3.1 Pharmacophore Model Generation
3.1.1 Cost analysis
21 compounds with different activities were included in the training set, and their IC50 values covered four orders of magnitude to generate 3D-QSAR pharmacophore model. The null cost of the generated pharmacophore was 284.43 and the fixed cost was 65.5657. Since the cost value of the ideal pharmacophore model should be as close as possible to the fixed cost, the larger the △cost, the better. Table 1 lists the cost values, △cost, RMSD and corrlation of the ten pharmacophore models generated. It is not difficult to see that among the 10 pharmacophores generated, hypo01 is the closest to fixed cost, △cost value is 195.736 (more than 60). Compared with other pharmacophore models, hypo01 has a certain predictive ability and low complexity. Also, from the analysis of other pharmacophore indexes, hypo01 has the largest correlation coefficient (0.950852) and the minimum RMSD (1.46439), which means that the pharmacophore is better than other models and can better predict the activity value of compounds.
Table 1. Results of HypoGen model generation with the training set data.
HYPO
|
Total Cost
|
△Cost 1
|
RMSD
|
Corrlation
|
Maximum Fit
|
Features 2
|
01
|
88.694
|
195.736
|
1.46439
|
0.950852
|
6.66393
|
HBA,HBA,HYA,HYD
|
02
|
90.0072
|
194.4228
|
1.49407
|
0.948795
|
6.28908
|
HBA,HBA,HYA,HYD
|
03
|
90.8522
|
193.5778
|
1.53574
|
0.9485
|
5.07409
|
HBA,HBA,HYA
|
04
|
91.346
|
193.084
|
1.55899
|
0.944088
|
5.34599
|
HBA,HBA,HYA
|
05
|
103.217
|
181.213
|
1.88939
|
0.916699
|
6.52523
|
HBA,HYA,HYD
|
06
|
110.053
|
174.377
|
2.04936
|
0.9012
|
5.2013
|
HBA,HYA,HYD
|
07
|
111.188
|
173.242
|
1.99687
|
0.906521
|
4.69033
|
HBA,HBA,HYA,HYD
|
08
|
112.159
|
172.271
|
2.09768
|
0.896214
|
5.19966
|
HBA,HYA,HYD
|
09
|
112.685
|
171.745
|
2.05
|
0.901195
|
5.04447
|
HBA,HBA,HBD,HYD
|
10
|
113.703
|
170.727
|
2.13622
|
0.892124
|
5.39697
|
HBA,HBA,HYA
|
1 Null cost = 284.43, Fixed cost= 65.5657, △Cost= Null cost-Total Cost.
2HBA-Hydrogen bond acceptor, HYA-Hydrophobic aromatic, HYD- Hydrophobic, HBD- Hydrogen bond donor.
The pharmacophore characteristics of hypo01 including four hydrogen bond receptor characteristics (HBA), one hydrophobic characteristic (HYD) and hydrophobic aromatic characteristic (HYA) (Figure 3). 3D-QSAR pharmacophores can be predicted by the degree of overlap between molecules and models. The activity prediction value (Estimate), matching value (Fit Valve) and grade classification of 21 training set compounds are listed in Table 2. The activity levels are divided into four orders of magnitude: the most active molecule (IC50 ≤ 0.1 μM, ++++), and the more active molecule (0.1 μM < IC50 ≤ 1 μM, +++), the moderate activity molecules (1μ m < IC50 ≤ 10 μM, ++), and lowest activity molecules (IC50 > 10 μM, +). The most active molecules can be well mapped on the pharmacophore model, while other molecules have some pharmacodynamic characteristics that are not mapped.
Table 2. Comparison between the experiment biological (IC50) and the estimated activity (IC50) of the Hypo01 model
NO
|
IC50 Value(μM)
|
Fit Valve
|
Error1
|
Activity scale 2
|
Experimental
|
Estimated
|
Experimental
|
Estimated
|
1
|
77
|
56
|
2.95
|
-1.4
|
+
|
+
|
2
|
90
|
34
|
3.16
|
-2.7
|
+
|
+
|
3
|
2
|
2.9
|
4.23
|
1.4
|
++
|
++
|
4
|
12
|
28
|
3.25
|
2.3
|
+
|
+
|
5
|
9.7
|
18
|
3.44
|
1.8
|
++
|
+
|
6
|
14
|
28
|
3.25
|
2
|
+
|
+
|
7
|
0.03
|
0.034
|
6.16
|
1.1
|
++++
|
++++
|
8
|
0.08
|
0.1
|
5.69
|
1.3
|
++++
|
++++
|
9
|
0.28
|
0.71
|
4.84
|
2.5
|
+++
|
+++
|
10
|
0.46
|
0.24
|
5.3
|
-1.9
|
+++
|
+++
|
11
|
0.77
|
0.73
|
4.83
|
-1.1
|
+++
|
+++
|
12
|
1.2
|
0.85
|
4.76
|
-1.4
|
++
|
+++
|
13
|
1.4
|
2.6
|
4.27
|
1.9
|
++
|
++
|
14
|
1.5
|
2.8
|
4.24
|
1.8
|
++
|
++
|
15
|
2.1
|
2.6
|
4.27
|
1.2
|
++
|
++
|
16
|
3
|
3.5
|
4.14
|
1.2
|
++
|
++
|
17
|
5.4
|
2.3
|
4.33
|
-2.3
|
++
|
++
|
18
|
5.8
|
2.8
|
4.24
|
-2
|
++
|
++
|
19
|
7.1
|
4.9
|
4
|
-1.5
|
++
|
++
|
20
|
7.3
|
2.7
|
4.27
|
-2.8
|
++
|
++
|
21
|
3.7
|
3.4
|
4.16
|
-1.1
|
++
|
++
|
|
|
|
|
|
|
|
|
|
1 The error coefficient is calculated as the ratio of measured activity to estimated activity; + indicates that the estimated IC50 is higher than the experimental IC50; - indicates that the estimated IC50 is lower than the experimental IC50.
2 The activity scale is the same as that defined in the 2.1. session.
3.1.2 Fischer randomization test
Fischer randomization test method can verify whether the pharmacophore model is statistically significant. By randomly corresponding the activity values to different molecular structures and comparing Hypo01 with 10 other randomly generated models, it is obvious that Hypo01 has a lower cost and a higher correlation coefficient (Figure 4), which proves that the model we constructed is not a result of chance.
3.1.3 Test Set Analysis
To further verify the predictive ability of the pharmacophore, the pharmacophore was used to predict the activity of the test set, which contained 11 compounds with different activities. The difference between the predicted activity value and the experimental activity value was compared. The regression analysis (Figure 5) demonstrated that the pharmacophore had a good predictive ability.
3.2 Database screening
The Zinc database containing 86 976 molecules was obtained by Lipinski's Rule of Five, and then SMART's filtration was used to reduce the number of molecules to 82 663.Using the validated hypo01 model as a 3D query, the processed database was screened, and 16181 molecules could be mapped on hypo01, of which 2763 molecules had predicted activity values less than 1 μM.
3.3 Molecular Docking
The FGFR1 protein complex (PDB ID: 4ZSA) contains the entire protein and its original ligand, which is immobilized at the active site of the FGFR1 protein, and key amino acid residues are determined by non-bonding action to define the active site. The ligand was extracted and then docked with the FGFR1 protein, with a high degree of molecular overlap and a RMSD value of 1.7. It can be concluded that the LibDock docking has a high reliability and can be used for the next virtual screening.
The well-known inhibitor of FGFR, E-3810, can effectively inhibit the activity of FGFR1, which is used as a reference compound. E-3810 inhibits the activity of FGFR1 by forming hydrogen bonds between methylamino group, phenoxy group and key residues such as Gly485, Asp641, Glu562, Glu531 on the phenyl ring(Figure 6). The docking score of E-3810 and FGFR1 is 145.145. A series of compounds with docking score greater than this value were screened using LibDock.
In LibDock, 659 compounds were successfully docked, and these compounds were docked on the receptor protein using the CDOCKER module. We selected the molecules with the highest LibDock score and -CDOCKER energy score for visual observation, and found 6 small molecules, and the docking results of 6 small molecules are shown in Table 3.
Table 3. The docking results of 7 potential FGFR1 inhibitors.
Name
|
2D Structure
|
LibDock Score
|
-CDOCKER
Energe
|
Estimated IC50
|
ZINC000038579824
|
|
158.868
|
44.3841
|
0.110979
|
ZINC000230274254
|
|
151.735
|
12.6032
|
0.143403
|
ZINC000011878462
|
|
146.958
|
30.956
|
0.181035
|
ZINC000219521723
|
|
148.581
|
31.1579
|
0.511363
|
ZINC000072436123
|
|
148.899
|
18.813
|
0.877224
|
ZINC000097657383
|
|
150.456
|
20.7042
|
0.88008
|
E-3810
|
|
145.145
|
3.35382
|
1.63942
|
All the selected molecules can be linked to the key residues of the receptor protein, thus inhibiting the activity of the protein. The hydrogen bond interaction between the six molecules and the protein is shown in Table 4.
Table 4. Hydrogen bond interaction parameters of each compound
Name
|
Donor Atom
|
Receptor Atom
|
H-Bond Distance (Å)
|
ZINC000038579824
|
Asp641:HN
Phf642:HN
Asn568:HD21
|
ZINC000038579824
O2
N25
|
3.04
2.39
2.20
|
ZINC000230274254
|
Phe489:O
Asp641:HN
Ala564:HN
Ala564:O
|
H43
O15
N11
H28
|
2.79
2.96
2.60
1.57
|
ZINC000011878462
|
Ala564:HN
Ala564:O
Ala564:O
|
O6
H31
H33
|
2.70
2.45
2.86
|
ZINC000219521723
|
Asp641:HN
Ala564:O
Tyr563:HA
|
O24
H28
O3
|
2.01
2.74
2.89
|
ZINC000072436123
|
Phe642:HN
Asp641:HA
Ala:564:O
|
N23
N24
H41
|
2.09
2.15
2.76
|
ZINC000097657383
|
Ala564:HN
Gly643:HN
Asp641:OD1
|
O10
N27
H46
|
2.53
2.41
2.92
|
The docking score was 158.868 for ZINC000038579824 (2-(1,3-dimethyl-2,6-dioxo-1,2,3,6-tetrahydro-7H-purin-7-yl)-N-(2-((6-methoxy-3,4-dihydroisoquinolin-2(1H)-yl)sulfonyl)ethyl)acetamide) has -44.3841 CDOCKER energy, and its predicted activity value is only 0.110979, which can be well mapped on the pharmacophore Figure 7A). The oxygen atom at the end and the nitrogen atom on the imidazole group form hydrogen bonds with Phf64 2 and Asn5688, respectively (Figure 8A). ZINC000230274254 ((R)-3-(2,5-dimethyl-7-oxo-4,7-dihydro-[1,2,4]triazolo[1,5-a]pyrimidin-6-yl)-N-(1-hydroxy-3-phenylpropan-2-yl)propanamide) can form four hydrogen bonds with Phe489, Ala564, Asp641 (Figure 8B), and the pharmacophore predicted activity is 0.143403 (Figure 7B). ZINC000011878462 ((R)-2-(3-oxo-1-phenethylpiperazin-2-yl)-N-(2-(2-oxooxazolidin-3-yl) ethyl) acetamide) had higher docking score and lower CDOCKER energy, formed three hydrogen bonds on Ala564 (Figure 8C), and could bind stably on the pharmacophore (Figure 7C). ZINC000219521723 ((R)-2-(5-phenyl-2H-tetrazol-2-yl)-N-(2-(5-(tetrahydrofuran-2-yl)-1,2,4-oxadiazol-3-yl)ethyl)acetamide) with docking score of 148.581 formed three hydrogen bonds with Asp641, Ala564, Tyr563, Oxadiazole and Phf489 formed a pi-pi conjugate system (Figure 8D), and tetrazolium group mapped on HBA (Figure 7D). ZINC000072436123 ((R)-2-(4-(2-(3-(1H-tetrazol-1-yl) phenoxy) acetyl) morpholin-3-yl)-N, N-dimethylacetamide) formed three hydrogen bond interactions with Ala564, Asp641, Phe642 (Figure 8E), and the tetrazolium group was mapped on HBA (Figure 7E). ZINC000097657383 ((R)-N-(3-(1H-imidazol-1-yl)propyl)-2,7-dioxo-N-(pyridin-3-ylmethyl)-1,3-diazepane-4-carboxamide) with docking score of 150.456 formed three hydrogen bonds with Ala564, Gly643, Asp641, and benzene ring formed a pi-pi conjugate system with Phf642 (Figure 8F), which could be mapped on the pharmacophore (Figure 7F).
3.4. ADMET and Toxicity Prediction
The ADMET module of DS2016 was used to predict the absorption, distribution, metabolism, excretion level of all selected small molecules and E-3810. The results are shown in table 5. The solubility level (in water at 25 ℃) indicates that all compounds are soluble in water. For human intestinal absorption, except ZINC00000385579824, it has good absorption level. CYP2 D6 is one of the important enzymes involved in drug metabolism. All compounds are non-inhibitors of cytochrome P4502D6 (CYP2D6). For hepatotoxicity, compared with E-3810 (toxic), ZINC0002219521723 was predicted to be toxic, and the other compounds were not hepatotoxic (Figure 9).
Table 5. ADME prediction
Name
|
Solubility
Level1
|
BBB
Level2
|
Absorption
Level3
|
CYP2D64
|
Hepatotoxity5
|
PPB
Level6
|
ZINC000038579824
|
3
|
4
|
1
|
0
|
0
|
0
|
|
ZINC000230274254
|
3
|
3
|
0
|
0
|
0
|
0
|
|
ZINC000011878462
|
4
|
4
|
0
|
0
|
0
|
0
|
|
ZINC000219521723
|
3
|
4
|
0
|
0
|
0
|
1
|
|
ZINC000072436123
|
4
|
3
|
0
|
0
|
0
|
0
|
|
ZINC000097657383
|
4
|
4
|
0
|
0
|
1
|
0
|
|
E-3810
|
2
|
3
|
0
|
0
|
1
|
1
|
|
1 footer. a. Solubility level: 0 (extremely low); 1 (very low, but possible); 2 (low); 3 (good). 2 BBB(Blood Brain Barrier level)Level: 0(Very High Penetrant); 1(High); 2(Middle); 3(Low); 4(undefined). 3 Absorption (Human Intestinal Absorption) Level: 0(good); 1(moderate); 2(Poor); 3(Very Poor).4 CYP2D6(Cytochrome P4502D6) level:0(non-inhibitor); 1(inhibitor).5 Hepatotoxicity: 0 (Nontoxic); 1 (Toxic). 6 PPB (Plasma Protein Binding): 0 (Binding is <90%); 1 (Binding is>90%); 2 (Binding is >95%).
To check the safety of the compound, the toxicity of all selected small molecules and E-3810 was predicted using the TOPKAT module, as shown in Table 6. The results showed that all compounds except ZINC0000038579824 were predicted to be non-mutagens. All of the three compounds have good aerobic biodegradation performance, four compounds have no potential developmental toxicity, and one is a non-carcinogen.
Table 6. Toxicity prediction
Name
|
Ames1
|
Mouse NTP2
|
DTP3
|
Aerobic Biodegradability4
|
Male
|
Female
|
ZINC000038579824
|
1
|
1
|
1
|
1
|
0
|
ZINC000230274254
|
0
|
0
|
0
|
0
|
1
|
ZINC000011878462
|
0
|
1
|
0
|
0
|
1
|
ZINC000219521723
|
0
|
1
|
0
|
0
|
0
|
ZINC000072436123
|
0
|
0
|
1
|
1
|
0
|
ZINC000097657383
|
0
|
1
|
1
|
0
|
1
|
E-3810
|
0
|
1
|
1
|
1
|
1
|
1 Mutagenicity (Ames text): 0 (non-toxic), 1 (toxic); 2 Rodent Carcinogenicity: 0 (non-carcinogenicity), 1 (carcinogenicity); 3 Developmental Toxicity Potential: 0 (non-toxic), 1 (toxic); 4 Aerobic Biodegradability: 0 (non-aerobic biodegradability), 1 (aerobic biodegradability).
Therefore, ZINC000230274254 can not inhibit CYP2D6, has no hepatotoxicity, mutagenicity and potential developmental toxicity, and has good aerobic biodegradation performance and absorption level. It is predicted to be a safe candidate drug and selected for further research.