing has been reported about their effects on cell viability. For in vivo probe or drug lead development significant selectivity and toxicity windows are extremely important; therefore, SR162808 represents a good starting point for a medicinal chemistry optimization effort. The HTS-based approach to metalloproteinase or any other inhibitor discovery has inherent limitations and strengths. The strength of HTS is in its ability to assess multiple chemical scaffolds for activity against the target of choice and the ability to select the most promising scaffolds for further optimization. In the event that the chosen scaffold is not amenable to optimization, the researchers can go back to the HTS campaign results and choose additional scaffolds. The main limitation of the HTS approach is that it relies heavily on the composition of the HTS library that is being screened. In our case, we found several novel scaffolds active and selective for both meprins, however, most of these scaffolds contain zinc-binding moieties such as hydroxamates. While hydroxamates have good binding affinity to zinc of the active site, they have their limitations (reviewed in [22]). More specifically, the zinc-binding property of hydroxamates can lead to off-target interactions with other members of metzincin superfamily (e.g., adamalysins, matrixins), unfavorable pharmacokinetic properties, dose-limiting toxicity and metabolic instability, to name a few. However, despite these limitations, there are multiple examples of hydroxamates being used in the clinic (reviewed in [23]). Examples include, but are not limited to histone deacetylase inhibitors (HDACI) panobinostat and bellinostat, deferoxamine, a chelating agent used to treat iron or aluminum toxicity [24].