Chem Biol Drug Des 2015; 86: 1411–1424 Research Article

Comparative Analysis of Pharmacophore Features and Quantitative Structure–Activity Relationships for CD38 Covalent and Non-covalent Inhibitors Shuang Zhang, Xiwen Xue, Liangren Zhang*, Lihe Zhang and Zhenming Liu* State Key Laboratory of Natural and Biomimetic Drugs, School of Pharmaceutical Sciences, Peking University, Beijing 100191, China *Corresponding authors: Zhenming Liu, [email protected]; Liangren Zhang, [email protected] In the past decade, the discovery, synthesis, and evaluation for hundreds of CD38 covalent and non-covalent inhibitors has been reported sequentially by our group and partners; however, a systematic structurebased guidance is still lacking for rational design of CD38 inhibitor. Here, we carried out a comparative analysis of pharmacophore features and quantitative structure–activity relationships for CD38 inhibitors. The results uncover that the essential interactions between key residues and covalent/non-covalent CD38 inhibitors include (i) hydrogen bond and hydrophobic interactions with residues Glu226 and Trp125, (ii) electrostatic or hydrogen bond interaction with the positively charged residue Arg127 region, and (iii) the hydrophobic interaction with residue Trp189. For covalent inhibitors, besides the covalent effect with residue Glu226, the electrostatic interaction with residue Arg127 is also necessary, while another hydrogen/nonbonded interaction with residues Trp125 and Trp189 can also be detected. By means of the SYBYL multifit alignment function, the best CoMFA and CoMSIA with CD38 covalent inhibitors presented cross-validated correlation coefficient values (q2) of 0.564 and 0.571, and non-cross-validated values (r2) of 0.967 and 0.971, respectively. The CD38 non-covalent inhibitors can be classified into five groups according to their chemical scaffolds, and the residues Glu226, Trp189, and Trp125 are indispensable for those non-covalent inhibitors binding to CD38, while the residues Ser126, Arg127, Asp155, Thr221, and Phe222 are also important. The best CoMFA and CoMSIA with the F12 analogues presented cross-validated correlation coefficient values (q2) of 0.469 and 0.454, and noncross-validated values (r2) of 0.814 and 0.819, respectively. Key words: 3D-QSAR, CD38, comparative analysis, covalent and non-covalent inhibitors, pharmacophore characteristics ª 2015 John Wiley & Sons A/S. doi: 10.1111/cbdd.12606

Received 19 December 2014, revised 9 May 2015 and accepted for publication 3 June 2015

Human CD38 is a member of ADP-ribosyl cyclase family, which is a type II or type III transmembrane glycoprotein (1) with multifunction such as enzymatic activity (2,3), receptor function (4), signal transduction (5,6), cell activation, cell factor mediation (7,8), and intercellular adhesion activity (9). It is widely expressed in different tissues and cells (10). CD38 can catalyzes the transformation of b-NAD+ and NADP to the Ca2+-mobilizing messengers cADPR and NAADP, respectively, and also the hydrolysis of cADPR and b-NAD+ to ADPR (11,12). These mentioned Ca2+ messengers can regulate the release and influx of cellular Ca2+ to maintain life activities (11,13). The substrate catalytic mechanism studies of CD38 revealed two enzyme–ligand transition states: one is that the residue Glu226 of CD38 forms the covalent intermediate with the C0 -1 carbon of ribose from substrate; the other transition state may form a non-covalent oxocarbenium ion intermediate (14–17). Based on the mentioned mechanism, the reported CD38 inhibitors up to now can be divided into two classes: the covalent and non-covalent inhibitors. The covalent inhibitors are mainly NAD+ analogues, among which, a series of fluoro-substituted NAD+ derivatives were firstly reported by Muller-Steffner et al. (18,19). Subsequently, Chen et al. (20) from our laboratory and his coworkers (21) synthesized a series of more potent and membrane-permeable NAD+ derivatives. The scaffolds of CD38 non-covalent inhibitors are diverse. Except for the discovered NAD+ analogues discovered (18,20–22), Zhou (23) from our laboratory designed and synthesized a series of new heterocyclic inhibitors, such as the indole compound F12. Subsequently, the heterocyclic inhibitors were further investigated and modified by Wu et al. (24). Several natural products such as the flavonoids were reported by Escande et al. (25) and Kellenberger et al. (26) as novel CD38 non-covalent inhibitors, while recently rhein and its water-soluble derivative rhein tripotassium salt (K-rhein) as low micromolar non-competitive CD38 inhibitors inhibited microglia activation function which is mediated by CD38 (27). According to the resolved crystallography of CD38–ligand complexes, the enzyme binding site is a big U-shaped outer (28). Compar1411

Zhang et al.

ison of CD38 covalent with non-covalent inhibitors showed the variety of chemical fragment and spatial volume, especially that the flavonoids may inhibit CD38 by binding to different sites. Despite the fact that the inhibition ability of non-covalent inhibitors (with most IC50 values at lM levels) is much weaker than that of covalent inhibitors (with the best IC50 values at nM levels at present), the chemical stability, membrane permeability, and the enzyme hydrolysis-resistant ability of non-covalent are obviously superior to covalent inhibitors (29), driving the non-covalent inhibitors to be more potent chemical tools in further research. Therefore, we intended to figure out some clues from the binding properties of covalent inhibitors to help improve the inhibition effect of non-covalent inhibitors. In this work, CD38 inhibitors including both covalent and non-covalent were collected and clustered into various groups based on the similarity of their chemical skeletons. Docking studies pharmacophore modeling and QSAR analysis were conducted on different inhibitor groups to investigate the commonalities and differences of the CD38 covalent and non-covalent inhibitors. Based on the comparative studies, our aim was to carry out a possible preliminary quantitative structure–activity relationship between CD38 covalent and non-covalent inhibitors, which might provide some insights in the design of CD38 inhibitors.

Methods and Materials As shown in Figure 1, the methodological studies are mainly composed of four sections: data collection and the 3D-structures generation; molecular clustering based on chemical similarity; molecular docking; and generation and validation of the pharmacophore and QSAR models of CD38 covalent and non-covalent inhibitors, respectively.

Data sets CD38 inhibitors For the data set of CD38 inhibitors, most of them were synthesized and reported by us (20–24,30). The others were collected from published literatures (25–27,31). The 3D structures of all compounds were constructed using the ‘sketch molecule’ module in SYBYL-X 2.0a, and energy minimization was performed by the Powell gradient algorithm with the tripos force field (32) and Gasteiger–Huckel charge (33).

Selection of CD38 crystal structures 35 CD38 crystal structures were available in Protein Data Bank (PDB) (34), including apo-CD38, 22 complexes cocrystalized with substrates and non-covalent inhibitors, and also eight complexes with covalent inhibitors. By comparing the crystal structures (see details in Results and Discussion section), we selected the cocrystalized structure with NMN (PDB ID: 3DZK) (15) for docking, and all water molecules and its ligand were removed from the original file, polar hydrogen atoms were added, and kollman charges (35) were assigned to the protein using AUTODOCK Tools (36), while the 3DZG (15), 3DZF (15), and 3ROP (21) were used for binding mode analysis of covalent inhibitors.

Chemical structure clusting The CD38 non-covalent inhibitors were clustered by PIPELINE PILOT 7.5 (Accelrys, San Diego, CA, USA) with the cluster ligands protocol. Clustering was employed to identify and divide the compounds into several groups based on their chemical similarity. The pipeline pilot maximum dissimilarity clustering algorithm was applied. This algorithm starts with a randomly chosen structure as the first cluster center. The molecule with the maximal distance from the first compound is selected as the next cluster center. The compound with the maximal distance

Figure 1: Flow diagram of the data set preparation protocol and computational procedures.

1412

Chem Biol Drug Des 2015; 86: 1411–1424

CD38 Covalent and Non-covalent Inhibitors

from both current cluster centers is selected after that. The process is repeated until the desired number of cluster centers is reached. The non-selected objects are then assigned to the nearest cluster center to determine the cluster membership. The structural clustering was based on the pairwise Tanimoto coefficient based on the FCFP_6 fingerprint.

Docking Molecular docking was performed using AUTODOCK 4.0 (37) to explore the detailed binding modes of CD38 with the inhibitors and then to develop receptor-based pharmacophore models. The molecular docking was performed with the size of the box, in which both the enzyme and ligand were embedded, set to 22.5  A 9 22.5  A 9 22.5  A along the x, y, and z dimensions of the Cartesian coordinate system. The number of docking runs was 100 and the default parameters in the AUTODOCK4.0 setting were used.

CoMFA and CoMSIA molecular field descriptors were used as independent variables, while the CD38 potential expressed as pIC50 formed the dependent variable in the PLS (partial least square) (39,40) regression analyses for the development of 3D-QSAR models. Cross-validation with the LOO (leave-one-out) (41,42) procedure was performed to internally validate the model, wherein each molecule was successively removed from the model derivation and its pIC50 value predicted using the model built from the remaining molecules. PLS was used to identify the optimal number of components to be retained for the derivation of the final 3D-QSAR models. The optimum number of components corresponds to the one that yielded the highest cross-validated (q2) value and had the smallest standard error of prediction. The non-cross-validated analyses were then performed using the optimum number of components to determine the conventional correlation coefficient r2, the standard errors (SE), and the F-value.

Results and Discussion Pharmacophore definitions of protein–ligand complexes To generate the structure-based pharmacophore models of CD38, the software LIGANDSCOUT3.1.2 (38) was applied. As input, for non-covalent inhibitors, we chose the representative compounds with the highest IC50 value of each cluster binding with 3DZK (15); for covalent inhibitors, the PDB file 3DZF (15) was used. Then, those complexes were imported into LIGANDSCOUT3.1.2 (38) to generate pharmacophore models. Subsequently, these pharmacophores were superimposed between each other and those compounds with similar pharmacophores were clustered to one class for the further QSAR study.

CD38 inhibitors data set Till this manuscript was written, the amount of known compounds with CD38 inhibitory activity is 219, including 170 non-covalent inhibitors and 49 covalent inhibitors. And those inhibitors can be divided into three main categories: NAD analogues, flavonoids, and other heterocyclic compounds. Among them, all the covalent inhibitors are NAD analogues, while the non-covalent inhibitors include 2 NAD analogues, 16 flavonoids, and 152 heterocyclic compounds. The IC50 values of known CD38 inhibitors are various from < 0.1 lM to more than 1 mM (2 in IC50 ≤ 0.1 lM; 14 in 0.1 lM < IC50 ≤ 10 lM; 35 in 10 lM < IC50 ≤ 100 lM; 101 in 100 lM < IC50 ≤ 1000 lM and 67 in IC50 > 1000 lM).

3D-QSAR modeling and validation The test sets and the training sets were chosen by hierarchical clustering from original data set. To test the predictive power of the derived CoMFA and CoMSIA models, biological activities of the test set molecules were, respectively, predicted using the two models derived from the training sets of covalent and non-covalent inhibitors.

According to the methodological criterion, 152 reported inhibitors were chosen for this study (see Tables S1 and S2), which covered all currently structure type of known CD38 inhibitors. According to requirement of CoMFA and CoMSIA study to the data set, 60 compounds were finally selected to build 3D-QSAR models.

The molecules were aligned to the template molecule on a common backbone using SYBYL-X2.0a. For alignment of the inhibitors, the structure of compounds G24 and F12 was used as the template for covalent and non-covalent inhibitors, respectively. The molecular conformation was generated by docking. The steric and electrostatic potential fields for CoMFA were calculated at each lattice intersection of a regularly spaced grid of 2.0  A in all three dimensions within the defined region using a sp3 hybridized carbon probe with a + 1.0 charge. CoMSIA similarity indices (steric, electrostatic, hydrophobic, hydrogen bond donor, and hydrogen bond acceptor) were derived within the same lattice box as used for CoMFA.

Clustering of CD38 non-covalent inhibitors Firstly, the covalent and non-covalent inhibitors were clustered based on their chemical similarity. As the covalent inhibitors were all NAD analogues, they were regarded as one class. For non-covalent inhibitors, the cluster result using the ‘cluster ligands’ protocol in PIPELINE PILOT7.5 (Accelrys) was deposited in Table S1. The representative individuals of the five categories were shown in Table 1. From the clustering result, it can be seen that the flavonoids and NAD analogues were divided into two categories, while the heterocyclic compounds can be further clustered as benzopyrroles (cluster 1) and benzothiazoles (cluster 4 and 5).

Chem Biol Drug Des 2015; 86: 1411–1424

1413

Zhang et al. Table 1: The representative compound of five CD38 non-covalent inhibitor categories Cluster

Representative compound

1

O

H N

O N H

Number 41 O

O

O

F12 OH

2

16

OH O

HO

OH

F46 NH2

3 N O

O HO P O O

O HO

19

NH N

N OH

F67 4

O

O

O

N H

N S

N H

22

NH

Docking studies

F98 HO2C

5 O N

O N H

S

28

O

NH

F126 The number indicated the amount included in each category.

CD38 structures selection The resolved CD38 crystal structures (15,16,28,43,44) show that the enzyme catalytic pocket is an U-sharp, composed of a5 and a6 helixes of N-terminal and a7 helix and b5 sheet of C-terminal, including key residues Trp125, Ser126, Arg127, Glu146, Asp155, Trp189, Ser193, and Thr221. The mutation studies indicated the residues Trp125, Ser193, Trp189, and Glu146 as the most important for CD38 catalytic function. The indole ring of residues Trp125 and Trp189 have hydrophobic effect on the aromatic ring of substrates; the residues Asp155 and Glu146 can help to identify and locate the substrates by hydrogen bond interaction. The residue Ser193 can co-ordinate and stabilize the catalytic transition intermediates. The residue Glu226 is the catalytic amino acid residue involved in the intermediates formation; mutation of Glu226 to any other amino acid will result in complete loss of CD38 activity (16,28). The CD38 crystal structures representing different catalytic transition intermediates have been resolved. In this study, we chose the structure complexes with NMN (PDB ID:3DZK) (15) for non-covalent CD38 inhibitor docking and binding mode analysis for the reason as follows. First of all, for most of resolved CD38 1414

structures, in order to capture the enzyme–substrate (such as NAD and cADPR) complex, the residue Glu226 was mutated to Gln226. As the result, it cannot fully reflect the CD38 real status of natural wide type. The NMN-CD38 complex with the key residues including Glu226 is in natural condition and it is the one most closely to the wide type protein. Secondly, although the NMN is only a partial fragment of NAD molecular, its binding mode is similar to NAD. In fact, the missing chemical groups compared with NAD are located outside CD38 catalytic pocket and would have no considerable impact for enzyme–ligand binding. Finally, the RMSD of NMN redocking to file 3DZK was 0.397. While the RF5-CD38 complex (PDB ID:3DZF) (15) was chosen for the pharmacophore analysis of covalent inhibitors, the RF5 is covalent binding mode of G24. And the 3DZK and 3DZF were resolved by the same research team. All mentioned above stated that the enzyme–NMN and enzyme–RF5 complexes were finally chosen for the following studies.

Docking of CD38 covalent inhibitors and binding mode analysis By comparing the binding modes before (Figure 2B,E) and after (Figure 2A,D) the formation of covalent bond, we can find that the main difference lies in whether or not the nicotinamide part remains and the covalent bond is formed with residue Glu226. In Figure 2A, the nicotinamide can form hydrophobic interaction with Trp189, and hydrogen bond interactions with residues Glu146 and Asp155, which is similar to the docking mode (Figure 2B). Due to the losing of nicotinamide group, the orientation of ligands in the active pocket became more freedom. Besides, the binding mode and conformation before and after the formation of covalent bond with the remaining parts of G24 are almost unchanged. For G15, due to the nicotinamide group leaving, the remaining part is more freedom so as to the sugar ring flip, but the phosphate group comparing with the docking model is stable. From Figure 2C, the binding mode can be regarded as a further conformation after the covalent bond forming. Due to the nicotinamide leaving, the binding capacity becomes stronger. Thus, the ligands can revolve around with Glu226 liking a butterfly. Although the location of the sugar ring changed, the original hydrogen bond and electrostatic interactions have not yet been affected. When investigating the covalent inhibitor binding modes (Figure 2A–E), the 50 -phosphate group can form possible hydrogen bond interactions with residues Arg127, Ser126, Phe222, and Thr221. Further studies showed the 50 -phosphate group located in the positively charged area of CD38 catalytic pocket, and this region can form a strong electrostatic interaction with the inhibitors. From the Figure 2F, we can see that the difference of the residues in Chem Biol Drug Des 2015; 86: 1411–1424

CD38 Covalent and Non-covalent Inhibitors

Figure 2: The binding patterns of compounds G24 and G15 with CD38 before and after covalent bond with residue Glu226 formation. (A) Compound G24 covalently bind to CD38 (PDB ID: 3DZG), with nicotinamide remained; (B) compound G24 non-covalently docked to CD38; (C) compound G24 covalently bind to CD38 (PDB ID: 3DZF), without nicotinamide; (D) compound G15 covalently bind to CD38 (PDB ID: 3ROP); (E) compound G15 noncovalently docked to CD38; (F) superposition of the CD38 key active site residues. (The color green indicates 3DZK, blue for 3DZF, red for 3DZG, and yellow for 3ROP).

A

B

C

D

E

F

binding pocket is very small. And by comparing the binding patterns of compounds G24 and G15 with CD38 before and after covalent bond with residue Glu226 formation, the changes of the ligands pose and interaction with key residues are relatively small. So the binding modes shown in Figure 2 can be used for the following pharmacophore generation.

Docking of CD38 non-covalent inhibitors and binding mode analysis As mentioned above, the non-covalent inhibitors of CD38 can be divided into five categories based on structure similarity. Their binding modes generated by docking strategy were shown in Figures 3 and 4D. Chem Biol Drug Des 2015; 86: 1411–1424

As seen in Figures 3 and 4D, we found that the common feature of the high active compounds is forming hydrogen bond with residue Glu226, including the cluster 1 (F12), flavonoids (F46), and the cluster 3 (F67). And the compounds F12 and F67 can also form the hydrogen bond interaction with residue Arg127. The flavonoids cannot perform a U-shape conformation, so they cannot form hydrogen bond interaction with residue Arg127. Besides, compounds F12 and F46 can also form hydrogen bond interactions with residues Asp155 and Asp156. The compound F98 belonging to the cluster 4 can form hydrogen bonds with residues Trp125, Leu145, and Glu146, while the compound F126 in the cluster 5 can form hydrogen bond with residue Arg127. Although they 1415

Zhang et al. A

B

C

A

C

1416

D

Figure 3: Binding modes of CD38 with non-covalent inhibitors. (A) Compound F46 forms hydrophobic interaction with residues Trp125 and Trp189. Also, the hydrogen bonds are formed with residues Trp125, Lys129, Asp155, Asp156, and Glu226; (B) compound F67 forms hydrophobic interaction with residue Trp189. Also, the hydrogen bonds are formed with residues Trp125, Ser126, Arg127, Asp155, Ser220, Thr221, Phe222, and Glu226; (C) compound F98 forms hydrophobic interaction with residues Trp125 and Trp189. Also, the hydrogen bonds are formed with residues Trp125, Leu145, and Glu146; (D) compound F126 forms hydrophobic interaction with residue Trp189. Also, the hydrogen bonds are formed with residue Arg127.

B

D

Figure 4: Binding mode of compound F12 with CD38. (A) The electrostatic surface of CD38 binding with compound F12. (Red: negatively charged region; blue: positively charged region). (B) The hydrophobic surface of CD38 binding with compound F12. The colors from blue to white indicate hydrophilic area, and the colors from orange to red indicate hydrophobic area; (C) the polar surface of CD38 binding with compound F12. (Purple: polar region; brown: hydrophobic region); (D) F12-CD38 binding mode with key residues generated by docking. (The compound F12 forms hydrogen bonds with residues Trp125, Arg127, Lys129, Asp156, Thr221, and Glu226, while it forms hydrophobic interaction with residues Trp125 and Trp189.)

Chem Biol Drug Des 2015; 86: 1411–1424

CD38 Covalent and Non-covalent Inhibitors

can also form hydrophobic interaction with residues Trp125 and Trp189 as others, the hydrogen bond interaction with residue Glu226 cannot be detected. The compound F12 is one of the potent non-covalent CD38 inhibitors reported by us. From Figure 4, it can be seen that the compound fits the pocket properly with suitable conformation; the benzopyrrole group locates in the junction of hydrophobic (see Trp125 region) and hydrophilic (see Glu226 region) area of the active pocket (Figure 4B). The carbonyl group of F12 locates in the positively charged region and can form hydrogen bond interaction with residue Arg127. When the inhibitors can form U-shaped conformation, it should bind with CD38 tightly. According to the above statement, it showed that the covalent/hydrogen bond interactions between residue Glu226 and inhibitors are a key factor associated with the final IC50 value. Besides the residue Glu226, the residue Arg127 within the catalytic pocket is electropositive, and this region can form possible hydrogen bond interactions with inhibitors. And the inhibitors can form the hydrophobic interactions with residues Trp125 and Trp189, so the residues Trp125 and Trp189 play an important role for the new CD38 inhibitor design. The 50 -phosphate group in CD38 covalent inhibitors exactly locates in the positively charged region and forms a strong electrostatic interaction with residue Arg127. So some chemical groups that are introduced to CD38 noncovalent inhibitors can interact with resides Glu226, Arg127, Trp125, and Trp189, for example, the phosphate group in CD38 covalent inhibitors that can form interactions with residues Ser126, Arg127, and Thr221. If the phosphate group is introduced to the non-covalent inhibitors, the activity of these inhibitors may improve.

The pharmacophore models of CD38 covalent and non-covalent inhibitors In Sections CD38 structures selection and Docking studies, we have investigated the CD38 binding site and the key residues. Then, the reasonable binding modes were obtained through the lowest energy and the frequency of docking poses. Subsequently, the binding modes were used to build the following pharmacophore models (Figure 5). From Figure 5A,B, we can see that besides the leaving nicotinamide group, the pharmacophore features of CD38 covalent inhibitors are almost the same before and after the covalent bond formation, which indicates that the binding modes of covalent inhibitors generated by AUTODOCK can be used in the following QSAR analysis. As shown in Figure 5, the compound G24 is the better covalent inhibitor with IC50 = 0.61 lM, and the compound F12 is a known potent non-covalent inhibitor with IC50 = 4.7 lM. The common feature of them is that they both can form an interaction with the catalytic residue Chem Biol Drug Des 2015; 86: 1411–1424

Glu226. Further investigation showed that due to the replacement of ribose by indole group, the pharmacophore features of F12 are more abundant compared with G24, such as F12 can form hydrogen bond interaction with residue Asp156 and hydrophobic interaction with residue Trp125. However, according to the known IC50 value, those additional interactions are not strong enough as the effect of a covalent bond formed with residue Glu226. For compound F67 (IC50 = 7.7 lM), which is NAD analogues, it can be seen that its feature is very similar with the covalent inhibitors from the pharmacophore models. However, due to the replacement of nicotinamide group by purine, the compound F67 cannot form covalent bond with residue Glu226, so its’ IC50 value is down nearly 10% comparing with compound G24’s. All those suggested that the covalent bond of NAD analogues formed with residue Glu226 is very important. The compound F46 (IC50 = 6.0 lM) was reported as CD38 non-covalent inhibitor with flavonoid skeleton. From its pharmacophore features (Figure 5E), it is more like the right part of compound F12, which suggests that compound F46 can be used as a basic chemical scaffold to achieve the potent CD38 inhibitors by chemical cross-linking. The compound F98 (IC50 = 44 lM) and compound F126 (IC50 = 23 lM) were CD38 non-covalent inhibitors, which cannot form hydrogen bond interaction with catalytic residue Glu226. At the same time, as those sharp molecules cannot match the CD38 binding pocket properly, the pharmacophore features of those compounds are different from others. However, those compounds can form hydrogen bonds with residues Trp125 and Arg127. By comparing the analysis, we can see that those inhibitors with good CD38 inhibitory activity can always form hydrogen bond interaction/or covalent bond with residue Glu226 and also form hydrophobic interaction with residue Trp189, no matter whether the covalent or non-covalent inhibitors. The difference between CD38 covalent and non-covalent inhibitors lies in that the phosphate group of covalent inhibitors can form electrostatic interactions and hydrogen bond interactions with residue Arg127, Thr221, and Phe222, while the benzopyrrole group of non-covalent inhibitors can form hydrogen bond interaction with residue Glu226 and hydrophobic interaction with residue Trp125. Without negatively charged chemical groups such as phosphate group, those non-covalent inhibitors cannot form strong possible hydrogen bond interactions and electrostatic interactions with residues Ser126, Arg127, Thr221, and Phe222.

3D-QSAR models CoMFA and CoMSIA statistical results The training and internal test sets for 3D-QSAR analysis of the CD38 covalent and non-covalent inhibitors (see 1417

Zhang et al. A

B

Figure 5: The pharmacophore models of CD38 covalent and non-covalent inhibitors generated by LIGANDSCOUT 3.12. Left: 2D pharmacophore models of the inhibitors, the green arrows indicate hydrogen bond donor, red arrows for hydrogen bond receptor, yellow area indicates hydrophobic center, blue arrow for aromatic ring center, and the red area indicates negatively charged region. Right: 3D pharmacophore models of the inhibitors, hydrogen bond donors were shown as purple balls, the hydrogen bond receptors were shown as green balls, the light blue balls indicate the hydrophobic region center, and the yellow balls indicate the aromatic ring center, while the blue balls indicate the negatively charged center. All the pharmacophore models were generated from the protein–inhibitor complexes mentioned above. (A) Compound G24 with CD38 (PDB ID: 3DZF); (B) the pharmacophore model of G24 based on docking complexes. (C– G) showed the pharmacophore models of other non-covalent inhibitors with CD38.

C

D

E

F

The statistical analysis results for the CoMFA and CoMSIA are given in Table 3. PLS analysis on the CD38 covalent and non-covalent inhibitors in the training sets resulted in the CoMFA models with the cross-validated q2 of 0.564 and 0.469. The two models gave an optimal number of components (ONC) of 5 and 2, and the conventional correlation coefficient r2 of 0.967 and 0.814. And the SEE (average standard error) of 0.246 and 0.211 suggested as a good model. The best CoMSIA model of the covalent and non-covalent inhibitors gave the cross-validated q2 of 0.571 and 0.454, the optimal number of components (ONC) is 4 and 3, and the conventional correlation coefficients r2 were 0.971 and 0.819. And the SEE were 0.222 and 0.212, respectively. When the CoMFA and CoMSIA models were applied to the test compounds, the models gave satisfactory results. The predicted correlation coefficient r2pred of CD38 covalent inhibitors for the CoMFA and CoMSIA models were 0.705 and 0.944, respectively. And the predicted correlation coefficient r2pred for the CoMFA and CoMSIA models of the CD38 non-covalent inhibitors were 0.614 and 0.901. The observed pIC50 and the predicted pIC50 of each inhibitor are compared and listed in range of more than 2 log units (Table 2). The plot of predicted versus observed pIC50 values for the four models is shown in Figure 6.

G

Table 2) were randomly selected, respectively. The pIC50 values for both the training set and the test set compounds covered a range of more than 2 log units (Table 2). 1418

Analysis of 3D-QSAR contour maps CoMFA contour maps. From the Figure 7A, it shows that the blue patch matches to the phosphate groups of CD38 covalent inhibitors (consistent with Figure 4A, the positively charged region around the residue Arg127). So introducing negative chemical groups at this position can help to improve the inhibitory activity, such as compound G10 (IC50 = 457 lM) VS compound G11 (IC50 = 800 lM). In Figure 7C, the amide group of compound F12 with blue patch is consistent with the covalent inhibitors’ feature. So increasing the positive charge at the amino group position may benefit the final inhibitory activity, such as Chem Biol Drug Des 2015; 86: 1411–1424

CD38 Covalent and Non-covalent Inhibitors Table 2: Observed and predicted activities for the training set and test set of CD38 covalent inhibitors by the 3D-QSAR models CoMFA

CoMSIA

Compound

Observed pIC50

Predicted pIC50 residual

Predicted pIC50 residual

G1t G2t G3 G4 G5t G6 G7 G8t G9 G10 G11 G12 G13 G14 G15 G16 G17 G18 G19 G20 G21 G22 G23t G24 F1 F2 F3 F4 F5 F6 F7 F8 F9t F10t F11 F12 F13 F14 F15 F16t F17 F18 F19 F20 F21t F22 F23 F24t F25 F26 F27 F28t F29 F30 F31 F32

4.00 3.66 5.66 5.51 4.95 3.10 3.92 3.52 3.03 3.34 3.10 5.82 5.89 5.73 6.22 3.39 5.71 4.03 4.94 4.25 4.63 5.06 4.85 6.21 4.20 3.61 3.72 3.71 3.32 3.55 3.55 4.28 3.29 3.17 3.72 5.33 3.06 3.13 3.16 3.55 3.76 3.24 3.11 3.82 3.32 4.11 3.18 3.11 4.26 3.95 3.88 3.19 3.92 3.89 3.86 3.52

4.14 3.49 5.39 5.48 5.34 3.10 3.57 2.93 3.10 3.74 3.24 5.98 5.89 6.11 5.99 3.28 5.77 4.03 4.98 4.48 4.45 5.13 4.25 5.99 4.40 3.45 3.90 3.73 3.47 3.29 3.66 4.10 3.41 3.36 3.86 4.61 2.77 3.19 3.14 3.57 3.77 3.26 3.47 3.79 3.38 4.23 3.19 3.62 4.19 4.17 4.11 3.5 3.88 3.86 3.92 3.46

0.14 0.17 0.27 0.03 0.39 0.00 0.35 0.59 0.07 0.40 0.14 0.16 0.00 0.38 0.23 0.11 0.06 0.00 0.04 0.23 0.18 0.07 0.60 0.22 0.20 0.16 0.18 0.02 0.15 0.26 0.11 0.18 0.12 0.19 0.14 0.72 0.29 0.06 0.02 0.02 0.01 0.02 0.36 0.03 0.06 0.12 0.01 0.51 0.07 0.22 0.23 0.31 0.04 0.03 0.06 0.06

Chem Biol Drug Des 2015; 86: 1411–1424

3.93 3.59 5.61 5.65 5.17 3.04 3.77 3.49 3.00 3.81 2.91 5.88 5.60 5.66 6.29 3.58 5.79 3.98 5.17 4.50 4.37 4.84 5.18 6.11 4.34 3.53 3.90 3.62 3.54 3.41 3.75 4.15 3.23 3.12 3.89 4.56 2.84 3.14 3.06 3.27 3.66 3.27 3.37 3.85 3.24 4.04 3.18 3.32 4.22 4.14 4.03 3.18 4.05 4.00 3.96 3.46

0.07 0.07 0.05 0.14 0.22 0.06 0.15 0.03 0.03 0.47 0.19 0.06 0.29 0.07 0.07 0.19 0.08 0.05 0.23 0.25 0.26 0.22 0.33 0.10 0.14 0.08 0.18 0.09 0.22 0.14 0.20 0.13 0.06 0.05 0.17 0.77 0.22 0.01 0.10 0.28 0.10 0.03 0.26 0.03 0.08 0.07 0.00 0.21 0.04 0.19 0.15 0.01 0.13 0.11 0.10 0.06

Table 2: continued CoMFA

CoMSIA

Compound

Observed pIC50

Predicted pIC50 residual

Predicted pIC50 residual

F33 F34 F35 F36t

3.55 3.31 3.86 3.05

3.38 3.40 3.93 3.54

3.40 3.20 4.05 3.24

0.17 0.09 0.07 0.49

0.15 0.11 0.19 0.19

The compounds G1–G24 are CD38 covalent inhibitors. Among them, the compounds G1, G2, G5, G8 and G23 were used as test set. The compounds F1–F36 are CD38 non-covalent inhibitors and the compounds F9, F10, F16, F21, F24, F28 and F36 were used as test set. Table 3: CD38 inhibitors’ summary of the CoMFA and CoMSIA results CD38 covalent inhibitors

CD38 non-covalent inhibitors

CoMFA value CoMSIA value

CoMFA value CoMSIA value

0.564 q2a ONCb 5 r2c 0.967 0.246 SEEd Fe 46.154 Field contributions (%) S 0.631 E 0.369 H D A

0.571 4 0.971 0.222 117.636

0.469 2 0.814 0.211 56.858

0.454 3 0.819 0.212 37.732

0.099 0.251 0.224 0.215 0.210

0.601 0.399

0.142 0.302 0.255 0.201 0.100

S, steric; E, electrostatic; H, hydrophobic; D, H-bond donor; A, Hbond acceptor. a Cross-validated correlation coefficient after leave-one-out procedure. b Optimal number of principal components. c Correlation coefficient. d Standard error of estimate. e Ratio of r2 explained to unexplained = r2/(1r2).

compound F117 (IC50 = 88 lM) VS compound F119 (IC50 = 152 lM). CoMSIA contour maps. From the Figure 8F, which indicates the hydrogen bond donor importance, it can be seen that the cyan color pieces matches to the benzene group of compound F12. And the blue patch and magenta patch are near to the phosphate group of compound G24 (Figure 8A,G). So more hydrogen bond interactions with residue Arg127 could help to improve the activity of inhibitors, such as compound F119 (IC50 = 152 lM) VS compound F121 (IC50 = 29 lM). Figure 8A,C suggest that introducing one hydrophobic substituent at the phosphate group position can enhance

1419

Zhang et al. A

C

B

Figure 6: Plots of predicted versus observed pIC50 values for CD38 inhibitors based on CoMFA (A and C) and CoMSIA models (B and D). (A) The scatter diagram of CD38 covalent inhibitors with CoMFA model; (B) the scatter diagram of CD38 covalent inhibitors with CoMSIA model; (C) the scatter diagram of CD38 non-covalent inhibitors with CoMFA model; (D) the scatter diagram of CD38 non-covalent inhibitors with CoMSIA model. The training sets were present with the black solid triangle, and the test set were present with black solid square.

D

A

B

C

D

Figure 7: CoMFA StDev*Coeff contour maps based on compounds G24 (upper) and F12 (down). (A) and (C) indicate electrostatic fields: electropositive (blue) and electronegative (red); (B) and (D) indicate steric fields: favored (green) and disfavored (yellow).

the inhibitory activity. For example, the compound with a substituent at phosphate group (compounds G3, G12, and G19) is better than compound G18, which have two substituents. For the non-covalent inhibitors, yellow patch is near to their benzene groups, and more hydrophobic 1420

substituent here could help to form hydrophobic interactions with residue Trp176. From Figure 8B, it indicates that increasing the electronegativity of para-substituted chemical group of benzene ring Chem Biol Drug Des 2015; 86: 1411–1424

CD38 Covalent and Non-covalent Inhibitors A

B

C

D

E

F

G

H

Figure 8: CoMSIA StDev*Coeff contour maps based on compounds G24 (left) and F12 (right). (A) and (B) indicate electrostatic fields and steric fields: electropositive (blue) and electronegative (red), favored (green) and disfavored (yellow); (C) and (D) indicate hydrophobic field: favored (yellow) and disfavored (white). (E) and (F) indicate hydrogen bond donor field: favored (cyan) and disfavored (purple); (G) and (H) indicate hydrogen bond acceptor field: favored (magenta) and disfavored (red).

Chem Biol Drug Des 2015; 86: 1411–1424

1421

Zhang et al.

Figure 9: Summary of relevant features identified by docking, pharmacophore, and 3D-QSAR analyses for CD38 inhibitors. Light blue: hydrophobic region; red: negatively charged region; blue: positively charged region.

positively charged region around residue Arg127; and (iii) hydrophobic interaction with residue Trp189. All those mentioned features are obtained consistently by the pharmacophore and QSAR studies. There are some differences between the covalent and non-covalent inhibitors. As the non-covalent inhibitors cannot form covalent bond with Glu226 and cannot have negatively charged groups, such as the phosphate group, the activity of non-covalent inhibitors is not good as the activity of covalent inhibitors. In this study, there are good 3D-QSAR models of CD38 covalent and non-covalent inhibitors, which further prove the above conclusion and provide tools to design and modify novel CD38 inhibitors. The non-covalent inhibitors are more stable and more likely to be modified. The obtained results and suggestions can help to provide some insights into further structural modification and development of new potent inhibitors. In our group, other members are designing and synthesizing new CD38 inhibitors and the results are going to be published.

can also help to enhance the inhibitors’ activity, such as compound F12 is better than compound F11.

Acknowledgments According to the study, the commonalities and differences of CD38 covalent and non-covalent inhibitors are clear. As mentioned above, the results from docking, pharmacophore modeling, and QSAR studies are consistent. In Figure 9, the indole of F12 can be instead of quinoline, quinazoline, flavonoids, benzothiazole, etc. that could form hydrophobic interaction with Trp125, and with Glu226, Glu146, and Trp125 form hydrogen bond interactions; the ethyl benzoate of F12 can form hydrophobic interaction with Trp189, and with Asp155 and Asp156 form hydrogen bond interactions, so the ethyl benzoate could be changed to other groups such as the nicotinamide of the CD38 covalent inhibitors, thiazole, and so on forming hydrophobic interaction with Trp189 and hydrogen bond interactions with Trp189, Asp155, and Asp156; and the amide groups of F12 near the Arg127 can be changed to phosphate negatively charged groups of CD38 covalent inhibitors, which can form electrostatic interaction with Arg127, and form hydrogen bond interactions with Thr221 and Phe222; the benzene ring of F12 on the left can be changed to other hydrophobic groups such as alkanes that could form hydrophobic interaction with Trp176. Thus, the pharmacophore characteristics and quantitative structure–activity relationships of CD38 covalent and non-covalent inhibitors achieved by this study can be further used for new CD38 inhibitor design and discovery.

Conclusions In summary, the comparative analysis and study showed three common features for both CD38 covalent and noncovalent inhibitors with better enzyme inhibitory: (i) hydrogen bond interaction with residue Glu226 and hydrophobic interaction with residue Trp125; (ii) interaction with the 1422

We would like to acknowledge Dr. Hongwei Jin from Peking University for scientific advices. We thank Dr. Jie Xia for the choice of techniques used in this work. This study was supported by the National Natural Science Foundation of China (21272017 and 81172917).

References 1. Zhao Y.J., Lam C.M.C., Lee H.C. (2012) The membrane-bound enzyme CD38 exists in two opposing orientations. Sci Signal;5:ra67. 2. Takasawa S., Tohgo A., Noguchi N., Koguma T., Nata K., Sugimoto T., Yonekura H., Okamoto H. (1993) Synthesis and hydrolysis of cyclic ADP-ribose by human leukocyte antigen CD38 and inhibition of the hydrolysis by ATP. J Biol Chem;268:26052–26054. 3. Tohgo A., Takasawa S., Noguchi N., Koguma T., Nata K., Sugimoto T., Furuya Y., Yonekura H., Okamoto H. (1994) Essential cysteine residues for cyclic ADP-ribose synthesis and hydrolysis by CD38. J Biol Chem;269:28555–28557. 4. Lund F.E., Yu N., Kim K.M., Reth M., Howard M.C. (1996) Signaling through CD38 augments B cell antigen receptor (BCR) responses and is dependent on BCR expression. J Immunol;157:1455–1467. 5. Howard M., Grimaldi J., Bazan J., Lund F., SantosArgumedo L., Parkhouse R., Walseth T., Lee H.C. (1993) Formation and hydrolysis of cyclic ADP-ribose catalyzed by lymphocyte antigen CD38. Science;262:1056–1059. 6. Aarhus R., Graeff R.M., Dickey D.M., Walseth T.F., Hon C.L. (1995) ADP-ribosyl cyclase and CD38 catalyze the synthesis of a calcium-mobilizing metabolite from NADP+. J Biol Chem;270:30327–30333. Chem Biol Drug Des 2015; 86: 1411–1424

CD38 Covalent and Non-covalent Inhibitors

7. Ausiello C.M., la Sala A., Ramoni C., Urbani F., Funaro A., Malavasi F. (1996) Secretion of IFN-c, IL-6, granulocyte-macrophage colony-stimulating factor and IL-10 cytokines after activation of human purified T lymphocytes upon CD38 ligation. Cell Immunol;173:192–197. 8. Ausiello C.M., Urbani F., la Sala A., Funaro A., Malavasi F. (1995) CD38 ligation induces discrete cytokine mRNA expression in human cultured lymphocytes. Eur J Immunol;25:1477–1480. 9. Dianzani U., Funaro A., DiFranco D., Garbarino G., Bragardo M., Redoglia V., Buonfiglio D., De Monte L.B., Pileri A., Malavasi F. (1994) Interaction between endothelium and CD4+ CD45RA+ lymphocytes. Role of the human CD38 molecule. J Immunol;153:952– 959. 10. Deaglio S., Aydin S., Vaisitti T., Bergui L., Malavasi F. (2008) CD38 at the junction between prognostic marker and therapeutic target. Trends Mol Med;14:210–218. 11. Guse A.H. (2005) Second messenger function and the structure–activity relationship of cyclic adenosine diphosphoribose (cADPR). FEBS J;272:4590–4597. 12. Zhang H., Graeff R., Lee H.C., Hao Q. (2013) Crystal structures of human CD38 in complex with NAADP and ADPRP. Messenger;2:44–53. 13. Lee H. (2011) Cyclic ADP-ribose and NAADP: fraternal twin messengers for calcium signaling. Sci China Life Sci;54:699–711. 14. Lin H. (2007) Nicotinamide adenine dinucleotide: beyond a redox coenzyme. Org Biomol Chem;5:2541–2554. 15. Liu Q., Kriksunov I.A., Jiang H., Graeff R., Lin H., Lee H.C., Hao Q. (2008) Covalent and noncovalent intermediates of an NAD utilizing enzyme, human CD38. Chem Biol;15:1068–1078. 16. Liu Q., Kriksunov I.A., Graeff R., Munshi C., Lee H.C., Hao Q. (2006) Structural basis for the mechanistic understanding of human CD38-controlled multiple catalysis. J Biol Chem;281:32861–32869. 17. Sauve A.A., Schramm V.L. (2002) Mechanism-based inhibitors of CD38: a mammalian cyclic ADP-ribose synthetase. Biochemistry;41:8455–8463. 18. Muller-Steffner H.M., Malver O., Hosie L., Oppenheimer N.J., Schuber F. (1992) Slow-binding inhibition of NAD+ glycohydrolase by arabino analogues of betaNAD. J Biol Chem;267:9606–9611. 19. Sauve A.A., Deng H., Angeletti R.H., Schramm V.L. (2000) A covalent intermediate in CD38 is responsible for ADP-ribosylation and cyclization reactions. J Am Chem Soc;122:7855–7859. 20. Chen Z., Kwong A.K.Y., Yang Z. (2011) Studies of the synthesis of nicotinamide nucleoside and nucleotide analogues and their inhibitions towards CD38 NADase. Heterocycles;83:2837–2850. 21. Kwong A.K., Chen Z., Zhang H., Leung F.P., Lam C.M., Ting K.Y., Zhang L.R., Hao Q., Zhang L.H., Lee H.C. (2012) Catalysis-based inhibitors of the calcium signaling function of CD38. Biochemistry;51:555–564. 22. Wang S., Zhu W., Wang X., Li J., Zhang K., Zhang L.R., Zhao Y.J., Lee H.C., Zhang L.H. (2014) Design, Chem Biol Drug Des 2015; 86: 1411–1424

synthesis and SAR studies of NAD analogues as potent inhibitors towards CD38 NADase. Molecules;19:15754–15767. 23. Zhou Y., Ting K.Y., Lam C.M.C., Kwong A.K.Y., Xia J., Jin H., Liu Z., Zhang L.R., Lee H.C., Zhang L.H. (2012) Design, synthesis and biological evaluation of noncovalent inhibitors of human CD38 NADase. ChemMedChem;7:223–228. 24. Wu D., Ting K., Duan Y., Li N., Li J., Zhang L.R., Lee H.C., Zhang L.H. (2013) Synthesis and activity of novel indole derivatives as inhibitors of CD38. Acta Pharm Sin B;3:245–253. 25. Escande C., Nin V., Price N.L., Capellini V., Gomes A.P., Barbosa M.T., O’Neil L., White T.A., Sinclair D.A., Chini E.N. (2013) Flavonoid apigenin is an inhibitor of the NAD+ase CD38: implications for cellular NAD+ metabolism, protein acetylation, and treatment of metabolic syndrome. Diabetes;62:1084–1093. 26. Kellenberger E., Kuhn I., Schuber F., Muller-Steffner H. (2011) Flavonoids as inhibitors of human CD38. Bioorg Med Chem Lett;21:3939–3942. 27. Kellenberger E., Kuhn I., Schuber F., Muller-Steffner H. (2014) Inhibition of glioma progression by a newly discovered CD38 inhibitor. Int J Cancer;136:1422–1433. 28. Liu Q., Kriksunov I.A., Graeff R., Munshi C., Lee H.C., Hao Q. (2005) Crystal structure of human CD38 extracellular domain. Structure;13:1331–1339. 29. Liu Z., Graeff R.M., Jin H., Zhang L., Zhang L. (2013) Studies on CD38 inhibitors and their application to cADPR-mediated Ca2+ signaling. Messenger;2:19–32. 30. Dong M., Si Y.Q., Sun S.Y., Pu X.P., Yang Z.J., Zhang L.R., Zhang L.H. et al. (2011) Design, synthesis and biological characterization of novel inhibitors of CD38. Org Biomol Chem;9:3246–3257. 31. Moreau C., Liu Q., Graeff R., Wagner G.K., Thomas M.P., Swarbrick J.M., Shuto S., Lee H.C., Hao Q., Potter B.V.L. (2013) CD38 structure-based inhibitor design using the N1-cyclic inosine 50 -diphosphate ribose template. PLoS ONE;8:e66247. 32. Clark M., Cramer R.D., Van Opdenbosch N. (1989) Validation of the general purpose tripos 5.2 force field. J Comput Chem;10:982–1012. 33. Gasteiger J., Marsili M. (1980) Iterative partial equalization of orbital electronegativity—a rapid access to atomic charges. Tetrahedron;36:3219–3228. 34. Berman H.M., Westbrook J., Feng Z., Gilliland G., Bhat T.N., Weissig H., Shindyalov I.N., Bourne P.E. (2000) The protein data bank. Nucleic Acids Res;28:235–242. 35. Weiner S.J., Kollman P.A., Case D.A., Singh U.C., Ghio C., Alagona G., Profeta S., Weiner P. (1984) A new force field for molecular mechanical simulation of nucleic acids and proteins. J Am Chem Soc;106:765–784. 36. Sanner M.F. (1999) Python: a programming language for software integration and development. J Mol Graph Model;17:57–61. 37. Morris G.M., Goodsell D.S., Halliday R.S., Huey R., Hart W.E., Belew R.K., Olson A.J. (1998) Automated 1423

Zhang et al.

docking using a Lamarckian genetic algorithm and an empirical binding free energy function. J Comput Chem;19:1639–1662. 38. Wolber G., Langer T. (2004) LigandScout: 3-D pharmacophores derived from protein-bound ligands and their use as virtual screening filters. J Chem Inf Model;45:160–169. 39. St ahle L., Wold S. (1987) Partial least squares analysis with cross-validation for the two-class problem: A Monte Carlo study. J Chemometr;1:185–196. 40. Wold S., Albano C., Dunn W.J. III, Edlund U., Esbensen K., Geladi P., Hellberg S., Johansson E., Lindberg W., Sjo¨stro¨m M. (1984) Multivariate Data Analysis in Chemistry. In: Kowalski B., editor. Chemometrics: Mathematics and Statistics in Chemistry: Springer Netherlands; p.17–95. 41. Wold S. (1978) Cross-validatory estimation of the number of components in factor and principal components models. Technometrics;20:397–405. 42. Cramer R.D., Patterson D.E., Bunce J.D. (1988) Comparative molecular field analysis (CoMFA). 1. Effect of shape on binding of steroids to carrier proteins. J Am Chem Soc;110:5959–5967.

1424

43. Liu Q., Kriksunov I.A., Moreau C., Graeff R., Potter B.V.L., Lee H.C., Hao Q. (2007) Catalysis-associated conformational changes revealed by human CD38 complexed with a non-hydrolyzable substrate analog. J Bioll Chem;282:24825–24832. 44. Liu Q., Kriksunov I.A., Graeff R., Lee H.C., Hao Q. (2007) Structural basis for formation and hydrolysis of the calcium messenger cyclic ADP-ribose by human CD38. J Biol Chem;282:5853–5861.

Note a

Sybyl, version X-2.0 (2012) St. Louis, MO: Tripos Associates.

Supporting Information Additional Supporting Information may be found in the online version of this article: Table S1. CD38 non-covalent inhibitors. Table S2. CD38 covalent inhibitors.

Chem Biol Drug Des 2015; 86: 1411–1424

Comparative Analysis of Pharmacophore Features and Quantitative Structure-Activity Relationships for CD38 Covalent and Non-covalent Inhibitors.

In the past decade, the discovery, synthesis, and evaluation for hundreds of CD38 covalent and non-covalent inhibitors has been reported sequentially ...
3MB Sizes 0 Downloads 11 Views