Pharmacophore Based Virtual Screening Approach to Identify Selective PDE4B Inhibitors

authors:

avatar Anand Gaurav a , * , avatar Vertika Gautam b

Faculty of Pharmaceutical Sciences, UCSI University, No. 1, UCSI Heights, Jalan Menara Gading, Taman Connaught, 56000 Kuala Lumpur, Federal Territory of Kuala Lumpur, Kuala Lumpur, Malaysia.
Department of Chemistry, Faculty of Science, University of Malaya 50603, Kuala Lumpur, Malaysia.

how to cite: Gaurav A, Gautam V. Pharmacophore Based Virtual Screening Approach to Identify Selective PDE4B Inhibitors. Iran J Pharm Res. 2017;16(3):e124985. https://doi.org/10.22037/ijpr.2017.2100.

Abstract

Phosphodiesterase 4 (PDE4) has been established as a promising target in asthma and chronic obstructive pulmonary disease. PDE4B subtype selective inhibitors are known to reduce the dose limiting adverse effect associated with non-selective PDE4B inhibitors. This makes the development of PDE4B subtype selective inhibitors a desirable research goal. To achieve this goal, ligand based pharmacophore modeling approach is employed. Separate pharmacophore hypotheses for PDE4B and PDE4D inhibitors were generated using HypoGen algorithm and 106 PDE4 inhibitors from literature having thiopyrano [3,2-d] Pyrimidines, 2-arylpyrimidines, and triazines skeleton. Suitable training and test sets were created using the molecules as per the guidelines available for HypoGen program. Training set was used for hypothesis development while test set was used for validation purpose. Fisher validation was also used to test the significance of the developed hypothesis. The validated pharmacophore hypotheses for PDE4B and PDE4D inhibitors were used in sequential virtual screening of zinc database of drug like molecules to identify selective PDE4B inhibitors. The hits were screened for their estimated activity and fit value. The top hit was subjected to docking into the active sites of PDE4B and PDE4D to confirm its selectivity for PDE4B. The hits are proposed to be evaluated further using in-vitro assays.

Introduction

Prevalence of inflammatory diseases of respiratory tract i.e. asthma and COPD has increased in recent years, with more than 200 million people affected by it worldwide. Most of the mortality related to these inflammatory disorders occurs in low and low-middle income countries (1).

Phosphodiesterase 4 (PDE4), a predominant family of phosphodiesterase (PDE) enzymes expressed in immune and inflammatory cells, includes three subtypes PDE4A, PDE4B and PDE4D. Inhibition of PDE4 has been shown to suppress a diverse spectrum of inflammatory responses in-vitro and in-vivo (2-5). More importantly, many PDE4 inhibitors in development are efficacious in animal models of various inflammatory disorders, such as asthma, COPD, psoriasis, inflammatory bowel diseases, and rheumatoid arthritis (3, 6, 7), as well as in clinical trials for asthma and COPD (8-10).

The development of PDE4 inhibitors has been slowed down due to narrow therapeutic window of most of the compounds. A major reason for their poor clinical results is the consequence of dosing limitation caused by side effects such as nausea and emesis (11). Recent findings in PDE4D knockout mice suggest that an inhibitor with PDE4B selectivity should retain many beneficial anti-inflammatory effects without the unwanted side effects (12). The study also established that PDE4D inhibition is responsible for the dose limiting side effects. Some other studies have proven that selective PDE4B inhibitors have potent anti-inflammatory effects in-vitro and in-vivo. Investigation in ferrets also showed significantly less emesis with this compound compared with the non-selective PDE4 inhibitor cilomilast (13). Thus, PDE4B has been established as an extremely attractive target for design of anti-inflammatory drugs, particularly for asthma and COPD.

The highly conserved catalytic domain of PDE4 isozymes makes the design of inhibitors with PDE4 subtype selectivity a challenging task, nevertheless subtype selective PDE4 inhibitors have recently been described (14, 15). Only a few amino acids are non-conserved in N-terminal regulatory domain UCR2 (i.e Phe in PDE4D vs Tyr in PDE4B) and C-terminal domain CR3 (i.e Leu in PDE4D vs Gln in PDE4B) (16, 17). These minor differences in the regulatory domains have been exploited to design selective PDE4B or PDE4D inhibitors so far (16-18).

The availability of PDE4B and PDE4D inhibition data for recently reported PDE4 inhibitors allows the development of pharmacophore models of PDE4B and PDE4D inhibitors (19-21). Pharmacophore models also help in the identification of structural features which can differentiate between the two receptor subtypes. The information obtained can be used for design of more selective and potent PDE4B inhibitors with hitherto new structures. The pharmacophore models of PDE4B and PDE4D inhibitors can be used to screen databases of drug like compounds in a sequential manner to identify novel leads as selective PDE4B inhibitors. Pharmacophore model based virtual screening has proved to be a useful strategy for identification of novel leads in the past (22-32). In the present study pharmacophore models of both PDE4B and PDE4D inhibitors has been developed and validated. The pharmacophore models were then used for sequential virtual screening to identify novel selective PDE4B inhibitors. The hits were screened for their estimated activity and fit value. Their selectivity for PDE4B was confirmed by docking studies.

Experimental

Data set

Selective PDE4B inhibitors belonging to thiopyrano[3,2-d] Pyrimidines,2 arylpyrimidines and triazines class reported recently, along-with their PDE4B and PDE4D inhibitory activities, were used for the present study (Figure 1) (19-21). The molecular structures and IC50 of the above series were taken from the original papers. Numbers used in original papers were used to denote molecules belonging to triazine series while numbers used in original papers for molecules belonging to 2-arylpyrimidine and thiopyrano[3,2-d]Pyrimidine series were suffixed with a and b respectively.

3D QSAR pharmacophore modeling

Pharmacophore generation

Pharmacophore modeling is the most widely used method for identification of essential structural features required for biological activity. In the present study, HypoGen algorithm was applied to build the 3D QSAR pharmacophore models for both PDE4B and PDE4D inhibitors using DS V2.0 software (Accelrys Inc., San Diego, CA, USA) (33).

For the study, 75 molecules, with activity values (IC50) between 3.0 nM and 18755 nM were selected as training set, which was used to engender the hypotheses. The training set selected complies with the requirements specified in the literature. To validate the hypothesis, the test set was prepared using the specified requirements. Test set contains 24 molecules having wide range of activity values. Sketch function of DS was used to sketch the two-dimensional (2D) chemical structures of all molecules which were later converted into 3D structures. Maximum of 250 conformations were generated for each molecule using the best conformation model generation method based on CHARMm force field and Poling algorithm (34). Those conformations with energy higher than 20 kcal/mol from the global minimum were rejected. Molecules with their conformational models were then submitted to DS for generating hypotheses.

General structures of 2-arylpyrimidines (A), triazines (A) and thiopyrano[3,2-d]Pyrimidine (B and C).
Hypo1B and Hypo1D chemical features with their geometric parameters
A. Most active PDE4B inhibitor (54b) aligned with Hypo1B, B. Least active PDE4B inhibitor (10) aligned with Hypo1B, C. Most active PDE4D inhibitor (29) aligned with Hypo1D, D. Least active PDE4D inhibitor (10) aligned with Hypo1D
The difference in costs between HypoGen runs and the scrambled runs for PDE4B and PDE4D. The 95% confidence level was selected
Structures of hits obtained using pharmacophore based virtual screening
Most selective PDE4B inhibitor (ZINC09157416) identified by virtual screening aligned with Hypo1B and Hypo1D
Table 1

Information of statistical significance and predictive power presented in cost values measured in bits for the top 10 hypotheses as a result of automated 3D QSAR pharmacophore generation for PDE4B

Hypo no.Total costCost differenceaRMSbCorrelationFeaturesbMax fit
Hypo1B365.722143.3781.860.9571HBA, HBA, HBA5.8678
Hypo2B369.858139.2421.890.9362HBA, HBA, H5.7607
Hypo3B376.722132.3781.930.9208HBA, HBA, H5.2594
Hypo4B382.828126.1721.970.9068HBA, HBA, H, H6.6554
Hypo5B383.112125.9881.980.9054HBA, HBA, H5.2458
Hypo6B386.676122.4241.990.8972HBA, HBA, H5.0619
Hypo7B387.872121.2282.010.8933HBA, HBA, H5.4563
Hypo8B391.358117.7422.030.8862HBA, HBA, H4.8644
Hypo9B391.861117.2392.030.8858HBA, HBA, H4.6476
Hypo10B393.315115.7852.040.8815HBA, HBA, H4.7971
Table 2

Information of statistical significance and predictive power presented in cost values measured in bits for the top 10 hypotheses as a result of automated 3D QSAR pharmacophore generation for PDE4D

Hypo no.Total costCost differencecRMSCorrelationFeaturesMax fit
Hypo1D334.571164.4191.660.9563HBA, HBA, H, H, H8.1515
Hypo2D335.16163.8301.660.9336HBA, HBA, H, H8.0525
Hypo3D339.583159.4071.700.9247HBA, HBA, H6.0897
Hypo4D340.037158.9531.670.8712HBA, HBA, H, H,5.6251
Hypo5D346.347152.6431.740.8341HBA, HBA, H, H, H6.9568
Hypo6D346.670152.3201.700.8027HBA, HBA, H, H, H4.5456
Hypo7D348.692150.2981.760.7991HBA, HBA, H, H6.2123
Hypo8D350.222148.7681.770.7892HBA, HBA, H, H6.7776
Hypo9D350.437148.5531.760.7554HBA, HBA, H, H, H6.3886
Hypo10D350.729148.2611.740.7332HBA, HBA, H, H, H4.8890
Table 3

Actual and estimated activity of the training set molecules based on the pharmacophore model Hypo1B.

Fit valuedLogIC50LogIC50 (predicted)Errore
13.90948.80997.80801.0018
25.08964.00734.0906-0.0832
44.44077.57306.58470.9884
54.55575.94805.31990.6282
84.23955.46816.0481-0.5800
94.05178.23488.4804-0.2455
103.90969.83928.80771.0316
123.90907.28626.80890.4773
144.00437.73286.58951.1433
154.46236.73935.53491.2045
184.66955.56455.05790.5066
193.82977.23996.99170.2483
204.59705.39365.22490.1688
224.14333.82864.2695-0.4409
234.26214.90535.9959-1.0906
243.90794.90535.8114-0.9062
275.04845.48485.8855-0.4007
284.59212.48492.23600.2489
294.81385.10604.72560.3804
314.62694.23414.15590.0782
325.10695.99655.65080.3457
334.68776.99396.01610.9779
10a4.39585.70385.68800.0158
12a3.90143.52644.8266-1.3002
13b4.65894.04314.0824-0.0393
14a4.01126.53676.5736-0.0369
14b5.23464.78754.9567-0.1692
15a4.67774.78755.0390-0.2515
16b4.66303.40124.0729-0.6717
17a4.64427.93747.11610.8213
17b5.46072.39792.5361-0.1382
18a3.89348.21617.84500.3711
18b4.93054.78754.45680.3307
19a4.03546.75696.51800.2390
1b5.66173.21893.4733-0.2544
20b5.13833.52643.9785-0.4522
21a3.91155.34716.8032-1.4561
21b4.99234.94164.31470.6270
22a5.06775.39365.5410-0.1474
22b5.14104.94164.9722-0.0305
23a4.61395.01065.1859-0.1753
24b4.53826.06385.36030.7035
26a3.91146.84596.80350.0424
26b5.45703.78423.54460.2396
27a3.91167.09016.80310.2870
27b5.57183.63763.9802-0.3426
28b5.35242.30262.4854-0.1829
29a5.10795.76835.04850.7198
29b5.44565.52155.27090.2506
2a3.88256.06386.8700-0.8062
2b5.09103.36733.28730.0800
30a5.13795.24704.67930.5678
31a5.41633.52643.33830.1881
31b5.51353.82863.51450.3142
32b5.51082.56502.32080.2442
33a5.08962.70812.49060.2175
33b5.55192.94443.0262-0.0817
34a5.18241.91691.87680.0401
35a5.20042.70812.43550.2725
35b5.45363.40123.25250.1487
36b5.56152.63913.0039-0.3649
37b5.57232.11632.3791-0.2628
39b5.61942.30262.8707-0.5681
3a3.90804.94164.81130.1303
3b4.80323.09104.7500-1.6590
44b5.53051.52612.0753-0.5493
45b5.65352.77262.7922-0.0196
47b5.64934.15893.80180.3571
48b5.48603.89183.57780.3140
49b5.62252.29252.8636-0.5711
4a3.90997.17016.80700.3632
53b5.52562.10412.08660.0176
54b5.75741.09861.3529-0.2542
55b5.74621.77501.57870.1962
8a4.05754.78755.4670-0.6795
Table 4

Actual and estimated activity of the training set molecules based on the pharmacophore model Hypo1D

Fit valueLogIC50LogIC50 (predicted)Error
15.80337.74077.11090.6297
25.75247.59947.22810.3713
46.38468.18318.7724-0.5893
55.83646.76277.0346-0.2718
85.93877.07416.79910.2751
95.44158.63258.9440-0.3115
105.549610.01059.69510.3154
125.80376.58897.1100-0.5211
145.70436.75587.3388-0.5830
156.17656.58766.25140.3361
185.90545.52555.8757-0.3502
195.60425.33275.5694-0.2367
205.99515.38915.6693-0.2803
226.91642.99572.54790.4478
235.60155.59105.57550.0155
245.94045.59105.7952-0.2042
276.70115.36135.04370.3176
287.64382.56502.8729-0.3079
297.65551.94591.84600.0999
316.90784.14314.5677-0.4246
326.64225.07525.1791-0.1039
336.09014.56444.45050.1138
10a5.75117.31327.23110.0822
12a5.67314.40674.4106-0.0039
13b5.62026.52216.5325-0.0104
14a5.57037.78327.64740.1358
14b5.72706.90787.2865-0.3788
15a5.65767.17017.4463-0.2761
16b5.76355.66995.20260.4673
17a5.93509.39279.8075-0.4149
17b5.86045.63485.9794-0.3446
18a5.60629.61589.56460.0512
18b5.90607.49557.8744-0.3788
19a4.86428.00648.2731-0.2668
1b5.90027.93747.88780.0496
20b5.65076.90786.46230.4455
21a6.04627.31327.5516-0.2384
21b5.65468.90928.45320.4560
22a5.80467.60097.10790.4930
22b5.80027.37787.11800.2598
23a5.86897.17016.95990.2102
24b5.79266.06386.1355-0.0717
26a5.76769.30579.19310.1125
26b5.89207.31326.90660.4066
27a5.83199.20039.04500.1553
27b5.96637.64977.7356-0.0859
28b5.85256.75696.9975-0.2406
29a5.82679.08259.05700.0255
29b5.96078.24288.7484-0.5056
2a4.65298.13158.7597-0.6281
2b5.55777.17017.6763-0.5062
30a5.79138.57558.13850.4370
31a5.70567.31327.3358-0.0225
31b5.91458.13158.8549-0.7233
32b5.91847.17017.8458-0.6757
33a5.75247.43847.22810.2103
33b5.92966.25386.8200-0.5662
34a5.88377.97257.92580.0467
35a5.74048.03927.25580.7834
35b5.89565.66996.8983-1.2284
36b5.94386.08686.7874-0.7006
37b5.85325.99156.9959-1.0044
39b5.91996.90786.84240.0654
3a5.66107.64977.43850.2112
3b5.92945.56076.8206-1.2599
44b5.93486.42976.8080-0.3782
45b5.93717.17016.80270.3675
47b5.95397.17016.76420.4059
48b5.94146.39696.7930-0.3960
49b5.89566.32796.8984-0.5704
4a5.57358.93597.63991.2960
53b5.90127.24426.88540.3588
54b5.81077.17017.09380.0763
55b5.92436.69706.8322-0.1352
8a5.71057.31327.3247-0.0114
Table 5

Actual and estimated activity of the test set molecules based on the pharmacophore model Hypo1B

Log (Activ)Log (Estimate)Error
75.52555.34550.1800
117.27037.23450.0358
167.52947.45670.0727
219.08427.35631.7279
266.66065.34551.3151
304.83633.83551.0008
347.36147.32530.0361
11a6.29165.73540.5562
16a4.21954.3323-0.1128
20a5.39365.34520.0484
24a4.35674.9752-0.6185
28a6.39695.34531.0516
32a2.94442.9968-0.0524
1a5.24704.86590.3811
15b4.78755.8364-1.0489
19b3.63764.7264-1.0888
23b4.94163.85631.0853
30b4.02544.3324-0.3070
34b3.63763.7254-0.0878
38b2.99573.2232-0.2275
46b1.64872.8675-1.2188
52b2.01492.2484-0.2335
56b1.33502.4543-1.1193
12b6.77997.2194-0.4395
Table 6

Actual and estimated activity of the test set molecules based on the pharmacophore model Hypo1D

Log (Activ)Log (Estimate)Error
77.30597.15330.1526
117.65307.9863-0.3333
167.72677.21210.5146
216.78676.9891-0.2024
266.52366.33340.1902
304.88284.62760.2552
346.39696.6676-0.2707
11a8.98728.22230.7649
16a6.89776.51220.3855
20a7.93747.22310.7143
24a6.63336.9098-0.2765
28a8.47648.29570.1807
32a7.37787.5762-0.1984
1a7.54967.8894-0.3398
15b6.04035.72040.3199
19b6.67206.8732-0.2012
23b7.69627.9909-0.2947
30b7.09016.79250.2976
34b6.67206.7623-0.0903
38b7.00317.4052-0.4021
46b6.36306.8437-0.4807
52b7.17017.4923-0.3222
56b6.39696.5427-0.1458
12b8.13158.4072-0.2757
Table 7

Fit values of hits with Hypo1B and Hypo1D

Fit value (Hypo1B)Fit value (Hypo1D)
ZINC091574164.388860.13199
ZINC195216604.334580.74828
ZINC723365584.335840.22303
ZINC784163714.502320.52406
ZINC195216584.360430.40399
Table 8.

CDOCKER energy of hits and standard (33b) with PDE4B and PDE4D

-CDOCKER energy (PDE4B)-CDOCKER energy (PDE4D)
34b -13.0258-22.3445
ZINC09157416-15.9889-26.7976

Automated 3D QSAR pharmacophores were produced by comparing the PDE4B and PDE4D inhibitory activity values of molecules in the training set separately. This helps in identifying the features that are common with the active compounds, but excludes common features for the inactive compounds within conformational allowable regions of space. Selecting the chemical features is one of the most important steps in generating a pharmacophore. While generating hypotheses, HBA (hydrogen bond acceptor), HBD (hydrogen bond donor), and H (hydrophobic), features were selected based on the training set molecules. The number of features allowed in the model were kept in the range 0-5. The ‘Uncertainty’ values for all the 75 molecules in the training set were set to 2.0, and the default values for other parameters were kept constant. Subsequently, pharmacophore models were computed and the 10 top scoring hypotheses for both PDE4B and PDE4D inhibition were selected for further study. The qualities of the hypotheses were reliant on the fixed cost, null cost, and total cost values (35).

Assessment of pharmacophore quality

Quality of the developed pharmacophore was assessed using three different methods. Initially, cross validation was performed by the Fischer’s randomization test. Secondly, the prediction of the activity values of the test set was performed. The correlation between the experimental and predicted activities was used to assess predictive ability of the model. All queries were addressed using the ligand pharmacophore mapping protocol.

Virtual Screening

The validated pharmacophore model (Hypo1B and Hypo1D) of PDE4B and PDE4D inhibitors was used as a query in a sequential manner to search the zinc database. Zinc is a comprehensive database of small molecules containing a total of 17,900,742 drug like molecules (36). In the first step ligand pharmacophore mapping module of DS was used along with Hypo1B as the pharmacophore model and zinc database as the database. In the next step, hits mapping to the pharmacophore model Hypo1B were retrieved and hit compounds showing Hypo1B estimated IC50 less than 20 nM were selected and subsequently subjected to screening using the validated pharmacophore model Hypo1D in the same manner as in the previous step. The hit compounds were chosen that showed Hypo1D estimated fit value less than 4.

Docking studies were used to confirm the selectivity of the hits obtained using the pharmacophore based virtual screening. The most PDE4B selective hit determined by the fit values for Hypo1B and Hypo1D and the most selective ligand from the series used for pharmacophore development i.e. 34b, were docked into the active sites of PDE4B (PDB ID: 4NW7) and PDE4D (PDB ID: 1Y2B). First the protein structures were prepared using the automatic protein preparation module of DS V2.0 software using the default parameters. The structures of the identified hit as well as the standard molecule (34b) were prepared using the prepare ligand module of DS V2.0 software. Docking of the prepared ligands into the active site of the prepared structures of PDE4B and PDE4B was carried out using CDOCKER program available in DS V2.0 software (Accelrys Inc., San Diego, CA, USA) with default parameters (37). The ratio of PDE4B/PDE4D docking scores was used as measure of PDE4B selectivity. The higher is the ratio the greater is the PDE4B selectivity.

Results and Discussion

3D QSAR pharmacophore modeling

Pharmacophore generation

The top scoring model (Hypo1B) for PDE4B inhibition consist of three HBA which established the highest cost difference (143.378), best correlation coefficient (0.9571), maximum fit value (5.8678) and lowest root mean square (RMS) of 1.86 (Table 1). The results revealed the importance of HBA in PDE4B receptor antagonist activity. The fixed and the null cost values were 236.38 and 509.10, respectively (Table 1). Difference between these two costs (143.378) was greater than 70 bits which showed that the model has over 90% statistical significance. A good pharmacophore model should also have the configuration cost lower than 17, and it was found to be 12.53 for the generated pharmacophore hypotheses. Hypo1B showed correlation coefficient value of 0.9571, demonstrating its good prediction ability.

Top scoring model (Hypo1D) for PDE4D inhibition consists of two HBA and three H with highest cost difference (164.419), best correlation coefficient (0.9563), maximum fit value (8.1515), and lowest root mean square (RMS) of 1.66 (Table 2). As in the case of Hypo1B, HBA was found to be important for PDE4D receptor antagonist activity although there is additional H in this case. Difference between fixed and null costs (164.419) showed that the model has over 90% statistical significance. The configuration cost was also sufficiently low at 12.49. Hypo1D showed correlation coefficient value of 0.9563 (Table 2). Based on statistical parameters Hypo1B and Hypo1D were selected as the best hypothesis for PDE4B and PDE4D inhibition respectively and were employed for further analyses.

Figure 2 shows Hypo1B, and Hypo1D chemical features with their geometric parameters while Molecules with highest and lowest activity in the training set aligned to Hypo1B and Hypo1D are shown in Figure 3. The prediction accuracy of both the models was verified using the training set and the activity of each molecule in training set was estimated by regression analysis.

The experimental and predicted activities by Hypo1B and Hypo1D for 75 training set molecules are shown in Tables 3 and 4 respectively. Data clearly shows the good agreement between predicted and experimental IC50 values.

dFit value indicates how well the features in the pharmacophore overlap the chemical features in the molecule. Fit value = weight x [max(0,1 - SSE)] where SSE = (D/T)2, D = displacement of the feature from the center of the location constraints and T = the radius of the location constraint sphere for the feature (tolerance).

eDifference between the predicted and experimental values. ‘+’ indicates that the predicted IC50 is higher than the experimental IC50; ‘-’ indicates that the predicted IC50 is lower than the experimental IC50; a value of 0 indicates that the predicted IC50 is equal to the experimental IC50.

Close examination of the pharmacophore models Hypo1B and Hypo1D reveals the structural features of an inhibitor which can differentiate well between the two receptors. The conformation which can allow –COOH at R3 and hydrophobic groups like halogen atoms in the aromatic ring (Ar) to orient properly for interaction with CR3 will show significant selectivity for PDE4B as compared to PDE4D. This is consistent with the findings described previously in the original papers in which these compounds have been reported (21).

Validation of Hypo1B and Hypo1D

The generated hypotheses were validated using standard methods to check whether the best hypotheses are statistically significant and have considerable predictive ability.

Fischer’s randomization method

Fischer’s randomization was used to evaluate the statistical significance of the Hypotheses. Validation was done by generating random spreadsheets for training set molecules, which randomly reassigned activity values to every molecule and subsequently generated the hypotheses using the same features and parameters originated for Hypo1B and Hypo1D. All the randomly generated spreadsheets had higher total cost values and lower correlation coefficient values as can be seen clearly from Figure 4. This suggests that Hypo1B and Hypo1D were not generated by chance.

Test set

Test set was prepared using the same protocol as training set and used to determine whether the hypotheses were able to predict the active molecules other than those present in the training set.

The correlation coefficient (r) for the test set given by Hypo1B was 0.8579 (Table 5) while that by Hypo1D was 0.8299 (Table 6). Test set molecules were classified using the same criteria as used for training set molecules. Thus Hypo1B and Hypo1D were able to estimate the PDE4B and PDE4D inhibition activities respectively with reasonable accuracy.

Virtual Screening

Zinc, a comprehensive database of small drug like molecules was used for the sequential virtual screening using the pharmacophore models. Screening of zinc database using the validated pharmacophore model (Hypo1B) of PDE4B inhibitors retrieved a set of 6015 hits, mapping to the pharmacophore model Hypo1B. The hits comprised of some compounds structurally similar to that of the existing PDE4B inhibitors, and some novel scaffolds.

The 397 hit compounds showing Hypo1B estimated IC50 less than 20 nM were selected and subsequently subjected to screening using the validated pharmacophore model Hypo1D. 5 hit compounds that showed Hypo1 PDE4D estimated fit value less than 4 were identified (Figure 5). Among the hits ZINC09157416 demonstrated the best PDE4B selectivity based on the hit values (Table 7). ZINC09157416 aligned with Hypo1B and Hypo1D is shown in Figure 6.

The results of docking studies of ZINC09157416 and 34b with PDE4B and PDE4D further confirmed the selectivity of ZINC09157416 for PDE4B over PDE4D (Table 8).

Conclusions

Ligand-based pharmacophore models for a diverse class of PDE4B and PDE4D inhibitors were developed. The best pharmacophore models Hypo1B and Hypo1D were validated using different methods to evaluate their predictive power over the diverse test set compounds. Hydrogen bond acceptors were identified to be mainly responsible for PDE4B inhibition while both hydrogen bond acceptors as well as hydrophobic groups were found to be responsible for PDE4D inhibition. The highly predictive pharmacophore hypotheses were further used in sequential virtual screening for identification of selective PDE4B inhibitors. Zinc drug like database was used in virtual screening. The hits from the virtual screening were filtered based on the estimated activity and fit value. Five molecules with different backbones were identified as final hits. The most selective hit molecule ZINC09157416 exhibited better selectivity for PDE4B than the standard compound 34b in the docking studies. The activity of the hit compound has not been reported in the literature as we explored by PubChem and SciFinder Scholar search tools. Thus, the sequential virtual screening strategy using 3D QSAR pharmacophores for PDE4B and PDE4D inhibitors proved to be an effective strategy to identify novel selective PDE4B inhibitors.

References

  • 1.

    Garg D, Gandhi T, Gopi Mohan C. Exploring QSTR and toxicophore of hERG K+ channel blockers using GFA and HypoGen techniques. J. Mol. Graph. Model. 2008;26:966-76. [PubMed ID: 17928249].

  • 2.

    Jin SL, Ding SL, Lin SC. Phosphodiesterase 4 and its inhibitors in inflammatory diseases. Chang. Gung. Med. J. 2012;35:197-210. [PubMed ID: 22735051].

  • 3.

    Press NJ, Banner KH. PDE4 inhibitors - a review of the current field. Prog. Med. Chem. 2009;47:37-74. [PubMed ID: 19328289].

  • 4.

    Souness JE, Aldous D, Sargent C. Immunosuppressive and anti-inflammatory effects of cyclic AMP phosphodiesterase (PDE) type 4 inhibitors. Immunopharmacology. 2000;47:127-62. [PubMed ID: 10878287].

  • 5.

    Baumer W, Hoppmann J, Rundfeldt C, Kietzmann M. Highly selective phosphodiesterase 4 inhibitors for the treatment of allergic skin diseases and psoriasis. Inflamm. Allergy Drug Targets. 2007;6:17-26. [PubMed ID: 17352685].

  • 6.

    Burgin AB, Magnusson OT, Singh J, Witte P, Staker BL, Bjornsson JM, Thorsteinsdottir M, Hrafnsdottir S, Hagen T, Kiselyov AS, Stewart LJ, Gurney ME. Design of phosphodiesterase 4D (PDE4D) allosteric modulators for enhancing cognition with improved safety. Nat. Biotechnol. 2010;28:63-70. [PubMed ID: 20037581].

  • 7.

    Spina D. PDE4 inhibitors: current status. Br. J. Pharmacol. 2008;155:308-15. [PubMed ID: 18660825].

  • 8.

    Barnette MS. Phosphodiesterase 4 (PDE4) inhibitors in asthma and chronic obstructive pulmonary disease (COPD). Prog. Drug Res. 1999;53:193-229. [PubMed ID: 10616299].

  • 9.

    Diamant Z, Spina D. PDE4-inhibitors: a novel, targeted therapy for obstructive airways disease. Pulm Pharmacol. Ther. 2011;24:353-60. [PubMed ID: 21255672].

  • 10.

    Torphy TJ. Phosphodiesterase isozymes: molecular targets for novel antiasthma agents. Am. J. Respir. Crit. Care. Med. 1998;157:351-70. [PubMed ID: 9476844].

  • 11.

    Beghe B, Rabe KF, Fabbri LM. Phosphodiesterase-4 inhibitor therapy for lung diseases. Am. J. Respir. Crit. Care. Med. 2013;188:271-8. [PubMed ID: 23656508].

  • 12.

    Jin SL, Goya S, Nakae S, Wang D, Bruss M, Hou C, Umetsu D, Conti M. Phosphodiesterase 4B is essential for T(H)2-cell function and development of airway hyperresponsiveness in allergic asthma. J. Allergy Clin. Immunol. 2010;126:1252-9 e12. [PubMed ID: 21047676].

  • 13.

    Robichaud A, Savoie C, Stamatiou PB, Tattersall FD, Chan CC. PDE4 inhibitors induce emesis in ferrets via a noradrenergic pathway. Neuropharmacology. 2001;40:262-9. [PubMed ID: 11114405].

  • 14.

    Donnell AF, Dollings PJ, Butera JA, Dietrich AJ, Lipinski KK, Ghavami A, Hirst WD. Identification of pyridazino[4,5-b]indolizines as selective PDE4B inhibitors. Bioorg. Med. Chem. Lett. 2010;20:2163-7. [PubMed ID: 20202838].

  • 15.

    Kranz M, Wall M, Evans B, Miah A, Ballantine S, Delves C, Dombroski B, Gross J, Schneck J, Villa JP, Neu M, Somers DO. Identification of PDE4B Over 4D subtype-selective inhibitors revealing an unprecedented binding mode. Bioorg. Med. Chem. 2009;17:5336-41. [PubMed ID: 19525117].

  • 16.

    Burgin AB, Magnusson OT, Singh J, Witte P, Staker BL, Bjornsson JM, Thorsteinsdottir M, Hrafnsdottir S, Hagen T, Kiselyov AS. Design of phosphodiesterase 4D (PDE4D) allosteric modulators for enhancing cognition with improved safety. Nat. Biotechnol. 2010;28:63-70. [PubMed ID: 20037581].

  • 17.

    Fox D, Burgin AB, Gurney ME. Structural basis for the design of selective phosphodiesterase 4B inhibitors. Cell. Signal. 2014;26:657-63. [PubMed ID: 24361374].

  • 18.

    Azam MA, Tripuraneni NS. Selective phosphodiesterase 4b inhibitors: a review. Sci. Pharm. 2014;82:453. [PubMed ID: 25853062].

  • 19.

    Naganuma K, Omura A, Maekawara N, Saitoh M, Ohkawa N, Kubota T, Nagumo H, Kodama T, Takemura M, Ohtsuka Y, Nakamura J, Tsujita R, Kawasaki K, Yokoi H, Kawanishi M. Discovery of selective PDE4B inhibitors. Bioorg. Med. Chem. Lett. 2009;19:3174-6. [PubMed ID: 19447034].

  • 20.

    Goto T, Shiina A, Murata T, Tomii M, Yamazaki T, Yoshida K-i, Yoshino T, Suzuki O, Sogawa Y, Mizukami K, Takagi N, Yoshitomi T, Etori M, Tsuchida H, Mikkaichi T, Nakao N, Takahashi M, Takahashi H, Sasaki S. Identification of the5,5-dioxo-7,8-dihydro-6H-thiopyrano[3,2-d]pyrimidine derivatives as highly selective PDE4B inhibitors. Bioorg. Med. Chem. Lett. 2014;24:893-9. [PubMed ID: 24412069].

  • 21.

    Hagen TJ, Mo X, Burgin AB, Fox 3rd D, Zhang Z, Gurney ME. Discovery of triazines as selective PDE4B versus PDE4D inhibitors. Bioorg. Med. Chem. Lett. 2014;24:4031-4. [PubMed ID: 24998378].

  • 22.

    Zhou N, Xu Y, Liu X, Wang Y, Peng J, Luo X, Zheng M, Chen K, Jiang H. Combinatorial pharmacophore-based 3D-QSAR analysis and virtual screening of FGFR1 inhibitors. Int. J. Mol. Sci. 2015;16:13407-26. [PubMed ID: 26110383].

  • 23.

    Xie H, Chen L, Zhang J, Xie X, Qiu K, Fu J. A combined pharmacophore modeling, 3D QSAR and virtual screening studies on imidazopyridines as B-Raf inhibitors. Int. J. Mol. Sci. 2015;16:12307-23. [PubMed ID: 26035757].

  • 24.

    Lu P, Wang Y, Ouyang PK, She J, He M. 3d-qsar based pharmacophore modeling and virtual screening for identification of novel G protein-coupled receptor40 agonists. Curr. Comput. Aided Drug Des. 2015;11:51-6. [PubMed ID: 26022066].

  • 25.

    Xie H, Qiu K, Xie X. Pharmacophore modeling, virtual screening, and 3D-QSAR studies on a series of non-steroidal aromatase inhibitors. Med. Chem. Res. 2015;24:1901-15.

  • 26.

    Kumar A, Roy S, Tripathi S, Sharma A. Molecular docking based virtual screening of natural compounds as potential BACE1 inhibitors: 3D QSAR pharmacophore mapping and molecular dynamics analysis. J. Biomol. Struct. Dyn. 2016;34:239-49. [PubMed ID: 25707809].

  • 27.

    Xie H, Qiu K, Xie X. 3D QSAR studies, pharmacophore modeling and virtual screening on a series of steroidal aromatase inhibitors. Int. J. Mol. Sci. 2014;15:20927-47. [PubMed ID: 25405729].

  • 28.

    The OECD 423 Guideline for Testing of Chemicals Acute Oral Toxicity—acute toxic class method. Paris; 2001.

  • 29.

    Gaurav A, Gautam V, Singh R. An overview on synthetic methodologies and biological activities of pyrazoloquinolines. Mini. Rev. Med. Chem. 2010;10:1194-210. [PubMed ID: 20854254].

  • 30.

    Gaurav A, Gautam V, Singh R. Quantitative structure–activity relationship and design of polysubstituted quinoline derivatives as inhibitors of phosphodiesterase 4. Med. Chem. Res. 2012;21:3087-103.

  • 31.

    Gaurav A, Singh R. 3D QSAR pharmacophore, CoMFA and CoMSIA based design and docking studies on phenyl alkyl ketones as inhibitors of phosphodiesterase 4. Med. Chem. 2012;8:894-912. [PubMed ID: 22741782].

  • 32.

    Gaurav A, Yadav MR, Giridhar R, Gautam V, Singh R. QSAR studies on 4-quinolone derivatives as high-affinity ligands at the benzodiazepine site of brain GABA(A) receptors. Med. Chem. 2009;5:353-8. [PubMed ID: 19689392].

  • 33.

    Li H, Sutter J, Hoffman R. HypoGen: An Automated System for Generating 3D Predictive Pharmacophore Models. In: Guner OF, editor. Pharmacophore Perception, Development and Use in Drug Design. International University Line, La Jolla; 2000. p. 171-89.

  • 34.

    Kristam R, Gillet VJ, Lewis RA, Thorner D. Comparison of conformational analysis techniques to generate pharmacophore hypotheses using catalyst. J. Chem. Inf. Model. 2005;45:461-76. [PubMed ID: 15807512].

  • 35.

    Evans DA, Doman TN, Thorner DA, Bodkin MJ. 3D QSAR methods: Phase and catalyst compared. J. Chem. Inf. Model. 2007;47:1248-57. [PubMed ID: 17477520].

  • 36.

    Irwin JJ, Shoichet BK. ZINC--a free database of commercially available compounds for virtual screening. J. Chem. Inf. Model. 2005;45:177-82. [PubMed ID: 15667143].

  • 37.

    Wu G, Robertson DH, Brooks CL, 3rd, Vieth M. Detailed analysis of grid-based molecular docking: a case study of CDOCKER-A CHARMm-based MD docking algorithm. J. Comput. Chem. 2003;24:1549-62. [PubMed ID: 12925999].