Discovery of Novel CYP 17 Inhibitors through Homology Modeling and Structure-Based Drug Design
Metin Türkay İ. Halil Kavaklı Koç University, İstanbul, Turkey
Outline Introduction Prostate Cancer
Computational Studies Molecular Dynamics Simulations Molecular Docking Studies Analysis of Computational Results
Experimental Studies Construction of Expression Plasmids Expression Mammalian Cells Activity Assays Toxicity Assays
Lead Optimization QSAR
Summary & Conclusions
Structure-based drug design 3-dimensional structure of the target protein •by x-ray crystallography or NMR spectroscopy •by homology modelling
Identify and model the binding pocket
Screening of compounds •Experimental high-throughput screening •Computational virtual screening Potency optimization •Chemistry based libraries •Structure based optimization
Facts about Prostate Cancer • Most common age-related cause of cancer death among males. • Age and family history are risk factors. • Clinical studies indicated that high levels of androgens is common.
Androgens • Role of androgens – Sex-specific organ development and control • Role of androgen in prostate cancer – Major growth factor for prostate cancer cells; especially testosterone and dihydrotestosterone • Androgens and risk – Testosterone levels are proportional to prostate cancer risk
Testesterone
Dihydrotestesterone
Treatment of Prostate Cancer Hypothalamus
Estrogens
• Anti-androgens for androgen receptors • GnRH analogs and antagonists • Prostatechtomy and Orchiechtomy
Gonadorelin
Gonadorelin Analogs
Pituitary
LH
Adrenal
FSH
Testes
4-DIONE TESTO
Prostate (Carcinoma)
Antiandrogens
Androgen Synthesis Pathway
Target – CYP17 • Cytochrome p450 monooxygenase 17 α hydroxylase / 17, 20 lyase • Microsomal enzyme • Expressed in adrenal tissues and testicles
Target – CYP17 – Structure
2c17
Target Analysis Molecular Dynamics RMSD - Simulation 4 3,5 3 RMSD
• Molecular Dynamics simulation is done with NAMD using CHARMM force field parameters • ~10ns, with 20ps intervals.
2,5 2 1,5 1 0,5 0 0
2
4
6 Time (ns)
8
10
Target Analysis Molecular Dynamics
Target Analysis Molecular Dynamics
In-Silico Drug-Protein Interaction Tests • Virtual screening was done using Autodock – A drug candidate should decrease the energy of the ligand – protein complex when docked to the active site.
• Compound libraries (total 1,500,000) – – – –
Maybridge Ambinter WDI NCI Database
• First, analyze the natural substrates
Docking of Substrates -10.12 kcal/mol
-10.39 kcal/mol
-9.88 kcal/mol
-10.42 kcal/mol
Virtual Screening
Unsuccessful dockings
Compound Library
Virtual Screening
Successful dockings
Virtual Screening Molecules
Docking Energy (kcal/mol)
20543
-15,08
4216
-14,06
7435
-13,94
26310
-13,89
21614
-13,7
Virtual Screening
Steroidal
Non-Steroidal
Detailed Docking Analysis Non-Steroidal Molecules
ID
Virtual Screening
Detailed Docking
N01
-11,54 kcal/mol
-11.99 kcal/mol
N02
-12,47 kcal/mol
-13.24 kcal/mol
N03
-11,01 kcal/mol
-11,93 kcal/mol
N04
-10,96 kcal/mol
-11,30 kcal/mol
N05
-10,72 kcal/mol
-11,30 kcal/mol
N06
-10,04 kcal/mol
-10,81 kcal/mol
N07
-10.96 kcal/mol
-11,02 kcal/mol
N08
-9,12 kcal/mol
-11,11 kcal/mol
N09
-11,58 kcal/mol
-11,77 kcal/mol
N10
-11,52 kcal/mol
-11,93 kcal/mol
N11
-11,35 kcal/mol
-11,83 kcal/mol
N12
-10,75 kcal/mol
-10.82 kcal/mol
N13
-10,96 kcal/mol
-11,77 kcal/mol
N14
-11,88 kcal/mol
-12,25 kcal/mol
N15
-11,58 kcal/mol
-12,07 kcal/mol
N16
-10,57 kcal/mol
-12,26 kcal/mol
N17
-10,17 kcal/mol
-11,91 kcal/mol
N18
-12,20 kcal/mol
-12,43 kcal/mol
N19
-11,30 kcal/mol
-11,72 kcal/mol
N20
-11,30 kcal/mol
12,76 kcal/mol
Detailed Docking Analysis
Steroidal Molecules ID
Virtual Screening
Detailed Docking
S1
-10,52 kcal/mol
-11,91 kcal/mol
S2
-11,08 kcal/mol
-11,72 kcal/mol
S3
-11,10 kcal/mol
-11,63 kcal/mol
S4
-10,83 kcal/mol
-11,36 kcal/mol
S5
-10,03 kcal/mol
-10,21 kcal/mol
S6
-10,09 kcal/mol
-11,12 kcal/mol
S7
-10,51 kcal/mol
-11,14 kcal/mol
Summary of Virtual Screening Results • 20 nonsteroidal and 7 steroidal molecules were further analyzed for docking interactions. • In detailed docking runs 9 of these compounds were eliminated since they lack reactive species close to heme iron. • 18 compounds were subjected to experimental analysis.
Inhibition Tests • Intact cells transfected with CYP17 expression vector are used for inhibition tests. – Adsorption – Distribution – Metabolism – Excression
Experimental Studies
Inhibition (%)
100 50 0 0
20
40 N15 Concentration (µM)
60
80
Relative Luminescence (%)
Cell Viability Assay N15 Concentration vs Inhibition
100 80 60
N15 S3
40 20 0
50 µM
100 µM
200 µM
N15
70
36,2
35,4
S3
23,4
16,3
12,5
Candidate Concentration (µM)
CYP17 Expression Plasmid Expression of CYP17 protein • Mammalian expression vector is constructed. – pcDNA4+cyp17 for mammalian cell expression
CYP17 Expression
HEK-293t cells with pcDNA4+CYP17
Initial Inhibition Tests
Inhibition Assays – IC50 % Inhibition vs. S3 concentration 100
120
% inhibition of CYP17
% inhibition of CYP17
% Inhibition vs. N15 concentration 100 80 60 40 20 0 0
20
40
60
N15 concentration(uM)
IC50 = 35.65 µM
80
80 60 40 20 0 -20
0
20
40
60
S3 concentration (uM)
IC50 = 46.30 µM
80
100
Cell Viability Assays
Lead Optimization • 2 simultaneous approaches – Expertise • Analyze the docking positions of the lead compound • Modify the molecular structure in such a way that the interaction between the active site and drug candidate is increased
– QSAR Analysis • A novel data mining and classification approach
LEAD STRUCTURE Alkyl group for hydophobic tail
H bond donor/acceptor or coordination with heme iron Aromatic group to interact with heme group
Electron donor for coordination with heme iron
Optimization of Molecules Based on N15
OPTIMIZED STRUCTURES O
O N H
N H
NH2
O
O N H
O
O O
OH
OH
QSAR • Quantitative effort of understanding the correlation between the chemical structure of a molecule and its biological and chemical activities – i.e. biotransformation ability, reaction ability, solubility or target activity
• main idea – molecules that are structurally similar should have similar activities also
• Attention! – studying chemical structures of the candidate molecules Æ predict the drug activity
• HOW? – Detect the most significant chemical and structural descriptors of the drug candidates that affect the drug activity
Outline of our approach • Construct molecular structures of each drug, optimize their energy by minimization Ædetermine their confirmation in 3-D space (Marvin Sketch) • Generate chemical descriptors (E-Dragon) • Selected the most significant ones among these descriptors (PLS study, MINITAB) • Classify the drugs in terms of activity(MILP based hyperboxes method) • Perform significance analysis to improve classification accuracy
CLASSIFICATION CONCEPT SVM
MIP
x2
x1
x1
OPTIMIZATION MODEL ∨ YPB l
∀i
il
∨ YPC k
∨ YPB l
∀i
ik
⇔ ∨ YPCik
il
∀i
k
YBl ⇒ ∨ YBClk ∀l k
YBClk ⇒ ∨ YPBil ∀lk i
YBClk ⇒ ∨ YPCik ∀lk i
∧ YPBN ∧ YPBM n
m
ilmn
ilm
⇒ YPBM ilm ⇒ YPCik
YPCik ⇒ YP1ik ¬YPCik ⇒ YP2ik
∀ilm ∀ilk
∀ik ∉ Dik ∀ik ∈ Dik
Classification of drugs… • Apply the MILP based hyper-boxes method – By using the selected descriptors from the previous step
• The strength of hyper-boxes classification method Æ ability to use more than one hyper-box to define a class – Æ prevents overlapping in the classes
• Training and test sets are assigned by 10-fold cross validation !RESULT! The candidate drugs are classified as having low or high IC50 value
% accuracy
CYP17
7-attribute
10-attribute
15-attribute
MILP based hyper-boxes
100.00
100.00
100.00
Bayes Network
81.25
81.25
81.25
Naive Bayes
62.50
71.88
53.13
Naive Bayes Updatable
62.50
71.88
53.13
Lojistic
71.88
56.25
62.50
Multilayer Perceptron
62.50
71.88
59.38
SimpleLogistic
75.00
75.00
81.25
SMO
81.25
81.25
81.25
IB1
59.38
59.38
81.25
Logit Boost
71.88
62.50
62.50
Multi Class Classifier
71.88
56.25
62.50
Threshold Selector
43.75
40.63
62.50
LMT
75.00
75.00
81.25
RandomForest
75.00
68.75
65.63
OneR
75.00
71.88
75.00
OTHER TARGETS % accuracy
ACHE
% accuracy
7-attr
10-attr
100
91.89
89.19
Bayes Network
79.28
77.48
78.38
Naive Bayes
80.18
80.18
81.08
Naive Bayes Simple
81.08
80.18
81.98
Naive Bayes Updatable
80.18
80.18
81.08
Lojistic
79.28
84.68
80.18
Multilayer Perceptron
82.88
81.08
81.08
SimpleLogistic
83.78
82.88
79.28
SMO (WEKA SVM)
79.28
80.18
80.18
IB1
70.27
80.18
77.48
IBk
70.27
80.18
77.48
Logit Boost
82.88
81.08
82.88
Multi Class Classifier
79.28
84.68
80.18
Threshold Selector
47.75
68.47
60.36
LMT
83.78
82.88
79.28
RandomForest
80.18
80.18
81.98
OneR
81.08
72.97
72.97
MILP based hyper-boxes
15-attri
COX-2
7-attr
10-attr
15-attr
MILP based hyper-boxes
98.13
97.2
90.65
Bayes Network
67.2
67.2
66.88
Naive Bayes
71.66
70.06
64.65
Naive Bayes Simple
72.29
70.06
64.65
Naive Bayes Updatable
71.66
70.06
64.65
Lojistic
72.29
70.38
70.06
Multilayer Perceptron
72.61
72.29
75.16
SimpleLogistic
72.29
71.97
68.47
SMO (WEKA SVM)
71.02
69.43
69.43
IB1
69.11
71.02
70.06
IBk
69.11
71.02
70.06
Logit Boost
71.66
70.06
70.7
Multi Class Classifier
72.29
70.38
70.06
Threshold Selector
68.47
65.29
64.65
LMT
71.34
71.02
68.15
RandomForest
71.97
74.2
70.06
OneR
70.7
70.38
70.06
NUMBER OF ATTRIBUTES
Optimization of Molecules Based on N15
Optimization of Molecules Based on N15 Derivative Molecule
Improvement in Docking Energy
-0.29 kcal/mol
-0,53 kcal/mol
-0,96 kcal/mol
Summary & Conclusions
• Structure-Based Drug Design for Prostate Cancer case. • Use of computational model of target protein, CYP17. • Inhibition tests on intact cells. • Novel inhibitory compounds were identified. • Lead compounds with good inhibition and no toxic properties. • Lead optimization – QSAR: computational – IC50
Acknowledgements • M. Emre Özdemir • Pelin Armutlu • Şule Özdaş
Questions?