Phytomedicine 38 (2018) 145–157

Contents lists available at ScienceDirect

Phytomedicine
journal homepage: www.elsevier.com/locate/phymed

Original Article

Predicting tyrosinase inhibition by 3D QSAR pharmacophore models and T

designing potential tyrosinase inhibitors from Traditional Chinese medicine
database
Hongwei Gao
Key Laboratory of Plant Resources and Chemistry in Arid Regions, Xinjiang Technical Institute of Physics and Chemistry, Chinese Academy of Sciences, Urumqi, 830011,
China

A R T I C L E I N F O A B S T R A C T

Keywords: Background: Tyrosinase plays a key role in the formation of skin melanin. The excessive accumulation of skin
Tyrosinase melanin will cause the serious aesthetic problems for human beings.
Inhibitors Hypothesis/purpose: To find the potent tyrosinase inhibitors using computational simulation from TCM
TCM Database@Taiwan.
Pharmacophore models
Study design: Inhibitors of tyrosinase have been thought as potential drugs for the decrease of melanin synthesis
Docking
in the process of pigmentation. To develop new tyrosinase inhibitors, we performed a virtual screening from
Traditional Chinese medicine (TCM) and Druglike Databases using the best 3D QSAR pharmacophore model as a
3D search query.
Methods: A total of 109 compounds were obtained after filtering by Lipinski's rule of five. Finally, 148 com-
pounds (22 from training set, 17 from test set, 109 from TCM and Druglike databases) were selected for further
docking studies. De Novo Evolution designed the top 10 candidates from the docking results.
Results: Hypo1 was selected as the best quantitative pharmacophore model, because Hypo1 has characters of the
highest cost difference (353.773), the lowest RMS (1.985), the lowest Error (121.440), and the best correlation
coefficient (0.933). By the analysis of interaction amino acids in the top 10 hits including two controls, HIS42,
HIS60, HIS204, HIS208, ARG209 and VAL218 are identified as the key binding site residues, ARG209 and
VAL218 are the critical residues for the inhibitory activity of tyrosinase. This finding is consistent with the
results from literatures.
Conclusion: De Novo Evolution study suggested Tyrosinase_1*_Evo_4, Tyrosinase_23*_Evo_7, magnolone.cdx_15_Evo_4,
compound_2.cdx_2_Evo_2, Compound_B_Evo_5, Compound_C_Evo_9, Compound_D_Evo_6 and malabaricone_
C.cdx_3_Evo_10 as the potential tyrosinase inhibitor candidates. De Novo Evolution study also suggested
compound_2.cdx_2_Evo_2 as the most potential tyrosinase inhibitor candidate. A total of ten novel leading compounds
were identified to have the favorable interaction with tyrosinase by the docking analyses.

Introduction Human tyrosinase is a membrane-bound glycoprotein and has 13%

Tyrosinase (polyphenol oxidase, EC 1.14.18.1) (Chase et al., 2000;
Sánchez-Ferrer et al., 1995) is a copper-containing enzyme which is
widely distributed in plants, animals and human beings. It plays a key
role in the formation of skin melanin (Solano et al., 2006). The ex-
cessive accumulation of skin melanin will cause the formation of mel-
anoma, which gives rise to the serious aesthetic problems for human
beings (Chompoo et al., 2012). The formation of melanoma may be
relevant to tyrosinase (Pan et al., 2011), which is also responsible for
Parkinson's and other neurodegenerative diseases (Zhu et al., 2011).
carbohydrate content. The active site of tyrosinase contains two copper
atoms, which can interact with dioxygen to form a highly reactive che-
mical intermediate and then oxidizes the substrate. Inhibiting the biolo-
gical activities of tyrosinase is an effective method to downregulate the
formation of skin melanin, and to prevent Parkinson's and other neuro-
degenerative diseases. Kojic acid as the commonly used tyrosinase in-
hibitor has been forbidden to use in many countries due to its serious side
effects and unstable property (Zhou et al., 2013). With the increasing
concern on the health issues, it is beneficial to find the effective, stable
and novel tyrosinase inhibitors with the lower side effects.

Abbreviations: TCM, Traditional Chinese medicine; HBA, hydrogen bond acceptor; HBD, hydrogen bond donor; GF, goodness of fit; EF, enrichment factor; HYAr, hydrophobic aromatic;
SAR, structure-activity relationship; QSAR, Quantitative Structure-Activity Relationship; DS, Dock score; LigS1, LigScore1; LigS2, LigScore2; DS2016, Discovery Studio 2016
E-mail address: gaohongw369@ms.xjb.ac.cn.

https://doi.org/10.1016/j.phymed.2017.11.012
Received 4 November 2016; Received in revised form 6 October 2017; Accepted 29 November 2017
0944-7113/ © 2017 Elsevier GmbH. All rights reserved.
H. Gao Phytomedicine 38 (2018) 145–157

Table 1
10 pharmacophore models generated by the HypoGen for tyrosinase inhibitors.

Hypo No. Total cost Cost differencea Error RMS Correlation Features

1 135.471 353.773 121.440 1.985 0.933 HBA, HBA, HBD, HYAr

2 159.989 329.255 145.002 2.371 0.902 HBA, HBA, HBD, HY
3 161.436 327.808 146.275 2.391 0.900 HBA, HBD, HBD, HYAr
4 165.975 323.269 151.555 2.468 0.894 HBA, HBA, HBA, HYAr
5 166.137 323.107 151.804 2.472 0.893 HBA, HBA, HBA, HYAr
6 167.209 322.035 151.852 2.472 0.893 HBA, HBA, HBD, HY
7 169.012 320.232 154.59 2.512 0.889 HBA, HBD, HBD
8 171.626 317.618 156.028 2.532 0.888 HBA, HBD, HBD, HYAr
9 172.812 316.432 157.382 2.551 0.886 HBA, HBA, HBD, HYAr
10 173.548 315.696 158.011 2.560 0.885 HBA, HBD, HBD, HY

a
(Null cost-total cost), null cost = 489.244, fixed cost = 80.301, for the hypo1 weight = 1.126, configuration = 12.905. HBA, HBD, HY and HYAr are represented as hydrogen bond
acceptor, hydrogen bond donor, hydrophobic and hydrophobic aromatic, respectively.

Table 2
Experimental and predicted activity of the training set compounds based on pharmacophore model Hypro1.

Compound no. Fit valueb Exp.IC50 nM Predicted IC50 nM Errora Experimental scalec Predicted scalec

tyrosinase-1 8.06 0.029 0.07 2.4 +++ +++

tyrosinase-2 6.86 1.7 1.10 −1.5 ++ ++
tyrosinase-3 6.21 2.7 4.90 1.8 ++ ++
tyrosinase-4 5.54 6.1 23 3.8 ++ ++
tyrosinase-5 5.78 8.2 13 1.6 ++ ++
tyrosinase-6 5.23 20 48 2.3 ++ ++
tyrosinase-7 5.70 24 16 −1.5 ++ ++
tyrosinase-8 5.56 26 22 −1.2 ++ ++
tyrosinase-9 5.19 28 52 1.9 ++ ++
tyrosinase-10 5.63 34 19 −1.8 ++ ++
tyrosinase-11 5.70 49 16 −3.1 ++ ++
tyrosinase-12 5.42 58 30 −1.9 ++ ++
tyrosinase-13 5.29 60 41 −1.5 ++ ++
tyrosinase-14 5.80 63 13 −4.9 ++ ++
tyrosinase-15 4.39 91 330 3.6 ++ +
tyrosinase-16 4.71 110 160 1.4 + +
tyrosinase-17 4.61 200 200 −1 + +
tyrosinase-18 4.69 200 170 −1.2 + +
tyrosinase-19 4.77 200 140 −1.5 + +
tyrosinase-20 4.45 250 280 1.1 + +
tyrosinase-21 4.46 250 280 1.1 + +
tyrosinase-22 4.75 300 140 −2.1 + +

a
Difference 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.
b
Fit value indicates how well the features in Hypro1 overlap the chemical features in the training set compounds.
c
Activity scale: IC50 <1 µM = +++ (highly active); 1 µM ≤ IC50 < 100 µM = ++ (moderately active); IC50 ≥ 100 µM = + (low active).

Table 3
Experimental and predicted activity of the test set compounds based on pharmacophore model Hypro1.

Compound no. Fit valueb Exp.IC50 nM Predicted IC50 nM Errora Experimental scalec Predicted scalec

tyrosinase-23 5.86 0.11 11.20 101.81 +++ ++

tyrosinase-24 6.03 1.96 7.51 3.83 ++ ++
tyrosinase-25 6.28 16.17 4.22 −3.83 ++ ++
tyrosinase-26 6.04 16.9 7.37 −2.29 ++ ++
tyrosinase-27 5.78 20.1 13.31 −1.51 ++ ++
tyrosinase-28 5.15 24.52 57.131 2.33 ++ ++
tyrosinase-29 4.34 27.1 368.99 13.62 ++ +
tyrosinase-30 4.51 32.81 250.26 7.63 ++ +
tyrosinase-31 5.70 36.36 15.98 −2.28 ++ ++
tyrosinase-32 4.40 54.40 316.86 5.82 ++ +
tyrosinase-33 4.45 79.05 282.04 3.57 ++ +
tyrosinase-34 4.25 83.17 455.85 5.48 ++ +
tyrosinase-35 4.22 200.00 480.39 2.40 + +
tyrosinase-36 4.58 200.00 209.32 1.05 + +
tyrosinase-37 4.47 250.00 271.07 1.08 + +
tyrosinase-38 4.47 250.00 270.05 1.08 + +
tyrosinase-39 4.63 300.00 186.49 −1.61 + +

a
Predicted IC50 is higher than the experimental IC50; '−' indicates that the predicted IC50 is lower than the experimental IC50.
b
Fit value indicates how well the features in Hypro1 overlap the chemical features in the test set compounds.
c
Activity scale: IC50 < 1 µM = +++ (highly active); 1 µM ≤ IC50 < 100 µM = ++ (moderately active); IC50 ≥ 100 µM = + (low active).

146
H. Gao Phytomedicine 38 (2018) 145–157

Table 4
Molecular docking results of tyrosinase inhibitors (top 10).* Control compound.

Rank Name DS LigS1 LigS2 -PLP1 -PLP2 -PMF Consensus

1 Tyrosinase_1* 44.043 3 4.53 62.46 67.83 99.65 1

2 Tyrosinase_23* 38.297 2.72 4.38 60.57 64.94 93.66 0
3 magnolone.cdx_15 58.411 4.53 5.44 76.51 67.76 110.69 0
4 Compound_A 82.907 2.68 3.54 63.53 62.66 86.27 0
5 compound_2.cdx_2 87.1 3.32 4.85 57.69 49.4 102.01 1
6 Compound_B 66.78 3.29 4.01 51.84 61.55 64.31 0
7 Compound_C 73.274 3.61 4.69 61.42 62.9 54.06 3
8 Compound_D 71.429 1.7 2.18 37.11 38.71 59.01 0
9 7874.cdx_31 64.939 1.96 3.67 44.75 46.36 103.55 0
10 malabaricone_C.cdx_3 64.193 1.97 3.56 55.26 48.57 91.31 0

Abbreviations:
(1) Compound_A: 1_5-dihydroxy-1_7-bis(4-hydr-oxyl-3-methoxyphenyl)-4_6-hepta-diene-3-one.cdx-12.
(2) Compound_B: 1_5-Dihydroxy-1-(4-hydroxy-3-methoxyphenyl)-7-(4-hydroxyphenyl)-4_6-heptadien-3-.
(3) Compound_C: 5_7-dihydroxy-3-(3-hydroxy-4-methoxybenzyl)-6-methoxychroman-4-one.cdx-9.
(4) Compound_D: 5_7-dihydroxy-3-(3-hydroxy-4-methoxybenzyl)-6-methoxychroman-4-one.cdx-7.

Table 5
The interaction amino acid in the ligand-protein for the top 10 docking compounds including two controls (*).

Rank Compound Interaction amino acids

1 tyrosinase-1* HIS42, HIS204, HIS208, ALA221, ASN205, ARG209, VAL218

2 tyrosinase-23* HIS42, HIS208, ALA221, VAL218, ARG209
3 magnolone.cdx-15 HIS42, HIS60, HIS204, HIS208, HIS231, MET61, VAL218, GLY216
4 1_5-dihydroxy-1_7-bis(4-hydr-oxyl-3-methoxyphenyl)-4_6-hepta-diene-3-one.cdx-12 HIS42, HIS204, HIS208, ARG209, GLY216, VAL218, ALA221
5 compound_2.cdx-2 HIS60, HIS204, HIS208, ARG209, VAL217, VAL218, ALA221
6 1_5-Dihydroxy-1-(4-hydroxy-3-methoxyphenyl)-7-(4-hydroxyphenyl)-4_6-heptadien-3- HIS60, HIS204, HIS208, ASN205, ARG209, VAL217, VAL218, GLY216
7 5_7-dihydroxy-3-(3-hydroxy-4-methoxybenzyl)-6-methoxychroman-4-one.cdx-9 HIS60, HIS204, MET61, ARG209, VAL217, VAL218
8 5_7-dihydroxy-3-(3-hydroxy-4-methoxybenzyl)-6-methoxychroman-4-one.cdx-7 VAL218, ASN205, ARG209, MET61
9 7874.cdx-31 HIS60, HIS69, HIS204, HIS208, VAL217, VAL218
10 malabaricone_C.cdx-3 HIS60, HIS204, HIS208, ASN205, GLY216, ALA221, VAL218, PRO219

*Represents the control compound.

Table 6
Molecular docking results for the derivatives generated by De Novo Evolution protocol from top 10 tyrosinase inhibitors.

Rank Name DS LigS1 LigS2 -PLP1 -PLP2 -PMF Consensus LUDI3

1 Tyrosinase_1*_Evo_4 54.530 3.61 5.01 69.16 69.08 111.75 5 669

2 Tyrosinase_23* _Evo_7 40.826 2.70 3.16 72.63 76.56 142.50 0 804
3 magnolone.cdx_15_Evo_4 74.086 5.42 6.6 110.72 99.67 148.08 5 1097
4 Compound_A_Evo_1 29.550 1.73 1.03 39.09 49.96 75.41 0 1026
5 compound_2.cdx_2_Evo_2 110.926 3.87 4.5 80.79 74.29 102.72 5 1049
6 Compound_B _Evo_5 85.445 5.11 5.16 60.42 71.02 61.73 0 863
7 Compound_C _ Evo_9 56.934 2.78 3.92 63.83 63.09 61.19 5 945
8 Compound_D_ Evo_6 87.488 3.80 3.87 79.25 77.04 87.63 0 703
9 7874.cdx_31_Evo_3 58.791 4.00 4.30 82.65 79.56 60.00 0 727
10 malabaricone_C.cdx_3_Evo_10 78.776 2.27 4.13 88.92 81.79 100.21 0 805

*Represents the control compound.

Traditional Chinese medicine (TCM) emerged numbers of data
supporting its therapeutic potential. Therefore, TCM plays an important
role in diagnosis and treatment of diseases in Asia. TCM contains three-
dimensional structural information of compounds, which are ready for
molecular docking simulation. TCM Database@Taiwan includes 51,564
small molecules, which can be free downloaded from http://tcm.cmu.
edu.tw. TCM database is currently the world largest traditional Chinese
medicine database for drug screening (Chen, 2011). It is possible to find
the potent tyrosinase inhibitors from TCM database.
Computational techniques provide a useful and powerful tool to find
the potent tyrosinase inhibitors from TCM database. Moreover, the ef-
fective computational approach plays an important role in designing
the inhibitors and drug development. Three dimensional quantitative
structure activity relationship (3D QSAR) pharmacophore modeling is a
three-dimensional computational approach, which is widely carried out
in drug design for the potential tyrosinase inhibitors. 3D QSAR models
can been generated by the 3D QSAR Pharmacophore Generation
protocol using the Catalyst HypoGen algorithm from a set of ligands
with known activity values on a given biological target. A ligand maps
the more features of a pharmacophore, the predicted activity of ligand
is more accurate.
The goal of this study is to generate 3D pharmacophore models
based on the known tyrosinase inhibitors, which can be evaluated by
preparing test set, Fischer's randomization method, goodness of fit (GF)
and enrichment factor (EF) etc. Furthermore, 3D pharmacophore
models will be used to screen the derivatives obtained from de novo
evolution. We aimed to find the potent tyrosinase inhibitors using
computational simulation from TCM Database@Taiwan.
Results and discussion
3D. QSAR pharmacophore hypotheses for tyrosinase inhibitors
The following chemical features were selected for constructing

147
H. Gao Phytomedicine 38 (2018) 145–157

Fig. 1. Chemical structures of tyrosinase inhibitors in the training set together with their biological activity data (IC50 value, µM).

pharmacophore models: hydrogen bond acceptor (HBA), hydrogen
bond donor (HBD), hydrophobic (HY) and hydrophobic aromatic
(HYAr). Ten 3D QSAR pharmacophore hypotheses were generated by
HypoGen protocol (shown in Table 1). The null cost and fixed cost were
489.244 and 80.301, respectively. Hypo1 has characters of the highest
cost difference (353.773), the lowest RMS (1.985), the lowest Error
(121.440), and the best correlation coefficient (0.933). Furthermore,
the value of 12.905 for the configuration cost of Hypo1 was smaller
than the standard value of 17, which providing the additional support
for the reliability of Hypo1. Therefore, Hypo1 ranked the top and has

148
H. Gao Phytomedicine 38 (2018) 145–157

Fig. 2. (Continued) Chemical structures of tyrosinase inhibitors in the training set together with their biological activity data (IC50 value, µM).

149
H. Gao Phytomedicine 38 (2018) 145–157

Fig. 3. Chemical structures of tyrosinase inhibitors in the test set together with their biological activity data (IC50 value, µM).

the best predictive power, which was considered as the best structure-
activity relationship (SAR) model.
The top scoring Hypo1 is mapped to the most active compound in
the training set (tyrosinase-1) (shown in Fig. 5). Hypo1 consists of two
hydrogen bond acceptor (HBA_1 and HBA_2), one hydrogen bond donor
(HBD) and one hydrophobic aromatic region (HYAr).
The effectiveness of Hypo1 was evaluated by the predicted activities
(IC50) of the training and test set compounds, Fischer validation and
Heat map for the test set compounds.

150
H. Gao Phytomedicine 38 (2018) 145–157

Fig. 4. (Continued) Chemical structures of tyrosinase inhibitors in the test set together with their biological activity data (IC50 value, µM).

Fig. 5. The top scoring Hypo1 is mapped to the most active compound in the training set (tyrosinase-1) (HBA, hydrogen bond acceptor; HBD, hydrogen bond donor; HY, hydrophobic).

Validation of the pharmacophore models plotted in Fig. 7.

In Tables 2 and 3, training and test set compounds were relatively
Training and test set classified into three sets according to their IC50:
The activity of the training and test set compounds was well pre- IC50 < 1 µM = + + +, highly active; 1 µM ≤ IC50 < 150 µM = + +,
dicted by Hypo1. Experimental and predicted activities (IC50) of the moderately active; IC50 > 150 µM = +, low active or inactive. No-
training and test set compounds based on the pharmacophore model tably, the errors for computational quality were within 1 order of
Hypo1 were listed in Tables 2 and 3, respectively. Experimental (lo- magnitude for almost all compounds. Except for a few compounds, the
gActiv) and predicted activity (logEstimate) of these compounds were activity of all the remaining compounds were correctly predicted by

151
H. Gao Phytomedicine 38 (2018) 145–157

Hypo1. The correlation coefficient (R2) for the training and test set
compounds given by Hypo1 were 0.913 and 0.745, respectively (Fig. 7).
In general, for the compounds of tyrosinase-1–14, 16–22, 24–28, 31
and 35–39, the correct activity prediction were obtained, respectively.
For the compounds of tyrosinase-15, 23, 29, 30 and 32–34, the activity
predictions were not correct.

Fig. 6. Fischer validation: the total cost of the initial and the 18 random spreadsheets on
the 95% confidence level.
Fischer validation
Fischer validation was performed for ten 3D QSAR pharmacophore
models on 95% confidence level using CatScramble module within
Catalyst. A total of 19 random spreadsheets were created to evaluate
the statistical relevance of ten 3D QSAR pharmacophore models. The
result is shown in Fig. 6. From Fig. 6, we can see that the total cost of all
pharmacophore models of random 1–19 are much higher than that of
ten 3D QSAR pharmacophore models (the original pharmacophore
models in Fig. 6: Costs). The result indicates that ten 3D QSAR phar-
macophore models were not generated by chance, and provides addi-
tional supports and reliability for the pharmacophore models.
The confidence level of 95% were calculated using 19 random
spreadsheets (random hypotheses) by the equations as follows:

S = [1 − (1 + X)/Y] × 100 (1)

S = [1 − (1 + 0)/(19 + 1)] × 100 = 95% (2)

Where, X represents the total number of hypotheses having a total

cost lower than the initial pharmacophore hypotheses; Y represents the
total number of hypotheses (initial + random runs). Here, X = 0 and
Y = (19 + 1).

Heat map of ligand profiler

Heat map of ligand profiler represents two-dimensional table (X-axis
represents ten 3D QSAR pharmacophore models, Y-axis represents test
set compounds). This is a popular plotting technique in DS2016 soft-
ware, used to depict the validation of Hypo1 for predicting the activity
of test set compounds. Heat map of "Ligand Profiler" predicted by
Hypo1 for the test set compounds was shown in Fig. 8. Fig. 8 clearly
showed that the maximum FitValue (white block) is located in the area
of Hypo1, and the bigger FitValue (red block) is also in the area of
Fig. 7. Correlation between the experimental and predicted activity (pIC50) by Hypo1 for
the training and test set compounds.
Hypo1. This result indicates that Hypo1 is the best 3D QSAR pharma-
cophore model for predicting the activity of test set compounds.

Fig. 8. Heat map of “Ligand Profiler” predicted by Hypo1 for the test set compounds. (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)
152
H. Gao Phytomedicine 38 (2018) 145–157

Fig. 9. Structures of the top 10 docking compounds for tyrosinase inhibitors.

Database screening (Virtual screening) 2087 compounds from Druglike) were retrieved from the first
A compound was considered to be a hit only if all features of Hypo1 screening. Druglike database was implanted in the Discovery Studio
were mapped. A total of 8529 compounds (6442 compounds from TCM, 2016 (DS2016) package, and it contains 5384 compounds. Due to the

153
H. Gao
drug-likeness property is an important indicator for virtual screening of
small molecules from TCM and Druglike databases, we further filtered
these 8529 compounds using the Lipinski's rule of five and a total of
7010 compounds passed this filtration. Finally 109 compounds were
selected by restricting the estimated activity smaller than 1 µM. At last
148 compounds (22 from training set, 17 from test set, 109 from TCM
and Druglike databases) were further considered for docking studies.
Molecular docking studies of tyrosinase inhibitors
The docking result was shown in Table 4 which lists Dock score
(DS), LigScore1 (LigS1), LigScore2 (LigS2), -PLP1, -PLP2, -PMF and
Consensus score. DS is used to examine the ligand internal energy and
the ligand-receptor interaction energy, which is the sum of van der
Waal force and electrostatic energy. LigS1 and LigS2 are used to eval-
uate the polar interactions between ligand and receptor. PLP (Piecewise
Linear Potential) (Gehlhaar et al., 1995) is used to evaluate the strength
of hydrogen bond in the ligand-receptor. PMF(Potential Mean Force)
(Muegge and Martin, 1999) is used to examine the binding free energy.
Consensus score is a fast way to identify ligands that score high in more
than one scoring function, which is generated by the Consensus score
protocol.
Structures of the top 10 docking compounds for tyrosinase in-
hibitors are presented in Fig. 9. Compond_2.cdx-2 has the highest
DockScore (79.548) among 26 candidates. However, the corresponding
LigS1 (2.8), LigS2 (4.23), -PLP1 (48.65), -PLP2 (42.86) and -PMF
(93.53) for compound_2.cdx-2 are significantly lower than two controls.
This result indicates that compound_2.cdx-2 may not has the most op-
timal interaction with tyrosinase. Consensus score can evaluate all
scoring functions, and represents the number of different scoring
functions that highly rank the ligand. According to which, magnolo-
ne.cdx-15 is selected as a most potent candidate.
The docking pose and interactions between magnolone.cdx-15 and
tyrosinase are shown in Fig. 10, in which showed eight key residues
(HIS42, HIS60, HIS204, HIS208, HIS231, GLY216, VAL218 AND
MET61) around magnolone.cdx-15. As well as presents the non-bonded
interactions between magnolone.cdx-15 and tyrosinase: hydrogen
bonds, electrostatic and hydrophobic.
The interaction amino acids in the ligand-protein for the top 10
docking compounds are listed in Table 5. The analysis of interaction
amino acids in the top 10 hits including two controls showed that
154
Phytomedicine 38 (2018) 145–157
Fig. 10. The non-bonded interactions between magnolo-
ne.cdx-15 and tyrosinase.
HIS42, HIS60, HIS204, HIS208, ARG209 and VAL218 are the key
binding site residues. Based on the results, ARG209 and VAL218 are the
critical residues for the inhibitory activity of tyrosinase.
Ligand de novo evolution analysis
Structures of the derivatives selected after De Novo Evolution and
LigandFit Docking are presented in Fig. 11. The docking results for the
derivatives generated by De Novo Evolution protocol were summarized
in Table 6. Comparing DockScore (DS) values with the previous docking
result in Table 4, almost DS values for top 10 tyrosinase inhibitors in-
creased, except for Compund_A_Evo_1 and 7874.cdx_31_Evo_3. Com-
paring with the previous docking result in Table 4, consensus score
values for Tyrosinase_1*_Evo_4, compound_2.cdx_2_Evo_2, magnolo-
ne.cdx_15_Evo_4 and Compound_C _ Evo_9 increase from 1 to 5, 1 to 5, 0
to 5 and 3 to 5, respectively.
The derivative, compound_2.cdx_2_Evo_2, had the highest
Dockscore and Consensus Score out of the top 10 candidates. This
drastic increase in Dockscore and Consensus Score might be due to the
addition of a naphthalene functional group to the C20 and C21 posi-
tions of the benzene ring in the molecule of compound_2.cdx_2. As
shown in Fig. 12, the addition of a naphthalene functional group not
only changes the binding conformation of compound_2.cdx_2, but also
enhances interactions with the protein of tyrosinase. Hence, addition of
a naphthalene functional group might have an important implication in
designing tyrosinase inhibitors.
Materials and methods
Data set
All molecular modeling calculations were performed by Dassault
Systèmes BIOVIA (2015). For the ligand-based study, 22 compounds of
tyrosinase inhibitors (listed in Figs. 1 and 2) obtained from literatures
(Bandgar et al., 2012; Dat et al., 2015; Liu et al., 2015; Lee et al., 2009;
Liu et al., 2016; Dong et al., 2016; Zheng et al., 2016) were used to
construct a training set for generating 3D QSAR pharmacophores, 17
compounds of tyrosinase inhibitors (listed in Figs. 3 and 4) obtained
from the same literatures with the training set were used to construct a
test set for validate 3D QSAR pharmacophores. For the structure-based
studies, the 3D structure of tyrosinase was obtained from Protein Data
H. Gao Phytomedicine 38 (2018) 145–157

Fig. 11. Structures of the derivatives selected after De Novo Evolution and LigandFit Docking. The functional groups added in De Novo Evolution are circled in blue. (For interpretation of
the references to color in this figure legend, the reader is referred to the web version of this article.)

155
H. Gao Phytomedicine 38 (2018) 145–157

Bank (PDB ID: 3NM8). The tyrosinase inhibitors of tyrosinase-1

(0.029 µM) and tyrosinase-23 (0.11 µM) were taken as the control
(Tang and Chen, 2015).

Pharmacophore modeling

Ten 3D Quantitative Structure-Activity Relationship (QSAR) phar-

macophore models were generated by the Catalyst HypoGen algorithm
[20] implemented in the Discovery Studio 2016 (DS2016) package
[21]. The 3D QSAR pharmacophore model has the predictive ability for
the activity of ligand. Training set (0.029–300 µM) and test set
(0.11–300 µM) span 4 orders of magnitude, which are very good for
HypoGen module.

Pharmacophore model evaluation

Ten 3D QSAR pharmacophore models were evaluated for their

ability by training and test set, Fischer validation (Lu et al., 2009), heat
map of ligand profiler and Debnath (2002) in terms of cost functions. A
good pharmacophore model should have a high correlation coefficient
for training and test sets, the highest cost difference (Δcost = Null
cost − Total cost) and the match of important pharmacophore features.
Fig. 12. The docking pose of compound_2.cdx_2 and derivative compound_2.cdx_2_Evo_2.

Fig. 13. Flow chart of the virtual screening approach performed in

this study.

156
H. Gao
Virtual screening
TCM and Druglike databases screening were performed to map into
Hypo1 for comparing the notable features. TCM and Druglike databases
consist of 51,564 and 5384 compounds, respectively. The maximum
hits were set to be 300; minimum interfeature distance was 0.5; and
search method was fast. All queries were performed using the search 3D
database within the Catalyst DBServer module.
Docking protocol
In order to further refine the retrieved hits, a LigandFit
(Venkatachalam et al., 2003) protocol was carried out with tyrosinase
for 148 compounds (22 from training set, 17 from test set, 109 from
TCM and Druglike databases). Scoring functions (LigScore1, LigScore2,
PLP1, PLP2, Jain and PMF) were used to evaluate the interactions be-
tween the ligand and protein. Maximum poses retained as 10.0. RMS
threshold for diversity was set to be 1.5, and score threshold for di-
versity was 20.0.
The 3D structure of tyrosinase was obtained from the Protein Data
Bank (PDB ID: 3NM8). A binding site was set around the six key re-
sidues (HIS42, HIS60, HIS204, HIS208, HIS230, HIS231). Tyrosinase
catalyzes two important reactions around these six key residues with
the assistant of Cu atoms in human body: hydroxylation and oxidation
(Tang and Chen, 2015). This information was used as reference for
evaluating the docking results. The high activity compounds of tyr-
osinase-1 (0.029 µM) and tyrosinase-23 (0.11 µM) were chosen as the
control.
De Novo Evolution protocol
The De Novo Evolution calculation by Ludi program (Deng et al.,
2008) was performed on the top 10 candidates, which had the higher
Consensus score than the controls. De Novo Evolution protocol can
create a collection of molecules with higher scores by fusing the fit
fragments to the initial ligands. Full Evolution was used in this protocol.
Evolution Scoring Function was set to be Ludi Energy Estimate 3.
Maximum RMSD for fitting was 1.0. The derivatives generated by De
Novo Evolution protocol were docked back to tyrosinase again. Flow
chart of the virtual screening approach performed in this study was
presented in Fig. 13.
Biological activity test
Tyrosinase inhibitory effects of the synthesized 8 candidates
(Tyrosinase_1*_Evo_4, Tyrosinase_23*_Evo_7, magnolone.cdx_15_Evo_4,
compound_2.cdx_2_Evo_2, Compound_B_Evo_5, Compound_C_Evo_9,
Compound_D_Evo_6 and malabaricone_C.cdx_3_Evo_10) are testing on
human tyrosinase for biological activity. The results will been reported
in our next work.
Conflict of interest
Authors declare no conflict of interest.
Acknowledgments
This work was financially supported by Recruitment Program of
Global Experts, and the Director Foundation of XTIPC, CAS, Grant No.
2015RC011. This work was also financially supported by Natural
Science Foundation of Xinjiang, China, Grant No. 2016D01A073.
References
Bandgar, B.P., Adsul, L.K., Chavan, H.V., Shringare, S.N., Korbad, B.L., Jalde, S.S.,
157
Phytomedicine 38 (2018) 145–157
Lonikar, S.V., Nile, S.H., Shirfule, A.L., 2012. Synthesis, biological evaluation, and
molecular docking of N-{3-[3-(9-methyl-9H-carbazol-3-yl)-acryloyl]-phenyl}-benza-
mide/amide derivatives as xanthine oxidase and tyrosinase inhibitors. Bioorg. Med.
Chem. 20, 5649–5657. http://dx.doi.org/10.1016/j.bmc.2012.07.001.
Chase, M.R., Raina, K., Bruno, J., Sugumaran, M., 2000. Purification, characterization and
molecular cloning of prophenoloxidases from Sarcophaga bullata. Insect Biochem.
Mol. Biol. 30, 953–967.
Chen, C.Y.-C., 2011. TCM Database@Taiwan: the World's largest traditional Chinese
medicine database for drug screening in silico. PLOS ONE 6, e15939. http://dx.doi.
org/10.1371/journal.pone.0015939.
Chompoo, J., Upadhyay, A., Fukuta, M., Tawata, S., 2012. Effect of Alpinia zerumbet
components on antioxidant and skin diseases-related enzymes. BMC Complement.
Altern. Med. 12, 106. http://dx.doi.org/10.1186/1472-6882-12-106.
Dassault Systèmes BIOVIA, 2015. Discovery Studio Modeling Environment, Release 2016.
Dassault Systèmes, San Diego.
Dat, L.D., Thao, N.P., Luyen, B.T.T., Tai, B.H., Jeong, M.H., Woo, M.H., Kim, Y.H., 2015.
Identification of six new lupane-type triterpenoids from Acanthopanax koreanum
leaves and their tyrosinase inhibitory activities. ResearchGate. doi:http://dx.doi.org/
10.1016/j.bmcl.2015.12.020.
Debnath, A.K., 2002. Pharmacophore mapping of a series of 2,4-diamino-5-deazapter-
idine inhibitors of Mycobacterium avium complex dihydrofolate reductase. J. Med.
Chem. 45, 41–53.
Deng, X.-Q., Wang, H.-Y., Zhao, Y.-L., Xiang, M.-L., Jiang, P.-D., Cao, Z.-X., Zheng, Y.-Z.,
Luo, S.-D., Yu, L.-T., Wei, Y.-Q., Yang, S.-Y., 2008. Pharmacophore modelling and
virtual screening for identification of new Aurora-A kinase inhibitors. Chem. Biol.
Drug Des. 71, 533–539. http://dx.doi.org/10.1111/j.1747-0285.2008.00663.x.
Dong, X., Zhang, Y., He, J.L., Zhang, S., Zeng, M.M., Chen, J., Zheng, Z.P., 2016.
Preparation of tyrosinase inhibitors and antibrowning agents using green technology.
Food Chem. 197, 589–596. http://dx.doi.org/10.1016/j.foodchem.2015.11.007.
Gehlhaar, D.K., Verkhivker, G.M., Rejto, P.A., Sherman, C.J., Fogel, D.R., Fogel, L.J.,
Freer, S.T., 1995. Molecular recognition of the inhibitor AG-1343 by HIV-1 protease:
conformationally flexible docking by evolutionary programming. Chem. Biol. 2,
317–324. http://dx.doi.org/10.1016/1074-5521(95)90050-0.
Lee, K.-H., Ab Aziz, F.H., Syahida, A., Abas, F., Shaari, K., Israf, D.A., Lajis, N.H., 2009.
Synthesis and biological evaluation of curcumin-like diarylpentanoid analogues for
anti-inflammatory, antioxidant and anti-tyrosinase activities. Eur. J. Med. Chem. 44,
3195–3200. http://dx.doi.org/10.1016/j.ejmech.2009.03.020.
Liu, J., Wu, F., Chen, C., 2015. Design and synthesis of aloe-emodin derivatives as potent
anti-tyrosinase, antibacterial and anti-inflammatory agents. Bioorg. Med. Chem. Lett.
25, 5142–5146. http://dx.doi.org/10.1016/j.bmcl.2015.10.004.
Liu, P., Shu, C., Liu, L., Huang, Q., Peng, Y., 2016. Design and synthesis of thiourea
derivatives with sulfur-containing heterocyclic scaffolds as potential tyrosinase in-
hibitors. Bioorg. Med. Chem. 24, 1866–1871. http://dx.doi.org/10.1016/j.bmc.2016.
03.013.
Lu, I.-L., Tsai, K.-C., Chiang, Y.-K., Jiaang, W.-T., Wu, S.-H., Mahindroo, N., Chien, C.-H.,
Lee, S.-J., Chen, X., Chao, Y.-S., Wu, S.-Y., 2009. Corrigendum to “A three-dimen-
sional pharmacophore model for dipeptidyl peptidase IV inhibitiors” [Eur. J. Med.
Chem. 43 (2008) 1603–1611]. Eur. J. Med. Chem. 44, 2763. http://dx.doi.org/10.
1016/j.ejmech.2008.09.008.
Muegge, I., Martin, Y.C., 1999. A General and fast scoring function for protein−ligand
interactions: a simplified potential approach. J. Med. Chem. 42, 791–804. http://dx.
doi.org/10.1021/jm980536j.
Pan, T., Li, X., Jankovic, J., 2011. The association between Parkinson's disease and
melanoma. Int. J. Cancer 128, 2251–2260. http://dx.doi.org/10.1002/ijc.25912.
ResearchGate.
Sánchez-Ferrer, A., Rodríguez-López, J.N., García-Cánovas, F., García-Carmona, F., 1995.
Tyrosinase: a comprehensive review of its mechanism. Biochim. Biophys. Acta 1247,
1–11.
Solano, F., Briganti, S., Picardo, M., Ghanem, G., 2006. Hypopigmenting agents: an up-
dated review on biological, chemical and clinical aspects. Pigment Cell Res. Spons.
Eur. Soc. Pigment Cell Res. Int. Pigment Cell Soc. 19, 550–571. http://dx.doi.org/10.
1111/j.1600-0749.2006.00334.x.
Tang, H.-C., Chen, Y.-C., 2015. Identification of tyrosinase inhibitors from traditional
Chinese medicines for the management of hyperpigmentation. SpringerPlus 4, 184.
http://dx.doi.org/10.1186/s40064-015-0956-0.
Venkatachalam, C.M., Jiang, X., Oldfield, T., Waldman, M., 2003. LigandFit: a novel
method for the shape-directed rapid docking of ligands to protein active sites. J. Mol.
Graph. Model. 21, 289–307. http://dx.doi.org/10.1016/S1093-3263(02)00164-X.
Zheng, Z.-P., Zhang, Y.-N., Zhang, S., Chen, J., 2016. One-pot green synthesis of 1,3,5-
triarylpentane-1,5-dione and triarylmethane derivatives as a new class of tyrosinase
inhibitors. Bioorg. Med. Chem. Lett. 26, 795–798. http://dx.doi.org/10.1016/j.bmcl.
2015.12.092.
Zhou, Z., Zhuo, J., Yan, S., Ma, L., 2013. Design and synthesis of 3,5-diaryl-4,5-dihydro-
1H-pyrazoles as new tyrosinase inhibitors. Bioorg. Med. Chem. 21, 2156–2162.
http://dx.doi.org/10.1016/j.bmc.2012.12.054.
Zhu, Y.-J., Zhou, H.-T., Hu, Y.-H., Tang, J.-Y., Su, M.-X., Guo, Y.-J., Chen, Q.-X., Liu, B.,
2011. Antityrosinase and antimicrobial activities of 2-phenylethanol, 2-phenylace-
taldehyde and 2-phenylacetic acid. Food Chem. 124, 298–302. http://dx.doi.org/10.
1016/j.foodchem.2010.06.036.