Structure-based design of allosteric calpain-1 inhibitors populating a novel bioactivity space

(S) binders that populate a novel chemical space. We have identi ﬁ ed compound 1 , Vidupiprant, which is shown to bind to the PEF(S) domain by the TNS displacement method, and it exhibited speci ﬁ city in its allosteric mode of inhibition. Compound 1 inhibited the full-length calpain-1 complex with a higher potency (IC 50 ¼ 7.5 m M) than the selective, cell-permeable non-peptide calpain inhibitor, PD150606 (IC 50 ¼ 19.3 m M), where the latter also inhibited the active site domain in the absence of PEF(S) (IC 50 ¼ 17.8 m M). Hence the method presented here has identi ﬁ ed known compounds with a novel allosteric mechanism for the inhibition of calpain-1. We show for the ﬁ rst time that the inhibition of enzyme activity can be attributed to an allosteric mode of action, which may offer improved selectivity and a reduced side-effects pro ﬁ le. © 2018 The Authors. Published by Elsevier Masson SAS. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

P u blis h e r s p a g e : h t t p:// dx.doi.o r g/ 1 0. 1 0 1 6/j.ej m e c h. 2 0 1 8. 0 8. 0 4 9 < h t t p:// dx.doi.o r g/ 1 0. 1 0 1 6/j.ejm e c h. S e e h t t p://o r c a .cf. a c. u k/ p olici e s. h t ml fo r u s a g e p olici e s.Co py ri g h t a n d m o r al ri g h t s fo r p u blic a tio n s m a d e a v ail a bl e in ORCA a r e r e t ai n e d by t h e c o py ri g h t h ol d e r s .

Introduction
Calpains are proteins that belong to the family of calciumdependent, non-lysosomal cysteine proteases expressed ubiquitously in mammals and other organisms [1].Although the physiological roles of calpains are still poorly understood, they have been shown to be involved in many processes such as cell motility, longterm potentiation in neurons and cell fusion in myoblasts [2].In particular, conventional dimeric calpains have been reported to be involved in the cell degeneration processes that characterize numerous disease conditions [3].
Calpain-1 (m-calpain) and calpain-2 (m-calpain) are heterodimeric proteases composed of a large subunit with a molecular mass of ~80 kDa, sharing a small subunit of mass ~30 kDa.The small Abbreviations: PEF(S), penta-EF hand calcium binding domain; TNS, p-toluidinylnaphthalene-6-sulfonate; CysPc, calpain-like protease domain; kDa, kilo Dalton; SAR, structure activity relationship; NR, no response; CNS, central nervous system; HBD, hydrogen bond donor; HBA, hydrogen bond acceptor; tPSA, topological polar surface area; MW, molecular weight.subunit consists of two domains, namely the penta-EF hand calcium binding PEF(S) and glycine rich (GR) domains, which are essential for stabilizing calpain-1 and calpain-2 [4].High sequence similarity of 62% is exhibited by the large subunits of calpain-1 and -2 in humans [4].And they consist of four different domains, an Nterminal anchor helix, the active site domain (CysPc), a domain that resembles the C2 membrane binding domains of phosphokinases, known as the C2L domain, and a second penta-EF hand calcium binding domain known as PEF(L).The PEF(L) domain is the determinant of the calcium concentration that is required for protease activation, which differentiates between the two isoforms: calpain-1 requires micromolar concentrations of Ca 2þ , whereas calpain-2 requires millimolar concentrations for activation [5].
Calpain-1 is a target dysregulated in many diseases such as neurodegenerative disorders, cardiovascular diseases, ischaemic disorders, arterial sclerosis, leishmaniasis and cancer [6,7].In most cases of disease, calpain-1 activity is elevated (and hence its inhibition would be beneficial in treatment).However, it has been recently suggested that the up regulation of calpain-1 appears to be beneficial in some cases, such as stage II Alzheimer's disease, where its activation could be neuroprotective and beneficial in controlling cellular damage [8].
Until recently, it has been challenging to design selective calpain-1 inhibitors, and this is attributed to the fact that most compounds that target the active site inhibit a broad spectrum of cysteine proteases, thereby resulting in undesirable side effects [9].For example, it has been previously reported that calpain-1 inhibitors [9], which also inhibit the proteasome may induce apoptosis, whereas selective calpain-1 inhibitors do not.Hence, it may be beneficial to design selective calpain-1 inhibitors to avoid off-target related side effects.
Classical allosteric inhibitors of calpain-1, which were originally reported to bind to PEF(S), exhibit a very specific type of chemistrythat is a-mercaptoacrylic acid-based, such as PD150606 and PD151746 [10].These inhibitors are potent, cell permeable and selective inhibitors of calpain-1 and calpain-2 exhibiting selectivity towards calpain over other cysteine proteases, with a slight selectivity for calpain-1 over calpain-2.It has been reported however that PD150606 could equally inhibit the active site domain of calpain-1 without the presence of PEF(S) [11].This obviously suggests that its mode of action is rather unclear.The reported amercaptoacrylic acid based calpain inhibitors (Fig. 1. A. PD150606 and B. PD151746) and their disulfide analogues (Fig. 1. C. and D.) were synthesized by Adams et al., and shown to bind to PEF(S) from X-ray diffraction analysis (PDB IDs: 1NX3, 4WQ2 and 4WQ3) [5,12e14].Additional calpain-1 inhibitors that were reported to inhibit the whole calpain-1 complex, which consists of the PEF(S) and CysPc (active site domain of calpain-1), are also depicted in Fig. 1, including their chemical structures.CHEMBL203568 [15], shown in Fig. 1.E., is a compound reported to inhibit the calpain-1 complex with an IC 50 value of 4.9 nM.While CHEMBL204883 [15], shown in Fig. 1.F., is a compound reported to inhibit the calpain-1 complex with an IC 50 value of 8 nM.However, CHEMBL203568 and CHEMBL204883 have not been confirmed as PEF(S) binders i.e. exhibit an allosteric mode of inhibition.Their reported confidence score is 7 (in ChEMBL) indicating that these compounds might be binding to any of the subunits involved in the full-length calpain-1 complex.Hence, the use of PEF(S) (calpain-1 small subunit) in structure-based virtual screening may be an appealing approach for the design of allosteric calpain-1 binders with completely different structural architecture from the classical allosteric binders and inhibitors.In addition, this approach is expected to answer the question of whether PEF(S) binding would confer inhibition, given that the shortlisting of candidates is based on the prediction of their binding to PEF(S).
In this work, we have used the PEF(S) (PDB ID: 4WQ2, calpain-1 small subunit) in a structure-based virtual screening protocol to ascertain whether novel chemical series bind to the allosteric pocket.To validate this approach, purchasable ligands of diverse and novel structural frameworks (very different from those that  D. respectively, which bind to PEF(S) and were reported [5] with improved potencies in comparison to their monomer compounds E. CHEMBL203568, a compound reported to inhibit the calpain-1 complex (Uniprot IDs: P04632 and P07384) with an IC 50 value of 4.9 nM and confidence score of 7 F. CHEMBL204883, a compound reported to inhibit the calpain-1 complex (Uniprot IDs: P04632 and P07384) with an IC 50 value of 8 nM and confidence score of 7.
have been previously investigated) were evaluated in silico, using ligand/protein docking against PEF(S), and the compounds were subsequently tested in relevant assays.This pipeline for the structure-based virtual screening protocol is depicted in Fig. 2.
As a general approach, we hypothesised that the functional effect of PEF(S) binders on the active site of calpain-1 may be predicted by carrying out Molecular Dynamics (MD) simulations [16,17] on the fulllength calpain-1 complex (PEF(S) and CysPc) in both the unbound and the ligand bound states to the PEF(S) domain.If the bound compound increases the average distance between the substrate and the interacting residues in the active site of calpain-1, then we further hypothesise that it would inhibit the activity of the enzyme.In contrast, if the compound decreases this distance, it would favour the interactions between the enzyme and the substrate, thus facilitating the enzyme reaction.A crystal structure of the full-length calpain-1 complex (PEF(S) and CysPc) is currently unavailable, and hence we approached the problem by using the pipeline shown in Fig. 2. Accordingly, the compounds that are predicted to bind to PEF(S) are either expected to inhibit or activate calpain-1.
To explore the small molecule chemical architecture that would most likely alter the geometry of the calpain-1 active site, to inhibit its substrate cleavage allosterically, compounds with diverse chemical structures were investigated computationally based on their predicted binding affinities towards PEF(S).The candidate PEF(S) binders, which were shortlisted by the structure-based virtual screening protocol depicted in Fig. 2 were sulphonamides, N-{3-[3-(2-alkoxyethoxy)-5-(4-methoxyphenyl)s, and [1,2,4]triazolo [4,3-b]pyridazin-6-yl]pyridines.Experimentally these compounds were also shown to bind to PEF(S) by displacing TNS (which binds to the allosteric site) [10].In addition, three compounds were able to inhibit the full-length calpain-1 complex (which includes PEF(S) and CysPc) allosterically, but not the active site domain of calpain-1 in the absence of PEF(S).The inhibitory activity, via a proposed allosteric mechanism is a crucial finding given that they show specificity in their mode of action, which is not the case for the classical allosteric inhibitors such as PD150606.The new inhibitors possess different scaffolds from the classical allosteric inhibitors.These compounds serve as a novel starting point for the expansion of the compound series (including SAR) to improve their potency.In addition, this finding suggests that compounds can inhibit the enzyme activity via the PEF(S) domain.An important aspect of this study is that by designing allosteric inhibitors, which do not inhibit the active site domain (that is common to a wide variety of cysteine proteases), these may be effective in treating calpain-1 related diseases without the side effects associated with inhibitors, which do inhibit the active site domain as well [18].

Structure-based virtual screening of purchasable ligands against PEF(S)
36,503 commercial compounds consisting of diverse chemical structures (sulphonamide-, amide-, pyridine-, urea-, enaminebased compounds), were docked using Glide [19] into the preprepared (see methods for details) protein crystal structure of human PEF(S) (PDB [20]ID: 4WQ2) [5].From the docking scores, the distribution for actives versus inactives was obtained.The active molecules displayed a more favourable distribution of scores, which allowed differentiation of actives and inactives (see methods for details).
Candidate PEF(S) binders from the purchasable database were shortlisted on the basis of a cut-off with the highest Mathews Correlation coefficient.The cut-off obtained was À6. 35, according to which compounds with more negative binding score were predicted to bind.The selected candidates were further screened against PAINs [21] (with regard to the recent analysis of the use of this approach (Tropsha)) [22] using the FAFDrug3 ADME-Tox Filtering Tool [23].Those compounds that didn't exhibit any potential PAINs liability were considered for evaluation of calpain-1 activity.As a result, five sulphonamides 1e5, two substituted N-{3-[3-(2-alkoxyethoxy)-5-(4-methoxyphenyl)s 6e7, and three substituted [1,2,4]triazolo [4,3b]pyridazin-6-yl]pyridines 8e10, which were the top ranked compounds according to the virtual screening criteria, were shortlisted as candidates for PEF(S) binding.

MDS plot shows that shortlisted PEF(S) binders occupy a novel region in chemical space
An MDS plot of the chemical space was generated consisting of Fig. 2. The pipeline of the structure-based virtual screening protocol followed for shortlisting candidates of PEF(S) binders started with the collection of a pool of compounds with diverse chemical structures (very different from those that have been previously investigated), then candidates were shortlisted based on docking into the crystal structure of PEF(S) (PDB: 4WQ2, calpain-1 small subunit).The top ranked candidates were assessed for their chemical novelty in comparison to the classical allosteric binders and inhibitors using an MDS plot.The subset of compounds was then investigated using relevant experimental assays.
the ChEMBL compounds inhibiting the full-length calpain-1 complex (PEF(S) and CysPc) with IC 50 values 1 mM, and the Adams library (a library of a-mercaptoacrylic acid calpain inhibitors and their disulfide analogues) [5], which were validated against the PEF(S) docking model (see methods for details) (Fig. 3).In addition, the shortlisted candidates 1e10 from the structure-based design protocol were also included in the plot, which enabled the assessment of the novelty of chemical space coverage of the shortlisted PEF(S) binders, in comparison to the classical calpain-1 inhibitors and PEF(S) binders.Fig. 3, which is a two dimensional MDS plot based on Morgan fingerprints of radius 2 with 90% ellipse-like confidence regions, shows that the shortlisted compounds 1e10 exhibited new structures in comparison to the previously reported compounds by occupying a novel region in the chemical space of calpain-1 actives.

FRET based inhibition assay
A fluorogenic assay of the full-length calpain-1 complex (which includes PEF(S) and CysPc) was used to determine the activity of compounds 1e10.Amongst the identified compounds, 1, 9,a nd10 inhibited the full-length calpain-1 complex with IC 50 values of 7.5 (±1.1), 20.5 (±1.9), and 29.7 (±5.2) mM, respectively (Table 1).The same experimental protocol was performed to measure the activity of compounds against the active site domain of calpain-1, without the presence of PEF(S), to investigate a possible allosteric mode of action.None of the compounds showed any inhibition in the absence of PEF(S), except for compound 1, which weakly inhibited the active site domain with an IC 50 value >100 mM.In contrast, compound 3 exhibited higher inhibitory activity against the active site domain of calpain-1 Fig. 3.A two dimensional MDS plot based on Morgan fingerprints of radius 2, with 90% ellipse-like confidence regions for 32 ChEMBL compounds, 33 compounds synthesized by Adams et al., [5] and the 10 shortlisted candidates from the structure-based design protocol.These were shortlisted from a purchasable database based on their high predicted binding affinities towards the PEF(S) domain of calpain via docking with Glide.The compounds identified belong to a new chemical space in comparison to the previously reported compounds that bind to the PEF(S) (Adams library) and the compounds which inhibit the full-length calpain-1 complex.Compounds 1, 9, and 10 were then validated as novel allosteric inhibitors of calpain-1.
with an IC 50 value of 41.1 (±15.4) mM as compared to the fulllength calpain-1 complex, which showed an activity of >100 mM, suggesting that in the presence of PEF(S) it preferentially binds to PEF(S).This could explain the reduction in its inhibitory activity since it is most likely unable to alter the geometry of the active site while it binds allosterically.The IC 50 values were also measured for the classical a-mercaptoacrylic acid based calpain inhibitor, PD150606, and these were 19.3 (±1.6) mM with the full-length calpain-1 complex, and 17.8 (±2.4) mM with the active site domain without the presence of PEF(S).
Hence, in contrast to compounds 1, 3, 9,a n d10, PD150606 exhibited an unspecific mode of action by equally inhibiting via both binding sites (active and allosteric sites).It is worth stating here that compound 1 is an asthma drug, Vidupiprant or AMG 853 [24], which exhibited higher potency than PD150606 in inhibiting the enzyme activity.The dose response curve of compound 1 (IC 50 ¼ 7.5 ± 1.1 mM) is displayed in Figure S1.
Interestingly, identifying Vidupiprant as an allosteric inhibitor of calpain-1 is in agreement with previous reports showing a direct link between calpain inhibition and anti-inflammatory properties, where it was shown that non-steroidal anti-inflammatory drugs (NSAIDs) inhibit calpain, and that calpain inhibition reduces allergic inflammation [25,26].Potentially, this finding could highlight the importance of considering calpain inhibitors for the development of new anti-asthmatic therapies.
The allosteric inhibitory activity exhibited by compounds 1, 9, and 10 validates the design approach, which shortlisted PEF(S) binders.The three active compounds include scaffolds, which are distinct from the classical allosteric inhibitors.The IC 50 values obtained could be improved by efficient choice of substituents using standard medicinal chemistry approaches.In particular compound 1, which is a sulphonamide, exhibited specificity in its allosteric inhibition of calpain-1, and was more potent (7.5 mM) than PD150606 (19.3 mM), which additionally inhibits the active site domain, in the absence of PEF(S), with a similar IC 50 value.

TNS displacement assay
TNS, which is a sensitive fluorophore that binds to PEF(S) was used to probe protein dynamics and conformational change.It fluoresces in the bound state i.e. in a hydrophobic environment, whereas when another compound displaces it, its fluorescence gets quenched.This fluorophore was used previously to assess the binding of PD150606 to PEF(S), a compound which has been already shown to bind to PEF(S) by X-ray crystallography (PDB ID: 1NX3) [10,12].In this work, PD150606 (as a control), compound 1, the most potent allosteric inhibitor, compound 3, a weak allosteric inhibitor, and compounds 2, 4 and 5, which did not exhibit any inhibitory activity, have been tested for PEF(S) binding by the TNS displacement method.The results for compounds 6e10 were unreliable as it became apparent that they fluoresced under the assay conditions, therefore for this reason these results were omitted.As shown in Fig. 4, all tested compounds, except for compound 5, quenched the fluorescence of TNS, exhibiting a similar trend in their quenching effect to that of the known PEF(S) binder, PD150606, hence confirming their binding to PEF(S).Therefore, it appears that compounds 2e4 do indeed bind to PEF(S) in a similar fashion to compound 1, but they either weakly inhibited or failed to allosterically alter the geometry of the active site.

Analysis of molecular docking studies of representative calpain-1 inhibitors 1 and 10, and compounds 2e5
Docking studies predicted molecular interactions of the sulphonamide 1 and the [1,2,4]triazolo [4,3-b]pyridazin-6-yl] pyridine 10 with the PEF(S) protein crystal structure (PDB ID: 4WQ2).Fig. 5. A. shows the 2-chloro-4-cyclopropylsulfonamido phenyl ring of compound 1 is p-stacked with His 131 , and the carbonyl of its carboxylic acid moiety H-bonds with the same residue, the carbonyl of its amide moiety H-bonds with Trp 168 , and the phenyl ring attached to the tert-butylcarbamoyl moiety is p-stacked with the same residue.Fig. 5. B. shows a p-stacking interaction between the pyridine ring of compound 10 and Trp 168 , and H-bonding of the nitrogen in that ring with the same residue.The hydrophobic interactions predicted for compounds 1 and 10 with Trp 168 are also seen in the co-crystallised ligand/protein crystal structure (PDB ID: 4WQ3) [5].In addition, a more favourable binding affinity towards PEF(S) was predicted for compound 1 (IC 50 ¼ 7.5 ± 1.1 mM) as compared to compound 10, (IC 50 ¼ 29.7 ± 5.2 mM), which could be the reason for the higher inhibitory activity exhibited by compound 1.In order to further explore the activities of compounds 1e10, MD simulations on the full-length calpain-1 complex, which includes the PEF(S) domain would be beneficial once the crystal structure is available.The inhibitory activity of each compound is predicted to correlate with the average distance between the substrate and the interacting residues in the active site of calpain-1.
Docking studies predicted molecular interactions of the sulphonamides 2e5 with the PEF(S) protein crystal structure (PDB ID: 4WQ2), which suggest that the hydrophobic interactions with Trp 168 are essential for PEF(S) binding.Fig. 6. A. shows that the phenyl ring of compound 2, which is attached to the sulfonyl moiety is p-stacked with Trp 168 ,a n dt h es a m er e s i d u eH -b o n d s with the carbonyl of the amide group, the hydroxyl of its cyclohexyl ring H-bonds with Gln 100 , and its sulfonyl group H-bonds with His 131 .Fig. 6.B. displays the predicted molecular interactions between compound 3 and the PEF(S) crystal structure, and these are H-bonding between its carboxyl moiety and the Lys 172 and Trp 168 residues, H-bonding between its sulfonyl group and His 131 , and p-stacking with the Trp 168 via its pyridine ring.Fig. 6.C .demonstrates the predicted molecular interactions for compound 4, the amino group of its amide moiety H-bonds with the Glu 97 and Trp 168 residues, the phenyl ring of its benzamide group pstacks with Trp 168 , and the carbonyl of its amide moiety H-bonds with Lys 172 .Fig. 6.D. shows the H-bonding interactions predicted for the carbonyl of the oxopyrrolidin-1-yl moiety in compound 5 and the Lys 172 and Trp 168 residues, and H-bonding interactions for its sulfonyl moiety with His 131 ,andp-stacking of the same residue with the with the aromatic ring of its trifluoromethylphenyl moiety.Hence, the hydrophobic interactions predicted for compounds 1e4 with Trp 168 are also seen in the co-crystallised ligand/ protein crystal structure (PDB ID: 4WQ3) [5].However, compound 5 was predicted to exhibit hydrophilic interactions with Trp 168 , which could explain why it didn't displace TNS, suggesting that the hydrophobic interactions with this residue are essential for PEF(S) binding.

Computational assessment of CNS permeability for representative calpain-1 inhibitors 1 and 10
Given that calpain-1 may serve as a therapeutic target for neurodegenerative disorders, the computational assessment of the CNS permeability for compounds 1 and 10 was performed with FAFDrug3 [23] to see whether these compounds could be considered as good starting points to target these diseases.Their physicochemical properties were calculated using FAFDrug3.Following this, CNS diagrams were obtained and are presented in Figure S2.AandB.Compound1 did not pass the CNS filter, which takes into consideration the assessment of its ability to pass the Table 1 IC 50 values for compounds 1e10, and PD150606 determined by the FRET based inhibition assay, with the full-length calpain-1 complex and the active site domain of calpain-1, reported as mean ± standard deviations from three independent experiments (NR ¼ no response).blood brain barrier.Hence, it is predicted not to exhibit the desired permeability [27], since the values of all the descriptors for compound 1 (logP, HBD, HBA, MW, tPSA (blue line)) fall outside the CNS filter area (light blue).As for compound 10,i ti s predicted to exhibit medium permeability since all the descriptors, except for the HBA, have passed the CNS filter.Hence, compound 10 might serve as a good starting point for analogue development to target neurodegenerative diseases.

Conclusions
In this work, we have successfully validated our structurebased design method, which has led to the discovery of chemically novel allosteric inhibitors of calpain-1, and we have shown for the first time an allosteric mode of action.Compounds 1, 9, and 10 inhibited the full length calpain-1 complex (which includes PEF(S) and CysPc) with IC 50 values of 7.5 (±1.1), 20.5 Fig. 4. PD150606, which was shown by X-ray crystallography to bind to PEF(S) (PDB ID: 1NX3), compound 1 (the most potent allosteric inhibitor), 3, which was the least potent among the identified allosteric inhibitors, 2 and 4, which did not exhibit any inhibitory activity, all quenched the fluorescence of TNS.All compounds showed a similar effect to that of PD150606 confirming their binding to PEF(S), except for compound 5, which neither exhibited any inhibitory activity nor displaced TNS.
(±1.9), and 29.7 (±5.2) mM respectively.Compounds 9 and 10 did not inhibit the active site domain of calpain-1 in the absence of PEF(S), and compound 1 inhibited the active site domain weakly with an IC 50 value >100 mM.In contrast, compound 3 exhibited higher inhibitory activity against the active site domain of calpain-1 with an IC 50 value of 41.1 (±15.4) mM as compared to the full-length calpain-1 complex, which was >100 mM suggesting that it preferentially binds to PEF(S).In addition, the IC 50 values were measured for PD150606, giving 19.3 (±1.6) mMwith the full-length calpain-1 complex and 17.8 (±2.4) mMw i t ht h e  active site domain (without the presence of PEF(S)).In comparison to the classical a-mercaptoacrylic acid based calpain inhibitor, PD150606, compounds 1, 9,a n d10, exhibited specificity in their allosteric mode of action, since they didn't inhibit the active site domain in the absence of PEF(S).
Furthermore, PD150606, compound 1, the most potent allosteric inhibitor, compound 3, a weak allosteric inhibitor, and compounds 2 and 4, (which did not exhibit any inhibitory activity) have been tested for PEF(S) binding by the TNS displacement method.All compounds (1e4) quenched the fluorescence of TNS, exhibiting a similar trend in their quenching effect to that of the known PEF(S) binder, PD150606.
IC 50 values obtained for compounds 1, 9, and 10 may be good starting points for optimisation, having novel scaffolds.Allosteric inhibitors discovered by this approach could exhibit more selectivity towards calpain-1 since they are unlikely to inhibit the active site domain, which is similar for a wide variety of cysteine proteases.This could translate to more effective treatments with less side effects for calpain-1 related diseases [18].

Ligand preparation
The entire set of extracted ligands were prepared for docking with LigPrep 2.5 [29] using the default settings and the Epik option, which introduces energy penalties associated with ionization and tautomerization [30].

Receptor preparation of PEF(S)
Docking with Glide [19] was performed against human PEF(S), calpain-1 small subunit (regulatory subunit) of the protein crystal structure (PDB [20] ID: 4WQ2) [5] bound to (Z)-3-(6-bromondol-3yl)-2-mercaptoacrylic acid.The structure was prepared using the protein preparation wizard of Maestro 9.3 [29], following the default protocol, which accounts for energy refinement, hydrogen addition, pK a assignment, and side-chain rotational isomer refinement.Resolved water molecules were discarded and the structure was centred using the co-crystallised ligand as the centre of the receptor grid generated for the protein structure.The co-crystal structure of the human calpain PEF(S) protein crystal structure (PDB ID: 4WQ2) bound to (Z)-3-(6-bromondol-3-yl)-2mercaptoacrylic acid was selected as the target structure.

Cut-off generation for compound selection from docking model
In an attempt to validate the docking model, a set of known active and inactive compounds were docked against the PEF(S) protein crystal structure to ensure that it enriched actives.32 ChEMBL [31] compounds with IC 50 values 1 mM (protein complex of the calpain-1, catalytic and small regulatory subunits: P07384, P04632 with confidence scores of 6 or 7) were docked against the PEF(S) model.In addition, 20 inactive compounds from PubChem [32] of the PEF(S) calpain-1 small regulatory subunit (Uniprot ID: P04632) were docked.
A good separation was obtained for the medians of docking score distribution for actives versus inactives for the docking model indicating that the actives are enriched.Figure S3.shows the separation of the medians for the PEF(S) docking model, À7.48 (actives) vs. À4.60 (inactives).In addition, the indole and the phenyl amercaptoacrylic acid-based inhibitors and their disulfide analogues, which were synthesized by Adams et al., [5] and shown by X-ray crystallography to bind to PEF(S), were docked against the PEF(S) docking model.The median of the docking score distribution obtained is À7.25 in comparison to the median of inactives, which is À4.60 (Figure S4., supporting information), which further validates that the model enriches these set of actives.Mann-Whitney test, which included statistical analysis on the active and inactive docking score distributions, was performed with R [33] using the script provided by Kalash et al. [34] The differences in the medians was significant with p values less than 0.05.
The Matthews correlation coefficient (MCC), which takes into account true and false positives and negatives, was computed using a Python script [34] for all the docking scores of the ChEMBL actives and PubChem inactives for the model.A search was performed for a docking score threshold that gave the highest MCC for the docking model in order to shortlist purchasable candidates of PEF(S) binders, which displayed docking scores that are lower than that with the highest MCC for the PEF(S) docking model (which was À6.35).

Docking
The purchasable compounds, prepared according to the protocol described, were docked against the PEF(S) protein crystal structure (PDB ID: 4WQ2).The Glide docking parameters included extra precision (XP) and the flexible ligand sampling option.These were deduced from docking experiments using known actives and inactives against the protein model.The highest ranked compounds with respect to predicted affinities towards PEF(S) were selected for binding assessment and calpain-1 activity evaluation i.e. those which displayed docking scores that are lower than that with the highest MCC for the PEF(S) docking model, which was À6. 35.The compounds that did not exhibit potential PAINs [21,22] liabilities upon virtual screening with the FAFDrug3 ADME-Tox filtering tool, were selected for experimental validation [23].The shortlisted compounds exhibited diverse structures, with five sulphonamides 1e5, two substituted N-{3-[3-(2-alkoxyethoxy)-5-(4-methoxyphenyl)s 6e7, and three substituted [1,2,4]triazolo [4,3-b]pyridazin-6-yl]pyridines 8e10.

Multi-dimensional scaling (MDS) analysis of the shortlisted compounds 1-10
This step of the analysis aimed to plot the chemical space of ChEMBL compounds, which are active against the full-length calpain-1 complex, with IC 50 values 1 mM (protein complex of calpain-1, catalytic and small regulatory subunits: P07384, P04632 and confidence scores of 6 or 7), and the Adams library [5], which were validated against the PEF(S) docking model.In addition, the shortlisted candidates from the structure-based design protocol were included in the same plot.This analysis enables an assessment of the novelty of the chemical space coverage of the shortlisted compounds 1e10.For this purpose, the SMILES of all compounds involved were standardized using the ChemAxon Command-Line Standardizer, where the following options were selected: "Remove Fragment" (keep largest), "Neutralize", "Remov-eExplicitH", "Clean2D", "Mesomerize" and "Tautomerize" [35].
Subsequently, Morgan fingerprints (radius 2, 1024 bits) were generated for all the compounds using KNIME 2.11.3 [36] A 2D-similarity matrix based on Euclidean distance was computed with R, using the generated Morgan fingerprints.Then, a metric multidimensional scaling of the similarity matrix was computed by embedding it into two dimensions (k ¼ 2).Subsequently, a 2D-plot was obtained, with the x-axis and the y-axis labelled, Dimension 1 and Dimension 2 respectively.These are two relative and unit-less dimensions that recapitulate the pairwise similarity of all points observed in the distribution of Euclidean distances in a lower dimensional space.For each data set in the plot (corresponding to Adams library, ChEMBL compounds, and the shortlisted compounds), 90% confidence ellipses were computed using the ellipse package.Finally, based on the generated Morgan fingerprints, an MDS plot with 90% ellipse-like confidence regions was obtained using the R ggplot2 package [37].

Expression and purification of PEF(S)
The codon optimised gene encoding human PEF(S) was purchased from Epoch Biolabs (Texas, USA) in a pET21d vector.Human PEF(S) was produced in E. coli BL21-CodonPlus (DE3)-RP (Agilent Technologies), and purified using the same procedure previously described for PEF(S) [5].

Evaluating calpain-1 activity of the shortlisted candidates of PEF(S) binders 1e10
This assay uses a fluorogenic peptide from the calpain-1 substrate a-spectrin, containing a FAM-DABCYL FRET pair (H 2 NeK(-FAM)-EVYGMMK(DABCYL)eOH).Cleavage by calpain-1 occurs between the Tyr-Gly residues and results in enhanced fluorescence as the quenching effect is relieved.The assays using purified porcine calpain-1 (CalBiochem, 25 nM) were performed in a buffer containing 1 mM calpain-1 substrate, 10 mM HEPES, 10 mM DTT, 0.5 mM EDTA, bovine serum albumin (0.1%) pH 6.8.The assay was carried out using a fluorescent plate reader (BMG Optistar) with a final assay volume of 100 ml at a temperature of 37 C, using an excitation band pass filter centred at 490 nm and emission detected at 520 nm.The compounds were added to the assay mixture before the reaction was initiated by the addition of CaCl 2 (5 mM).None of the compounds had significant fluorescence at this wavelength.The compounds were dissolved in DMSO at 40 mM and diluted into assay buffer to give range of concentrations from 5 nM to 200 mM.
In each assay run, the effect of DMSO alone over the concentration used was also measured.Although there was no effect of DMSO at lower concentrations, in some assay runs, DMSO at 0.005%e0.5% produced some inhibitory effect.This DMSO effect (which was only relevant for compounds with poor inhibitory ability) was subtracted before constructing the inhibition curves [5].The IC 50 values were obtained by fitting the data with non-linear regression with the SigmaPlot software [38], and the reported results are the mean ± standard deviation of three independent experiments.The plates were then incubated in a FLUOstar Omega plate reader at 25 C for 5 min then were analysed using an excitation wavelength of 355 nm and an emission wavelength of 450 nm with 10 flashes per well with orbital averaging.
The baseline (just TNS) was subtracted from the fluorescence (B).B was then subtracted from the B max (with no inhibitor) to invert the data.The data was plotted in excel as bar diagrams for the mean ± standard deviation of three independent experiments (n ¼ 3) in triplicate as shown in Fig. 4. Whereas for compounds 6e10, their fluorescence interfered with TNS (refer to Figure S5.for the blanks obtained for compounds 1e10), for this reason their results were omitted.
Compounds and reagents Compound 1 was purchased from Tocris, and compounds 2e10 were purchased from Ambinter, and used without further purification.PD150606 was purchased from Sigma-Aldrich and used without further purification.

Cloning of CysPC gene
The pET28a-GB1-MAAKLVFF plasmid, a kind gift from Dr Cornelius Krasel of the Institut für Pharmakologie (Marburg, Germany), was used to create a Golden Gate acceptor plasmid by standard PCR, overlap extension PCR, endonuclease digestion and T4 DNA ligase reactions.When subjected to Golden Gate digestion/ ligation with BsaI and T4 DNA ligase with an appropriate complimentary PCR product, the resulting plasmid has a section of DNA encoding RF under a constitutive promotor removed and the PCR product is incorporated in such a fashion that the translation product contains the protein of interest with an N-terminal hexahistidine tag GB1 fusion tag that can be removed by the action of tobacco etch virus protease (TEV).
Digestion/ligation proceeds as expected despite the presence of an additional BsaI site within the sequence encoding RFP.The CysPC domain of calpain-1 was inserted into this acceptor by PCR with the following primers: Fwd: CGACTAGTGGTCTCCAGTCCATGGGTCGCCATGAGAA Rev: CGACTAGTGGTCTCCATCAGTCCGGGGTCAGGTTACA for ligation using the golden gate protocol into the pET28a-GB1-MAAKLVFF plasmid [39].

Expression and purification of CysPC
BL21-CodonPlus (DE3) RP cells containing the human calpain-1 CysPC gene were grown at 37 C in kanamycin selective LB media until OD 600 ¼ 0.6 then induced with 1 mM IPTG.The protein was expressed overnight at 20 C and cells harvested by centrifugation in a Sorvall RC6 Plus centrifuge (Thermo Fisher Scientific, Inc, MA, USA) using an SLA-3000 rotor at 6080 RCF for 20 min at 4 C.The cells were re-suspended in 20 mM HEPES, 100 mM NaCl, 0.5 mM TCEP pH 7.6 (buffer A) and lysed by sonication for 5 min (pulsed 5 s on, 10 s off).The lysate was clarified by centrifugation at 4 C for 40 min at 30,310 RCF in a Sorvall RC6 Plus centrifuge.The supernatant was passed through a 0.2 mm syringe filter and applied to a Ni-NTA column.The bound protein was washed with 15 C V buffer A and eluted with 10 C V buffer A containing 250 mM imidazole, which was further dialyzed in buffer A overnight in a 10 kDa membrane containing 1 mL aliquot of TEV protease.The cleavage product was then passed back through a Ni-NTA column to remove the 6xHis-GB1 solubility tag and TEV protease, with the flow through containing active CysPC as confirmed by SDS-PAGE, mass spectrometry and calpain-1 activity assay.

Expression and purification of TEV protease
BL21 (DE3) cells containing the TEV gene codon optimised for E. coli expression was obtained from Prof. Nigel Richards (Cardiff University).
The cells containing the TEV protease gene were grown at 37 C in ampicillin selective LB media until OD 600 ¼ 0.6 then induced with 1 mM IPTG.The protein was expressed overnight at 20 C and cells harvested by centrifugation in a Sorvall RC6 Plus centrifuge (Thermo Fisher Scientific, Inc, MA, USA) using an SLA-3000 rotor at 6080 RCF for 20 min at 4 C.The cells were re-suspended in 20 mM HEPES, 100 mM NaCl, 0.5 mM TCEP pH 7.6 (buffer A) and lysed by sonication for 5 min (pulsed 5 s on, 10 s off).The lysate was clarified by centrifugation at 4 C for 40 min at 30,310 RCF in a Sorvall RC6 Plus centrifuge.The supernatant was passed through a 0.2 mm syringe filter and applied to a Ni-NTA column.The bound protein was washed with 15 C V buffer A and eluted with 10 C V buffer A containing 250 mM imidazole.The eluent was mixed with 20% v/v glycerol (20 mL final volume), and stored at À80 C in 1 mL aliquots.

2 0 1 8 . 0 8 . 0 4 9 >
Pl e a s e n o t e: C h a n g e s m a d e a s a r e s ul t of p u blis hi n g p r o c e s s e s s u c h a s c o py-e di ti n g, fo r m a t ti n g a n d p a g e n u m b e r s m a y n o t b e r efl e c t e d in t his ve r sio n.Fo r t h e d efi nitiv e ve r sio n of t hi s p u blic a tio n, pl e a s e r ef e r t o t h e p u blis h e d s o u r c e.You a r e a d vis e d t o c o n s ul t t h e p u blis h e r's v e r sio n if yo u wi s h t o cit e t hi s p a p er. Thi s v e r sio n is b ei n g m a d e a v ail a bl e in a c c o r d a n c e wit h p u blis h e r p olici e s.

Fig. 1 .
Fig. 1. A. PD150606 B. PD151746 bind to the PEF(S) of calpain, and show modest selectivity for calpain-1 over calpain-2.Adams et al., 2015, synthesized their disulphide analogues C.and D. respectively, which bind to PEF(S) and were reported[5] with improved potencies in comparison to their monomer compounds E. CHEMBL203568, a compound reported to inhibit the calpain-1 complex (Uniprot IDs: P04632 and P07384) with an IC 50 value of 4.9 nM and confidence score of 7 F. CHEMBL204883, a compound reported to inhibit the calpain-1 complex (Uniprot IDs: P04632 and P07384) with an IC 50 value of 8 nM and confidence score of 7.

Fig. 5 .
Fig. 5. Docking studies predicted molecular interactions of compounds 1 and 10 with the human PEF(S) of calpain-1 small subunit (regulatory subunit) protein crystal structure (PDB ID: 4WQ2).A. The 2-chloro-4-cyclopropylsulfonamido phenyl ring of compound 1 shows p-stacking with His 131 and the carbonyl of its carboxylic acid moiety H-bonds with the same residue.The carbonyl of its amide moiety H-bonds with Trp 168 and the phenyl ring attached to the tert-butylcarbamoyl moiety is p-stacked with the same residue B. The pyridine ring of compound 10 is p-stacked with Trp 168 and the nitrogen of that ring H-bonds with same residue.The hydrophobic interactions of 1 and 10 with Trp 168 are also seen in the co-crystallised ligand/protein crystal structure (PDB ID: 4WQ3) [5].

Fig. 6 .
Fig.6.Docking studies predicted molecular interactions of the sulphonamides 2e5 with the PEF(S) protein crystal structure (PDB ID: 4WQ2), which suggest that hydrophobic interactions with Trp 168 are essential for PEF(S) binding A. The phenyl ring of compound 2, which is attached to the sulfonyl moiety is p-stacked with Trp 168 , and the same residue H-bonds with the carbonyl of its amide group, the hydroxyl of its cyclohexyl ring H-bonds with Gln 100 , and its sulfonyl group H-bonds with His 131 B. H-bonding interactions are predicted to occur between the carboxyl moiety of compound 3 and the Lys 172 and Trp 168 residues, H-bonding between its sulfonyl group and His 131 ,a n dpstacking with the Trp 168 via its pyridine ring C. The amino group of the amide moiety for compound 4 H-bonds with Glu 97 and Trp 168 , and the phenyl ring of its benzamide group p-stacks with Trp 168 , and the carbonyl of its amide moiety H-bonds with Lys 172 D. H-bonding interactions are predicted to occur between the carbonyl of the oxopyrrolidin-1-yl moiety for compound 5 and the Lys 172 and Trp 168 residues, H-bonding between its sulfonyl moiety and His 131 ,andp-stacking of the same residue with the with the aromatic ring of its trifluoromethylphenyl moiety.

4. 9 .
TNS displacement for compounds 1e5 10 mM PEF(S) in 20 mM Tris base, 1.1 mM CaCl 2 , 1 mM EDTA and pH 7.4 was incubated with 46.7 mM 2-p-toluidinylnaphthalene-6sulfonate (TNS, 1 mM stock in 40% ethanol) in a Greiner CELLSTAR 96 well black flat bottom plate for 5 min at 25 C. Compounds 1e10 were stored as 40 mM stock solutions in DMSO, then diluted from 500 mM to 500 nM by serial dilution over 10 wells with an Integra Viaflow 96 multichannel pipette in triplicate.
. The workflow generated Morgan fingerprints in the following sequence: It (a) read chemical data from an SDF file, (b) generated RDKit molecules from a molecule string representation (SDF), (c) generated hashed bit-based fingerprints for an input RDKit Mol column (d) converted RDKit molecules into string based molecule representations (SDF or smiles) (e) excluded columns from the input table (f) renamed columns (g) and saved data table into a CSV file.