Drug Design

Drug design is an iterative process of search for suitable candidate(s) among several chemical compounds, which were initially chosen for having similar molecular structure, physico-chemical parameters and biological activity (-ies). Drug design is best defined as ligands design with the perspective of identification of drug targets. A target on the other hand is a protein, function of which can be selectively modified by interaction with a drug, to affect the symptoms or underlying cause of disease. Identification of an ideal target provides the focus for subsequent steps in the process of drug design. Single QSAR Model
Training Set
MLR, PLS, PCR Stepwise Linearetc

QSAR Model

Candidate Molecules

Prioritized Candidates

A goal in the early stages of drug discovery is the identification of one or more lead compounds. A lead compound is any substance that shows the biological activity of interest. It demonstrates that the particular compound have some interesting chemical structures that possesses at least some of the desired properties.

Fig 1: Amino acids in the dopamine receptor (and their position in the overall receptor primary structure) important for the binding interaction with dopamine

Quantitative Structure Activity Relationships

Figure 2 - An example of an antifungal base structure used for QSAR studies by varying R and R'.

Investigations of the relationship between chemical structure and the activity of compounds (structure activity relationships, SAR) helps in understanding the activity of interest and may enable prediction of the activity of new compounds based on knowledge of the chemical structure alone. These predictions can be achieved by quantifying the SAR. These Quantitative Structure-Activity Relationships (QSAR) have been pioneered by Hansch and co-workers in the 1960s. It is assumed that the sum of substituent effects on the steric, electronic and hydrophobic interaction of the compounds with some endogenous receptor, determines the biological activity BA, as illustrated roughly in Equation 1. BA = { (steric) + (electronic) + (hydrophobic) } interactions BA: Biological Activity

Explanation:

This linear free energy relationship is called extra thermodynamic, because it is only a rough approximation that the biological activity can be reduced to free energy of binding to or reaction with a biomolecule. When examining the potency of a drug (the dosage required to produce a biological effect), the change in free energy [ Go] can be calculated to be proportional to the inverse logarithm of the concentration [S] of the compound. Go = - 2.3RTlogK = log 1/[S] The relations are evaluated with the help of statistical techniques like linear and non-linear regression, discriminant, cluster and factor analysis. The successes of this simple approach are manifold [see Quantitative Drug Design, 1990]. It is evident that the choice of physicochemical properties is of crucial importance for satisfactory results. QSAR without knowledge of physicochemical properties is possible using the Free-Wilson procedure [Kubinyi 1988]. Statistical considerations for the derivation of a meaningful QSAR are given by Craig [1971] and Topliss [1972, 1979].

The steric substituent effects can be modelled with bulk properties like molar volume or molar refraction. The Es parameter as tabulated by Taft [1956] is thought to model steric effects at least equally good. To account for different geometries at equal overall size, Verloop et al. [1976] tabulated the length and width in four directions of the spacefilling Corey Pauling Koltun (CPK) molecule models, in order to reflect the shape of a particular substituent in their STERIMOL approach. A more three-dimensional description of the geometry of the substituents would lead to far more parameters and hence, possibly lower statistical significance of the QSAR. Lehmann [1987] proposed the use of Quantitative Stereo-Structure Activity Relationships using an eudismic index to redefine the observed biological activities of chiral drugs [see also Arins 1989]. How to obtain QSAR equations: 1. A congeneric series of compounds : This is a series of compounds with a similar basic structure but with varying substituents (side chains). These are known to act on the same target in the same manner (see Fig 2)
2. Some quantitative activity data for this series of compounds. This

data is generated experimentally, assuming that all compounds have the same mode of action. Eg. - kcat data for enzymes - Measuring receptor activation through measurement of molecules known to arise from that receptor's signal transduction pathway. 3. Some quantitative chemical, physical or structural parameters need to be known for the varying substituents. Since the structures only vary with respect to certain substituents, only the structural parameters of these

substituents need to be known. Each substituent is described by a substituent constant, which quantifies a particular aspect of structure. Examples: - Hammett Electronic Substituent Constant (sigma): quantifies the electron withdrawing or electron donating effect of a substituent. - Taft Steric Substituent Constant (Es): quantifies the substituent size relative to Me (Minimize energy). - Hansch/Fujita Hydrophobicity substituent constant (Pi): measures the contribution, which a substituent makes to the total partition coefficient of the molecule. - Charge on certain atoms.
4. A powerful computer capable of performing complex statistical

analysis. The activity is usually dependent on more than one aspect of structure. This will be seen if a non-linear plot arises from Activity vs. Hammett Electronic Substituent Constant ( ) for example, as shown in Graph 1.

Graph 1 - Activity vs Hammett Electronic Substituent Constant is non-linear, thus activity dependent on more than one aspect of structure.

Thus, more powerful statistical analysis is required. A commonly used technique is Multiple Linear Regression, which plots each substituent constant on a multidimensional graph, against activity. A line of best fit is then found in multiple dimensions with a corresponding Regression Coefficient (R2). The regression coefficient indicates---how well the points fit the proposed line. R2 lies between -1 and 1, with a value of -1 or 1 indicating that all the variation on the line can be explained. If R2 lies close to 0, no variation observed can be explained by the variation in substituent constants. In order to be statistically significant, the original group of congeneric molecules should number at least 5. If the substituent constants are highly correlated or are very numerous, it is possible to exploit a partial least squares method, which expresses a dependent variable, y, in terms of a linear combination of the independent variables:

Equation to explain Partial Least Squares analysis. The overall results are a QSAR equation as shown in

An example of a QSAR equation, as first derived by Hansch (1969).

Conclusion: This can be used to predict the activity of a novel compound once the substituent constants are known. It may also give insight into the mechanism of action of the drug. Short Notes: Hammett Sigma Constant: Electronic substituent effects have been quantified by Hammett [1937]. He tabulated the change in acidity of substituted benzoic acids as compared to the non-substituted benzoic acid BA (Equation 2). log (K/KBA) = ------- Equation 2 (rho factor). Still, (sigma factor) does not The susceptibility of other solute systems to these electronic substituent effects is reflected in tabulated adapted represent a completely transferable substituent effect. Other investigators values for different solute compounds or reaction constants

mechanisms. Swain and Lupton [1968] analysed most of these

statistically and tabulated F and R for the field inductive and resonance effects. Quantum mechanically computed properties have also been used to reflect certain electronic interactions. Quantum mechanical analyses of pKavalues have been reported [Karaman 1990, Mitchell 1990]. It would be preferable, however, to use parameters that more reflect the mechanism of action. If biological activity is attributed to electron or proton transfer, the E1/2 or pKa values of the compounds would be more meaningful. Since electrostatic interaction can also be based on dipole-dipole or dipole-charge interaction, the dipole moment can also be used [Lien 1982].

The majority of QSARs involves hydrophobicity properties like log Po/w or o/w (the corresponding substituent constant of benzene substitution, [Fujita 1964]), where o/w is the n-octanol/water partitioning system. Such relationships either reflect transport to the site of action or binding to a hydrophobic site in a receptor molecule. The first QSAR may have been the relationship between the narcotic activity and the partition coefficient of anaesthetics as found by Meyer [1899] and Overton [1897]. Measurement of partition coefficients can be achieved with a shake-flask procedure using noctanol/water [Dearden 1985] or with reversed-phase HPLC (RP-HPLC) [Braumann 1986]. The stationary phase is usually an octadecyl ether of silanol; the use of n-octanol produced technical problems [Mirrlees 1976]. A detailed description by Martin [1978] shows that in QSARs the partition coefficient cannot be simply replaced by a distribution coefficient in case of ionic species. De novo calculation of partition coefficients, that cannot be determined reliably, can be performed with the approach of additivity of fragment constants, as proposed by Rekker [Nys 1973, Rekker 1979, Hansch 1979 a]. The additivity is improved by a 'magic constant' of 0.28 log units for which no physical explanation has yet been given. Others have used a questionable constant of 0.07 log units for corrections, approaching the standard error of most measurements of partition coefficients [Van de Waterbeemd 1983]. The thermodynamic relationship between the fragment constants and was discussed by Janssen [1976]. Several authors have proposed the use of indicator variables, counting the possible number of hydrogen bonds that can be formed [Ou 1986, Yang 1986], in order to correct for differences in hydrogen bonding behaviour between the lipid parts of the biological system of interest and n-octanol. Quantum chemical

calculations of the partition coefficient have also been used with some success [Cammarata 1971, Kamlet 1988, Bodor 1989]. Role of Linear Regression in Determination of Chemical Activity: In chemistry, linear regression has provided means for identifying which structural features may be important in determining chemical activity over a group of reactants in a poorly understood interaction. This is achieved by forming a linear relationship between those variables that describe the structural variation within the group of reactants and those that describe the activities of the reactants. The relationship is denoted as a quantitative structure activity relationship (QSAR). QSAR studies have therefore formed a close association with combinatorial chemistry studies in which variations in activities caused by the systematic modifications of structures can yield insight into the reaction activity mechanism. A particularly interesting example of such a QSAR is comparative molecular field analysis (CoMFA), in which reactants with common structural backbones varying by residue substitutions may be aligned with one another, and physical characteristics such as electrostatic potential and steric energies may be measured for each reactant on a common grid with thousands of points. This is doubly interesting because it explicitly uses descriptors that depend only on the 3D character of a molecule rather than any information on its 1D topology. CoMFA QSAR regression systems are grossly underdetermined. However, it should be expected that the variations at the various field points should be correlated well enough (because of the relatively small number of residue substitutions as well as the discrete character of residue substitution in combinatoric studies) that a meaningful relationship between those residue

substitutions and the activities may be determined. Then those grid points around the residue sites that are important to the determination of the activity will make large contributions to the regression.