1.
Activity RelationshipsActivity RelationshipsQuantitative StructureQuantitative Structure--Saramita ChakravartiSaramita ChakravartiResearch Scientist II (iResearch Scientist II (i((Chembiotech LaboratoriesChembiotech Laboratories

2.
Why QSAR?The number of compounds required forsynthesis in order to place 10 different groupsin 4 positions of benzene ring is 104Solution: synthesize a small number ofcompounds and from their data derive rules topredict the biological activity of othercompounds.

3.
QSAR and Drug DesignCompounds + biological activityNew compounds withimproved biological activityQSAR

4.
What is QSAR?A QSAR is a mathematical relationshipbetween a biological activity of a molecularsystem and its geometric and chemicalcharacteristics.QSAR attempts to find consistent relationshipbetween biological activity and molecularproperties, so that these “rules” can be used toevaluate the activity of new compounds.

5.
Statistical ConceptsInput: n descriptors P1,..Pn and the value ofbiological activity (EC50 for example) form compounds.Bio P1 P2 ... PnCpd1 0.7 3.7Cpd2 3.2 0.4…Cpdm

6.
Statistical ConceptsThe problem of QSAR is to find coefficientsC0,C1,...Cn such that:Biological activity = C0+(C1*P1)+...+(Cn*Pn)and the prediction error is minimized for a list ofgiven m compounds.Partial least squares (PLS) is a technique used forcomputation of the coefficients of structuraldescriptors.

7.
3D-QSARStructural descriptors are of immense importance inevery QSAR model.Common structural descriptors are pharmacophoresand molecular fields.Superimposition of the molecules is necessary.3D data has to be converted to 1D in order to usePLS.

8.
3D-QSAR AssumptionsThe effect is produced by modeled compound and not it’smetabolites.The proposed conformation is the bioactive one.The binding site is the same for all modeled compounds.The biological activity is largely explained by enthalpicprocesses.Entropic terms are similar for all the compounds.The system is considered to be at equilibrium, andkinetics aspects are usually not considered.Pharmacokinetics: solvent effects, diffusion, transport arenot included.

9.
QSAR and 3D-QSARSoftwareTripos – CoMFA, VolSurfMSI – Catalyst, SeriusDocking SoftwareDOCK – KuntzFlex – LengauerLigandFit – MSI Catalyst

10.
3D molecular fieldsA molecular field may be represented by 3Dgrid.Each voxel represents attractive and repulsiveforces between an interacting partner and atarget molecule.An interacting partner can be water, octanol orother solvents.

11.
Common 3D molecular fieldsMEP – Molecular Electrostatic Potential (unit positivecharge probe).MLP – Molecular Lipophilicity Potential (no probenecessary).GRID – total energy of interaction: the sum of steric(Lennard-Jones), H-bonding and electrostatics (anyprobe can be used).CoMFA – standard: steric and electrostatic, additional:H-bonding, indicator, parabolic and others.

12.
Comparative Molecular FieldAnalysis (CoMFA) - 1988Compute molecular fields gridExtract 3D descriptorsCompute coefficients of QSAR equation

13.
CoMFA molecular fieldsA grid wit energy fields is calculated by placinga probe atom at each voxel.The molecular fields are:Steric (Lennard-Jones) interactionsElectrostatic (Coulombic) interactionsA probe is sp3carbon atom with charge of +1.0

14.
CoMFA 3D-QSAREach grid voxel corresponds to two variablesin QSAR equation: steric and electrostatic.The PLS technique is applied to compute thecoefficients.Problems:Superposition: the molecules must beoptimally aligned.Flexibility of the molecules.

15.
3D-QSAR of CYP450camwith CoMFA• Training dataset from 15 complexes ofCYP450 with different compounds was used.• The alignment of the compounds was done byaligning of the CYP450proteins from thecomplexes.

16.
3D-QSAR of CYP450camwith CoMFAMaps of electrostaticfields:BLUE - positive chargesRED - negative chargesMaps of steric fields:GREEN - space fillingareas for best KdYELLOW - spaceconflicting areas

17.
VOLSURFThe VolSurf program predicts a variety ofADME properties based on pre-calculatedmodels. The models included are:• drug solubility• Caco-2 cell absorption• blood-brain barrier permeation• distribution

18.
VOLSURFVolSurf reads or computes molecularfields, translates them to simple moleculardescriptors by image processing techniques.These descriptors quantitativelycharacterize size, shape, polarity, andhydrophobicity of molecules, and the balancebetween them.

19.
VOLSURF DescriptorsSize and shape: volume V, surface area S, ratio volumesurface V/S, globularity S/Sequiv (Sequiv is the surface area of asphere of volume V).Hydrophilic: hydrophilic surface area HS, capacity factorHS/S.Hydrophobic: like hydrophilic LS, LS/S.Interaction energy moments: vectors pointing from thecenter of the mass to the center of hydrophobic/hydrophilicregions.Mixed: local interaction energy minima, energy minimadistances, hydrophilic-lipophilic balance HS/LS, amphiphilicmoments, packing parameters, H-bonding, polarisability.

20.
VOLSURFhydrophobic (blue) and hydrophilic (red) surfacearea of diazepam.

21.
CatalystCatalyst develops 3D models (pharmacophores) from acollection of molecules possessing a range of diversity inboth structures and activities.Catalyst specifies hypotheses in terms of chemicalfeatures that are likely to be important for binding to theactive site.Each feature consists of four parts:Chemical functionLocation and orientation in 3D spaceTolerance in locationWeight

22.
Catalyst Features• HB Acceptor and Acceptor-Lipid• HB Donor• Hydrophobic• Hydrophobic aliphatic• Hydrophobic aromatic• Positive charge/Pos. Ionizable• Negative charge/Neg. Ionizable• Ring Aromatic

23.
Catalyst HipHopFeature-based pharmacophore modeling:uses ONLY active ligandsno activity data requiredidentifies binding features for drug-receptorinteractionsgenerates alignment of active leadsthe flexibility is achieved by using multipleconformersalignment can be used for 3D-QSAR analysis

24.
Catalyst HipoGenActivity-based pharmacophore modeling:uses active + inactive ligandsactivity data required (concentration)identifies features common to actives missed byinactivesused to “predict” or estimate activity of new ligands

25.
Catalyst CYP3A4substrates pharmacophoreHydrophobic area, h-bonddonor, 2 h-bond acceptorsSaquinavir (most active compound)fitted to pharmacophore

26.
Catalyst CYP2B6substrates pharmacophore3hydrophobic areas, h-bondacceptor7-ethoxy-4-trifluoromethylcoumarinfitted to pharmacophore

27.
Catalyst Docking – Ligand FitActive site findingConformation search of ligand against siteRapid shape filterdetermines whichconformations should bescoredGrid-based scoring for thoseconformations passing the filter

28.
Catalyst Docking – LigandFlexibilityMonte Carlo search in torsional spaceMultiple torsion changes simultaneouslyThe random window size depends on thenumber of rotating atoms

29.
Catalyst Docking – ScoringpKi = – c – x (vdW_Exact/ Grid_Soft)+ y (C+_pol)– z (Totpol^ 2)• vdW = softened Lennard-Jones 6-9 potential• C+_pol = buried polar surface area involved in attractiveligand-protein interactions• Totpol^ 2 = buried polar surface area involved in bothattractive and repulsive protein-ligand interactions

30.
3D-QSAR of CYP450camwith DOCKGoal:• Test the ability of DOCK to discriminate betweensubstrates and non-substrates.Assumption:• Non-substrate candidate is a compound thatdoesn’t fit to the active site of CYP, but fits to thesite of it’s L244A mutant.

31.
MethodsDocking of 20,000 compounds to ‘bound’ structure ofCYP and L244A mutant.11 substrate candidates were selected from 500 highscoring compounds for CYP.6 non-substrate candidates were selected from adifference list of L244A and CYP.Optimization of compounds 3D structures by SYBYLmolecular mechanics program and re-docking. As a result 2compounds move from “non-substrate” list to “substrate”list and one in the opposite direction.

32.
Prediction ResultsAll compounds predicted as “non-substrates” shown nobiological activity.4 of the 11 molecules predicted as “substrates” werefound as non-substrates.The predictions of DOCK are sensitive to the parameterof minimum distance allowed between an atom of theligand and the receptor (penetration constrains).

33.
Prediction Results

34.
ReferencesCruciani et al., Molecular fields in quantitative structure-permeationrelationships: the VolSurf approach, J. Mol. Struct. (Theochem), 2000,503:17-30Cramer et al.,Comparative Molecular Field Analysis (CoMFA). 1. Effect ofshape on Binding of steroids to Carrier proteins, J. Am. Chem. Soc. 1988,110:5959-5967Ekins et al., Progress in predicting human ADME parameters in silico, J.Pharmacological and Toxicological Methods 2000, 44:251-272De Voss et al., Substrate Docking Algorithms and Prediction of theSubstrate Specifity of Cytochrome P450cam and its L244A Mutant, J. Am.Chem. Soc. 1997, 119:5489-5498Ekins et al., Three-Dimensional Quantative Structure Activity RelationshipAnalyses of Substrates for CYP2B6, J. Pharmacology and ExperimentalTherapeutics, 1999, 288:21-29