Available online on 15.03.2025 at http://jddtonline.info

Journal of Drug Delivery and Therapeutics

Open Access to Pharmaceutical and Medical Research

Copyright  © 2025 The  Author(s): This is an open-access article distributed under the terms of the CC BY-NC 4.0 which permits unrestricted use, distribution, and reproduction in any medium for non-commercial use provided the original author and source are credited

Open Access Full Text Article  Research Article

Identification of Some DPP-4 Inhibitors Using QSAR Modeling Based Drug Repurposing Approach

Sonu ¹, Arijit Bhattacharya ², Mohan Lal Kori *³

¹ Ph. D. Research scholar RKDF University Bhopal (M.P.), India

² Arijit Bhattacharya, DST-SERB Junior Research Fellow, Punjabi University, Patiala (PB), India

³ Vice Chancellor, Tantya Bhil University Khargone (M.P.), India

 

 

1. INTRODUCTION 

Diabetes is a group of metabolic diseases characterized by hyperglycemia caused by inadequate insulin secretion with or without a simultaneous decrease in hormone action at its receptor ¹. 

Currently, diabetes is the fifth deadliest disease. As per WHO report, about 422 million people worldwide have diabetes, the majority living in low-and middle-income countries, and 1.5 million deaths are directly attributed to diabetes each year. Both the number of cases and the prevalence of diabetes have been steadily increasing over the past few decades². Post-prandial hyperglycemia still remains a problem in the management of type 2 diabetes mellitus. Of all available anti-diabetic drugs, Dipeptidyl peptidase - IV (DPP-4) inhibitors seem to be one of the most effective in reducing post-prandial hyperglycemia³. DPP- is a serine protease, which is present in membrane bound form and plasma soluble form⁴. The enzyme is responsible for degradation of number of biologically important peptides. DPP-IV deactivates GLP-1, so the DPP-IV inhibitors increase the activity of GLP-1. Inactivation of DPP-IV causes the increase in half-life of GLP-1. Most of the DPP-IV inhibitors are peptide derivatives of α-amino acyl pyrrolidines⁵. Currently numbers of DPP-IV inhibitors are available in the market due to high oral bioavailability like Sitagliptin, Vildagliptin, Saxagliptin, Linagliptin, Alogliptin, Gemigliptin, Anagliptin, Teneligliptin, Alogliptin, Trelagliptin and Omarigliptin ⁶. Some of the FDA approved are displayed in Fig. 1

On the basis of these literature observations, it was thought worthwhile to identify some new α-glucosidase inhibitors with better safety profile therefore drug repurposing approach in combination with QSAR was considered to be better choice.

 

 

 

 

Figure 1: FDA approved DPP-4 inhibitors

 

 

Drug repurposing is gaining popularity as a quick and effective method of identifying new therapeutic indications of approved drugs unrelated to their original medical intent, and is successfully moving towards the second phase of clinical trials. In this study, drug repurposing with QSAR based virtual screening was implemented for identification of some DPP-4 inhibitors as new anti-diabetics. To carry of QSAR modeling against DPP-4 inhibitors, a congeneric series of (S)-1-((S)-2-amino-3-phenylpropanoyl) pyrrolidine-2-carbonitrile^7-9, as shown in Fig. 2, having two different types of substitutions i.e. R₁ on phenyl and R₂ on pyrrolidine as well as proper variation in the biological activity was selected on the basis the of thumb rules described by Hansch in his manual¹⁰.

 

Figure 2:  Basic scaffold of DPP-4 inhibitors  used in QSAR modeling.

2.  MATERIALS AND METHOD

The study of DDP4 inhibitors was carried out using conventional various QSAR approaches including Free Wilson, Hansch, and Mixed modeling. For this purpose, various QSAR descriptors were collected from different sources like Hansch Manual, Medicinal chemistry books etc.^{10, 11} and PaDEL software¹². Indicator variables for deriving Free Wilson approach were formulated from the various substituents present on the parent scaffold. Hansch models were developed using substituent’s constants collected from Hansch manual¹⁰ and global properties of the inhibitors, which were calculated from the PaDEL software. QSAR models were derived by DTC QSAR modeling tool¹³. Internal and external validations were carried out by calculating various statistical parameters like Q², R²_traing, R² test, PRESS, F values etc. 

3 RESULTS AND DISCUSSION 

For QSAR modeling, a data set of DPP-4 inhibitors^7-9, was selected on the basis of thumb rules described by Hansch in his manual¹⁰. Data set containing 60 molecules was divided into training set of 45 molecules and test set of 15 molecules. Details about training set and test set are given in the Table 1. Training set was used for determining internal predictive ability whereas test set was used for external predictive ability of the QSAR model. Inhibitory activity data i.e. IC₅₀ was collected from the literature. Here IC₅₀ of the compounds represent their doses in nanomolar concentration required to produce 50% inhibition of DPP-4 enzyme. The given IC₅₀ data is first converted into pIC₅₀ by taking negative log of IC₅₀, where IC₅₀ is in molar concentration. The values of pIC₅₀ of all molecules in the data set are described in Table 1. 

 

 

 

 

 

 

 

 

 

 

 

 

Table 1: Training set and test set data for QSAR analysis of DDP4 inhibitors

 

   * Test set compounds, ^aDose in nanomolar concentration required to produce 50% inhibition of    

            DPP-4, ^b –log IC₅₀

Total number of compounds: 60

Number of trainings:45, number of tests: 15

   

 

3.1 QSAR Model Development 

QSAR modeling was started with Free Wilson approach.  For this purpose various indicator variables were recorded for different functionality at R₁ by assigning value 1 for presence of the particular group and value 0 for absence of that group. Various Free Wilson models were developed taking pIC₅₀ as dependent variable and various combination of indicator variables of R₁ as independent variables using multiple linear regression analysis. No model was found to be significant for predicting activity accurately. Thereafter, study was followed to develop Hansch QSAR models using some local properties of the R₁ substituents. For this purpose stepwise multiple linear regression analysis was performed by considering pC₅₀ as dependent variables and various substituent’s constants which were collected from Hansch manual and Burger’s Medicinal chemistry^10,11, as independent variables. In this analysis also, no models was found to be significant. Study was further subjected to Hansch type QSAR analysis by regression analysis using global proprieties of the inhibitors which were calculated by PaDEL software. The best model generated in this attempt is given in Equation 1. Correlation matrix of best model equation is given in Table 2 to determine mutual correlation among the parameters present in this equation.   These are free from mutual correlation. Values of Dependent (pIC₅₀) and independent variables (Descriptors) which were utilized in deriving Equation 5.1 are given in Table 5.3.

 

 

pIC50 = 14.84288(+/-0.28854)   -0.10092(+/-0.00655) apol -0.20639(+/-0.05306) ATSC8p +0.48633(+/-0.14694) nssO +0.06997(+/-0.02333) VE3_D ……………………………………1

Descriptions about selected variables are as follows: 

apol(PaDEL; 2D)=> 'Negative Contribution' =>Sum of the atomic polarizabilities (including implicit hydrogens)

ATSC8p(Dragon; 2D autocorrelations)=> 'Negative Contribution' =>Centred Broto-Moreau autocorrelation of lag 8 weighted by polarizability 

nssO(PaDEL; 2D)=> 'Positive Contribution' =>Count of atom-type E-State: -O-

VE3_D(Dragon; 2D matrix-based descriptors)=> 'Positive Contribution' =>logarithmic coefficient sums of the last eigenvector from topological distance matrix 

Internal Validation Parameters:

SEE  :0.34922, r^2 :0.91955, r^2 adjusted :0.91151, PRESS :4.87812, F :114.30494 (DF :4, 40)

Leave-One-Out(LOO) Result :

Q2 :0.90415

Rm^2 metrics (after scaling the data):

Average rm^2(LOO):0.86499, Delta rm^2(LOO):0.06214

External Validation Parameters(Without Scaling):

r^2 :0.92569, r0^2 :0.92332, reverse r0^2:0.9083, RMSEP:0.36579, Q2f1/R^2(Pred) :0.92043, Q2f2 :0.92001

External Validation Parameters (After Scaling):

Average rm^2(test) :0.82937

Delta rm^2(test) :0.06883

Error based judgments of test set predictions:

Mean Absolute Error (MAE; 95% data): 0.28167

Standard Deviation of Absolute Error (SD; 95% data): 0.13951

Model Quality based on MAE-based criteria: 'GOOD'

Golbraikh and Tropsha acceptable model criteria's (7) :

***************************************************

1. Q^2 0.90415 Passed   (Threshold value Q^2>0.5)

2. r^2 0.92569 Passed   (Threshold value r^2>0.6)

3. |r0^2-r'0^2| 0.01501          Passed   (Threshold value |r0^2-r'0^2|<0.3)

4. k 0.99239    [(r^2-r0^2)/r^2]   0.00256 OR*

k' 1.0063    [(r^2-r'0^2)/r^2]   0.01878 Passed  (Threshold value: [0.85<k<1.15 and ((r^2-r0^2)/r^2)<0.1 ] OR* [0.85<k'<1.15 and ((r^2-r'0^2)/r^2)<0.1] )

 

 

 

 

Table 2: Correlation matrix for the best QSAR model Equation 5.1

	apol	ATSC8p	nssO	VE3_D
apol	1.000
ATSC8p	-0.1334	1.000
nssO	0.3206	0.2770	1.000
VE3_D	-0.5889	-01256	-0.2826	1.000

 

Table 3:  QSAR Descriptors of DPP-4 inhibitors

Name	pIC50	apol	ATSC8p	nssO	VE3_D
2	10.74	39.96769	-1.76855	0	-6.00711
3	9.61	39.96769	-1.85127	0	-5.05077
4	10.96	39.96769	-1.59695	0	-4.27629
5	10.77	43.17107	-3.04585	0	-4.27629
7	10.12	41.84427	-2.30391	0	-4.27629
8	10.7	42.11469	-2.3054	0	-4.63571
9	10.68	42.27069	-3.15359	0	-4.13924
10	10.51	42.84169	-3.04369	0	-5.7751
11	11.4	41.59069	-3.93977	0	-4.27629
12	11.4	42.46069	-5.3656	0	-4.27629
13	9.84	53.30465	-3.25712	0	-6.16482
15	10.57	42.27069	-1.68551	0	-7.95386
16	10.34	42.84169	-3.04436	0	-6.23449
18	9.68	42.84169	1.326809	0	-10.7791
20	11.52	39.85789	-1.707	0	-4.0548
22	10.82	43.86327	-1.48249	1	-3.94653
25	11.22	42.73189	-3.15395	0	-5.2645
26	9.9	52.34203	-1.55704	0	-5.2645
27	10.03	57.09045	-1.96379	1	-4.66624
28	10.66	42.16089	-1.7747	0	-6.50901
29	10.7	42.73189	-3.13558	0	-7.88199
30	11.22	39.7481	-1.7376	0	-5.61252
31	10.77	39.63831	-1.85004	0	-8.02489
32	10.64	39.63831	-1.85004	0	-8.50818
33	10.22	39.63831	-2.10229	0	-15.2955
34	9.58	50.65786	-2.02214	0	-6.25004
35	9.47	53.75145	-2.63328	0	-6.35282
36	9.43	56.84503	-3.83216	0	-6.13481
38	9.28	56.84503	-3.26119	0	-7.20549
39	9.24	59.93862	-3.90149	0	-9.30993
40	9.32	54.55345	-3.07147	0	-6.13481
41	9.61	55.51145	-3.10086	0	-7.23285
42	9.47	58.60503	-3.95397	0	-7.64249
44	9.06	64.7922	-4.55289	0	-9.8782
45	8.03	63.88503	-2.52144	0	-7.93614
48	8.08	61.93145	-2.6132	0	-7.52736
49	7.71	63.77524	-2.61461	0	-10.5495
51	8.42	63.77524	-2.49978	0	-8.0663
52	7.82	65.39824	-1.2517	0	-10.5495
53	7.98	65.39824	-2.00415	0	-9.11949
55	7.68	67.78062	-1.98304	1	-14.7656
56	8.12	67.78062	-2.97056	1	-9.87002
57	8.36	67.78062	-0.74157	1	-7.90727
59	8	66.91145	-2.38925	0	-10.0602
60	7.8	71.6762	-1.23532	2	-14.3555
1*	10.57	40.07748	-1.74242	0	-4.17759
6*	10.54	43.97307	-1.36641	1	-4.13924
14*	10.38	43.17107	-2.7733	0	-6.00711
17*	10.2	42.27069	-0.7191	0	-5.33592
19*	10.77	39.96769	-1.85127	0	-3.97136
21*	11.4	43.06127	-3.154	0	-4.0548
23*	10.54	42.00489	-2.41429	0	-4.36414
24*	11.3	42.16089	-3.25218	0	-3.94653
37*	9.48	59.93862	-3.36961	0	-5.76647
43*	9.48	61.69862	-4.76556	0	-9.0776
46*	7.59	66.97862	-1.56279	0	-6.88244
47*	7.75	62.55824	-2.79883	0	-7.93614
50*	8.12	63.77524	-2.55671	0	-9.11949
54*	8.22	65.39824	-2.87984	0	-8.0663
58*	8.07	63.66545	-2.53519	0	-10.0602

 

 

Statistical evaluation of  Equation 1 clearly demonstrated that model is having acceptable values of primary statistical parameters including SEE: 0. 0.34922, r² : 0.91955, r² adjusted : 0.91151, PRESS : 4.87812, F : 114.30494, Q² : 0.90415 which determine internal consistency of the best model, Equation 1,  and  r² : 0.92569, r₀² : 0.92332, reverse r₀² : 0.9083, RMSEP: 0.36579, Q²f1 or R2(Pred) : 0.92043, Q2f2 : 0.92001 Average rm^2(test) : 0.82937, Delta rm^2(test) : 0.06883 which determine external predictive ability of the best model. Other criterion including Model Quality based on MAE-based criteria and Golbraikh and Tropsha acceptable model criteria's[ also pass the model for its acceptability to use it for designing of new DPP-4 inhibitors and prediction their activities. Predicted activities of training and test set molecules from the best model, Equation 1, along with residual values are given in Table 4.   Graph of observed vs predicted activities from the best model of the training and test set molecules is shown in Fig. 3 and compound vs residual is shown in Fig. 4. These graphs clearly indicate that most of the compounds predicted within ± 0.5 pIC₅₀ units. 

 

 

 

 

Table 4: Predicted activities of training and test set molecules along with residual values

* Test compounds, ^a - logIC_50, where IC₅₀ is experimental reported in the literature, 

^b predicted –log(IC₅₀) from the best model Equation 5.1 

Figure 3: Graph of observed vs predicted activity from the best model Equation 1.

Figure 4: Resisual plot for training and test set

 

 

Some DPP-4 inhibitors were identified by QSAR model based virtual screening (VS) protocol. VS is a computational technique used in identification new bioactive molecules. It deals with the quick search of large libraries of chemical structures in order to identify those structures which are most likely to map over the query in silico model. For this purpose, the best QSAR model of DPP-4 inhibitors, given in Equation 1, was used to screen out some α-glucosidase inhibitors as NCE with anti-diabetic effect. These best models were used as filters for screening DRUGBANK using Predict Module of DTC QSAR tool^{13, 14}. To predict activities of the screened out molecules, descriptors of these were calculated by PaDEL software¹². Some identified DPP-4 inhibitors along with predicted pIC₅₀ from Equation1 is given in Table 5. Top ten repurposed DPP-4 inhibitors screened out by virtual screening using Equation 1 as query against DRUGBANK are shown in Fig.5.

 

 

Table 5: Newly identify DPP-4 inhibitors as anti-diabetic drug

 

 

Figure 5: Top ten repurposed DPP-4 inhibitors screened out by virtual screening using Equation 1 as query against DRUGBANK.  

 

 

 

 

 

4. CONCLUSION

On the basis of this QSAR modeling of DDP-4 inhibitory activity, it can concluded that a Hansch type two dimensional QSAR model has been successfully developed by utilizing some PaDEL descriptors for a set of (S)-1-((S)-2-amino-3-phenylpropanoyl) pyrrolidine-2-carbonitrile derivatives. Generated model was thoroughly evaluated by means of all reported statistical parameters. This validation results of the best model Equation 1 are in acceptable criterion and therefore suggest model’s reliability to be used in VS for identifying repurposed DPP-4 inhibitors which may be further develop as new effective anti-diabetic in management of post-prandial hyperglycemia in the type II diabetes without additional safety measurement.  

Acknowledgements: None 

Conflict of Interest: The authors declare no potential conflict of interest with respect to the contents, authorship, and/or publication of this article.

Author Contributions: All authors have equal contribution in the preparation of manuscript and compilation.

Source of Support: Nil

Funding: The authors declared that this study has received no financial support.

Informed Consent Statement: Not applicable. 

Data Availability Statement: The data presented in this study are available on request from the corresponding author. 

Ethical approvals: This study does not involve experiments on animals or human subjects.

REFERENCES

1. Pizzi RA. Defying diabetes: The discovery of insulin. Modern Drug Discovery 2000; 3(6); 77-80.

2. APh A Special Report. New approaches to insulin therapy for diabetes. American Pharmaceutical Association, Washington DC 2001.

3. Derosa G, Maffioli P. α-Glucosidase inhibitors and their use in clinical practice, Arch Med Sci 2012; 5:899-906 https://doi.org/10.5114/aoms.2012.31621 PMid:23185202 PMCid:PMC3506243

4. Wehmeier U, Piepersberg W. Biotechnology and molecular biology of the α - glucosidase inhibitor acarbose, Appl. Microbiol. Biot. 2004; 63:613-625. https://doi.org/10.1007/s00253-003-1477-2 PMid:14669056

5. Narita T, Yokoyama H, Yamashita R, Sato T, Hosoba M, Morii T., et al., Comparisons of the effects of 12-week administration of miglitol and voglibose on the responses of plasma incretins after a mixed meal in Japanese type 2 diabetic patients, Diabetes. Obes. Metab. 2011; 14:283-287. https://doi.org/10.1111/j.1463-1326.2011.01526.x PMid:22051162

6. Derosa G, Mereu R, D'Angelo A, Salvadeo S, Ferrari I, Fogari E, et al., Effect of pioglitazone and acarbose on endothelial inflammation biomarkers during oral glucose tolerance test in diabetic patients treated with sulphonylureas and metformin, J. Clin. Pharm. Ther. 2010; 35:565-579. https://doi.org/10.1111/j.1365-2710.2009.01132.x PMid:20831680

7. Derosa G, Maffioli P. Mini-Special Issue paper Management of diabetic patients with hypoglycemic agents α-Glucosidase inhibitors and their use in clinical practice, Arch. Med. Sci. 2012; 5:899-906. https://doi.org/10.5114/aoms.2012.31621 PMid:23185202 PMCid:PMC3506243

8. Holt R, Lambert K. The use of oral hypoglycaemic agents in pregnancy, Diabet. Med. 2014; 31:282-291. https://doi.org/10.1111/dme.12376 PMid:24528229

9. Syahrul I. et al. Synthesis of novel flavone hydrazones: In-vitro evaluation of α-glucosidase inhibition, QSAR analysis and docking studies Eur. J. Med. Chem., 2015; 105:156-170. https://doi.org/10.1016/j.ejmech.2015.10.017 PMid:26491979

10. Muhammad T. et al. Synthesis of novel inhibitors of α-glucosidase based on the benzothiazole skeleton containing benzohydrazide moiety and their molecular docking studies, Eur. J. Med. Chem, 2015; 92:387-400. https://doi.org/10.1016/j.ejmech.2015.01.009 PMid:25585009

11. Farman A, et al. Hydrazinyl arylthiazole based pyridine scaffolds: Synthesis, structural characterization, in vitro α-glucosidase inhibitory activity, and in silico studies, Eur. J. Med. Chem., 2017; 138:255-272 https://doi.org/10.1016/j.ejmech.2017.06.041 PMid:28672278

12. Flynn GL. Substituent constants for correlation analysis in chemistry and biology. By Corwin Hansch and Albert Leo. Wiley, 605 Third Ave., New York, NY 10016. 1979.

13. Golbraikh A, Tropsha A., Beware of Q2, J Mol Graph Model, 2002; 20:269-76. https://doi.org/10.1016/S1093-3263(01)00123-1 PMid:11858635

14. Krzywinski M, Altman N. Classification and regression trees. Nat Methods. 2017; 14(8):757. https://doi.org/10.1038/nmeth.4370

15. Costa VG, Pedreira CE. Recent advances in decision trees: an updated survey. Artif Intell Rev. 2023; 56:4765-4800. https://doi.org/10.1007/s10462-022-10275-5