Transcript Lecture 12, computers CORRECTED

Modern Tools of Drug Discovery
Combinatorial synthesis
• Efficient way to make structurally diverse compounds fast
• Can rely on automation and solid phase techniques
• Identification of active compound can be challenging
• Useful in lead identification; SAR; lead optimization
High throughput screening - rapid screening of lots of compounds
Computers in drug design: Molecular modeling, computer algorithms
Computers in drug design: Where do they play a role?
A. Molecular visualization
B. Energy of different conformations (Conformational analysis).
Bioactive conformation determination
C. 3D Pharmacophore generation, molecular mimicry
• alignment of molecules
• Calculation of/comparison of physiochemical features
such as electrostatic potential or logP
D. Receptor-based drug design (docking; de novo construction)
E. Receptor Mapping
F. Quantitative Structure-Activity Relationships (QSAR)
Two major Computational methods for the Calculation of
Structure and Property Data
1. Molecular mechanics:
• atoms are spheres, bonds are springs
•
•
•
•
•
Giant classical mechanics (classical physics) problem
Calculated Energies/”force fields” are relative - good for
comparing two conformations of the same molecule
Fast
Cannot be used to calculate electronic properties (no
electrons in the model!)
Useful for: Energy minimization, Identification of stable
conformations, Energy calculations for specific
conformations, Molecular motion
Two major computational methods for the Calculation of
Structure and Property Data
2. Quantum mechanics
• Nuclei and electrons of the molecules are considered
• Giant quantum physics problem
HY = EY
• Requires many approximations to make these problems
tractable. (Example: nuclei are motionless, electrons move;
electrons move independently of one another)
• Two quantum mechanical approaches
Ab initio - more rigorous, from first principles (no stored
parameters or data), takes a long time, restricted to small
molecules
Semi-empirical - faster, but less accurate, can be used on
larger molecules (MNDO, AM1, PM3)
• Useful for MO energies, partial charges, electrostatic
potentials, dipole moments
Computers in drug design: Where do they play a role?
A. Molecular visualization - 3D experimental data from
crystal structures or NMR (x, y, z coordinates)
Acetylcholinesterase + Aricept
(anti-Alzheimer drug)
Computers in drug design: Where do they play a role?
A. Molecular visualization - 3D structure from 2D drawing.
A. Molecular visualization - 3D structure from 2D drawing.
How is it done?
1. Draw the molecule in 2D in the modeling program.
2. Perform an energy minimization to “fine tune” the structure by
removing unfavorable bond lengths, bond angles, torsion angles,
sterics. See movie on MC A5.5.1-8
A. Molecular visualization - 3D structure from 2D drawing.
Energy minimization (continued)
Minimize energy of molecule by optimizing the following:
Bond Stretching:
Bond angle bending:
Torsion angles:
A. Molecular visualization - 3D structure from 2D drawing.
Energy minimization (continued)
Minimize energy of molecule by optimizing the following:
Van der Waals Interactions:
Dipole-dipole Interactions
Hydrogen Bonding:
A. Molecular visualization - 3D structure from 2D drawing.
Energy minimization (continued)
After performing this energy minimization, you have a reasonable
structure for the molecule in 3D.
Note: Energy minimization stops at the first stable
conformation it finds - closest in structure to the starting
structure. This may be a local minimum, and not the global
minimum. It may or may not be the bioactive conformation.
Spartan: Example of molecular modeling software
2. Minimize (molecular mechanics)
1. Build structure in 2D
3. Set up and perform quantum
calculations
4. Examine results: here, an
electrostatic potential surface
Computers in drug design: Where do they play a role?
B. Energy of different conformations (Conformational analysis).
Bioactive conformation determination. In general, one wants to
know the lowest energy conformation plus all other
conformations within 12kJ/mole of the global minimum.
Recall a flexible molecule may have more than one stable
conformation.
B. Energy of different Conformations (continued)
Molecule with one rotating bond.
A few possible stable
conformations:
Molecule with >1 rotating bond
More possible stable
conformations:
B. Energy of different conformations (continued)
Stable conformations occur at local and global minima:
Many, MANY possible
Stable conformations
B. Energy of different conformations (continued)
Goal: To generate by computer all possible different stable
conformations of the molecule (to help identify possible bioactive
conformations, for example).
Method 1: molecular dynamics = simulated heating and cooling of
the molecule to accumulate all possible conformations.
Method 2: stepwise rotation of bonds (rotate each bond of the
molecule by a set number of degree increments).
B. Energy of different Conformations (continued)
The resulting representative conformations for each local
minimum should be minimized to obtain the lowest energy
conformation.
Energy
Minimization
•Now a group of conformations with structures and energies at local
and global minima has been generated.
•They can be analyzed to determine the bioactive conformation
(systematically compare reasonable minimized conformations of all
active compounds; look at X-ray data; compare activity of rigid and
nonrigid analogs).
•Better molecules can be synthesized based on this information.
B. Energy of different Conformations (continued)
Example: Captopril
Rigid analogs were designed to keep amide and CO2 in place,
but providing a range of locations for SH; tested for activity...
A group of stable conformations (structures local and global
minima) were generated by computation for all active species and
compared. The overlap of all active compounds reveals a bioactive
conformation.
Computers in drug design: Where do they play a role?
C. 3D Pharmacophore generation, molecular mimicry
Recall alignment of molecules to generate a pharmacophore:
Example: Overlay of procaine and cocaine: 2D overlap is misleading
A computer algorithm can then use the 3D pharmacophore to perform
a 3D database search to find new leads!
C. 3D Pharmacophore generation, molecular mimicry (continued)
Calculation of/comparison of physiochemical features such
as electrostatic potential or logP
Example 1:
The electrostatic potential surface above explains why
binding a cationic (positively charged) compound through the
cation-pi interaction decreases from left to right. The cationpi interaction is electrostatic, so the additional fluorines
substituted on the aromatic ring decrease the negative charge
that stabilizes the cationic compound.
C. 3D Pharmacophore generation, molecular mimicry (continued)
Calculation of/comparison of physiochemical features such
as electrostatic potential or logP
Example 2: Prediction of pharmacokinetic properties (ADMET)
Prescreening based on “Rule of Five”: MW<500; logP<5; no more
than 5 H-bond donors; no more than 10 H-bond acceptors
Computers in drug design: Where do they play a role?
D. Receptor-based drug design (docking; de novo construction)
• 3D structure of the receptor is known:
D. Receptor-based drug design (continued)
De novo design by
3D database searching computer algorithms
Computer-based structure building or linking (atoms, fragments)
building
linking
Example: Thymilidate synthase inhibitors bind at coenzyme binding
site of the target enzyme - antitumor agents that block DNA synthesis.
de novo structure
D. Receptor-based drug design (continued)
Docking. If the binding groups on the ligands and the target are
known (or automatically selected by the program), a program can
move a rigid ligand around a rigid binding site to optimize binding
interactions. Once the docking is finished, the energy of the system is
minimized (ligand and binding site).
Docking
Docking provides information about:
Role of each functional group in the ligand and/or binding site
Active conformation of ligand
Modifications of known ligands that will enhance binding
Prediction of activities of unknown compounds - virtual screening
of libraries of compounds. Challenge: comparison of binding
energies of structurally different ligands (called “scoring”).
Computers in drug design: Where do they play a role?
E. Receptor Mapping
If primary amino acid sequence of the receptor is known and the Xray structure of a related protein has been determined: Construct a
model receptor:
1. The known structure is used as a template. The backbone of the new receptor is
constructed on a computer to match that of the known protein.
2. Side chains are added in favorable conformations, and the energy is minimized
by computer.
3. Key residues in the new receptor are identified and tested experimentally (sitedirected mutagenesis).
E. Receptor Mapping (continued)
If no homologous proteins are known: Construct a model binding site:
1. Use experimental activity data for a diverse range of small molecules to construct a
3D pharmacophore.
2. Compute interaction energies for hydrophobic, ionic, polar interactions all around
the molecule. Significant interactions are highlighted by “isoenergy contours.” SAR
confirms these.
3. Suitable amino acids can be positioned around the molecule.
4. This model can be tested by comparing experimental binding data to calculated
binding energies of docked ligands to this model binding site.
After constructing and testing the
validity of a model receptor or
model binding site, new molecules
can be designed for improved
binding, selectivity, or other
properties.
Case study: Development of a drug using multiple techniques
Target: serotonin receptor. An antagonist will help treat anxiety,
depression, migraines. This antagonist should be selective for a
certain type and subtype of receptor (for 5HT2C over 5HT2A) to
avoid side effects.
Lead: drug already developed by Lilly; it was insoluble. Modified
structure to improve solubility.
What is the chemical modification here? How does it
increase solubility?
Case study (continued)
This structure was also modified: What is the modification and why?
H
N
N
O
N
(CH2 )n
N
CH3
If n=5, affinity increased; if n = 6, affinity decreased.
Molecular modeling was used to understand why. Minimized
structures:
n=5
n=6
Better affinity, better selectivity;
but metabolic instability...
Case study (continued)
Additional chemical modifications: What are they called? What do
they suggest?
N - ethyl or propyl:
Similar affinity, slightly increased
selectivity.
N-benzyl: decreased affinity and
selectivity.
Molecular modeling was used to generate a pharmacophore for
affinity AND selectivity. The region crucial for selectivity was
found to be where the N-methyl substituent is located.
Molecular modeling was used to generate a model receptor based
on the structure of a related protein.
Case study (continued)
Molecular modeling was used to generate a model receptor based
on the structure of a related protein.
It was thought that the serine residues H-bond to the carbonyl group,
while the hydrophobic pockets fit the rings. Docking calculations
were performed to determine the most stable binding mode.
Case study (continued)
Best binding mode:
Note: N-methyl group is in a pocket with two valine residues. The
other receptor subtype has leucine residues there. How might that
explain the increased selectivity with N-ethyl vs. N-methyl?
Next, combinatorial synthesis was used to generate multiple structures
for testing to improve selectivity
Case study (continued)
Quantitative structure-activity relationships show these
substituents lead to the best compromise of binding affinity and
selectivity
Docking was performed again:
Note: -CF3 acts as a
conformational blocker (forcing
SCH3 into hydrophobic pocket for
selectivity enhancement)
What further changes in structure
could improve binding affinity?
Case study (continued)
Analogs were synthesized to add groups to fill the hydrophobic
pocket of the target, to improve metabolic susceptibility, and to
maintain selectivity of binding to the correct receptor subtype.
Binding data as well as further molecular modeling of ligand and
binding site of analogs were used to develop the compound below.
That compound was sent on to clinical trials.
Case study 2: J. Med. Chem. 2001, 44, 4615-4627
Cyclin-dependent kinases (Cdk): regulate the cell cycle. Cancer is
associated with abnormal cell cycle regulation. Inhibitors for G1
phase of cell cycle (over other phases) are desirable: must inhibit
Cdk4.
Possible structure-based approaches:
3D database search:
commercially available, so synthetically accessible
Perhaps not correct ionization (databases are neutral)
Not all conformations are searched
No novel structures
De novo:
Computers generate structures that are difficult to synthesize
So, these investigators chose de novo plus a program to find
synthetically accessible related compounds.
Step 1: Use homologous protein structure of Cdk2 to create a receptor
model on computer. (DON’T FORGET: Cdk4 is the TARGET!)
Inhibitors of Cdk2 (A - C) and predicted binding region for Cdk4 (D)
De novo design program LEGEND is used to generate scaffolds:
builds atom-by-atom.
Below, I is one output of LEGEND. II is a simpler structure to use
in 3D computer database searches.
Automated de novo
structure building program
Special program to find
easier compounds to
synthesize
382 compounds were
purchased and
screened for activity
Hits from the
computer database
search were divided
into four structural
classes
One of the classes with multiple hits that also will lend itself to
parallel synthesis was picked: Diaryl ureas
Structure-activity
relationships (SAR)studies
were performed. One
compound was chosen as
the new lead…
This lead was
docked to the model
site of Cdk4 on
computer. Multiple
conformations and
binding modes were
examined and
minimized.
Comparison of the four modeled binding modes (A-D) to
experimental data (SAR) ruled out two of the modes. The remaining
binding modes were compared, and mode A was determined to be a
stronger binding mode.
Examining this mode
revealed a steric
repulsion in the ligand as
well as two areas to
“grow” the ligand to
better fit the site. New
analogs were synthesized
and tested.
One ligand was chosen as a new lead IC50 for Cdk4 = 0.042 mM
However, there was no selectivity for Cdk4 over Cdk1/2.
Xray structure of the new lead
compound with Cdk2.
Future work involved comparing receptor structures to design an
inhibitor with increased selectivity for Cdk4 over other subtypes.
References
Patrick, G. L. An Introduction to Medicinal Chemistry; Oxford University
Press: New York, NY, 2001.
Guidebook on Molecular Modeling in Drug Design; Claude Cohen, N.
Ed.; Academic Press: San Diego, CA, 1996.
Chemoinformatics; Gasteiger, J.; Engel, T., Eds.; Wiley-VCH: Weinheim
2003.
Molecular Conceptor Volume A