|Click here for information on how to order reprints of this article.|
Computational Chemistry Studies on a New Class of Peptide Antidepressants*
Joseph J. Hlavka, PhD
1 Maynard Drive, Suite 205
Park Ridge, NJ 07656
*This study was sponsored by Innapharma, Inc.
KEY WORDS: computational chemistry, antidepressant peptides, regression coefficients and quantitative structure activity relationships
Computational chemistry studies were performed to determine the quantitative structure activity relationships (QSAR) of a new series of small peptides that have demonstrated potent antidepressant activity in animal models for depression and in Phase II clinical trials. When comparing graphically the calculated activity value (Z*) with the observed activity value (Z), we observe an excellent correlation between the two, resulting in a good predicator of activity.
Heterogeneous unipolar depression is a common psychiatric disorder that is mediated by neurochemical changes in the central nervous system. Administration of antidepressant drugs for the treatment of unipolar depression has gained wide acceptance in the medical community over the past decades. Endogenous depression, which is a genetically determined biochemical disorder, is classified as unipolar depression.
We have synthesized a series of new peptides1 that have demonstrated potent antidepressant activity in animal models for depression2,3 and in Phase II clinical trials.4,5
MATERIALS AND METHODS
Computational chemistry studiesa were performed to determine the quantitative structure activity relationships (QSAR) of peptides in this series. A three-dimensional structure was computed for a neutral species in the gas phase for each of eleven peptides in our series. An example for our lead compound, nemifitide, is given in Figure 1. The structures were then used to determine twelve physiochemical parameters, called descriptors, for the eleven peptides. Some of the descriptors have marked interdependencies. For example, molar refraction is highly correlated with molecular volume, so one may not want to use both. After removal of these dependencies, the number of descriptors was easily reduced to the following seven: heat of formation, ionization energy, dipole moment, polarizabilityb (second derivative term), molecular volume, octanol/water partition coefficient (log P), and acid dissociation constant in the physiological range (pKa).
Biological data in terms of the experimental Z in Table 1 are represented by stress-induced depression testing2 measurements of Z score and percentage of responders.b
We reduced the physicochemical units to a common scale (i.e., the deviation from the group mean divided by the standard deviation) to enable comparison of descriptors with disparate units, and then performed three types of correlational analyses: multiple linear regression analysis (MLR), principal component analysis (PCA), and partial least squares analysis (PLS).
The MLR technique gave a good fit, but the coefficients were poorly determined (almost all were nonsignificant by t‑test). After standardization of the descriptors, three principal components were computed and antidepressant activity was best correlated with the third component. Table 2 lists the regression coefficients derived after translation from PCA analysis back to the original variables. Finally, data were subjected to PLS analysis; one dimension was sufficient. The PLS coefficients are also shown in Table 2. For ease of comparison among and between the methods, the regression coefficients for all descriptors for both methods of analysis are expressed in arbitrary units, adjusted to set the coefficient of log P equal to 2.0.
RESULTS AND DISCUSSION
Results of both PCA and PLS analysis agree that the octanol/water partition (log P) is the most important factor in predicting antidepressant activity. This is not unexpected since compounds with higher log P more readily cross biological membranes in general and the blood-brain barrier in particular.
In addition, both methods predict that nemifitide should be the most active and that Innapharma compound (IC) 00955 and IC 00924 have comparable activity. This predicted profile of activity was observed in our animal models for depression.2
The calculated (predictive fuction) Z, when expressed in standardized terms (i.e., in units of standard deviation), is as follows:
Z 5 O.65 Heat - 0.96 Ionization - 0.015 Dipole moment 1 0.77 Polarz'nB 1.89 Volume 1 1.76 LogP - 0.46 pKa
When comparing graphically the calculated Z* with the observed Z, we observe an excellent correlation between the two, resulting in a good predicator of activity (Figure 2). We next plan to do the computational chemistry calculations on peptides that have not yet been synthesized. Those molecules that show a good calculated Z* will be synthesized and tested in order to confirm the reliability of this potentially predicative method.
1. Hlavka JJ, Abajian H, Noble J: Pharm Res 13(Suppl 9), 143, 1996.
2. Hlavka JJ, Nicolau G, Abajian H, Noble J: Drugs of the Future 22(12):1314-1318, 1997.
3. Hlavka JJ, Abajian H, Nicolau G, Feighner JP: XVIth Int Symp Med Chem [Abstracts]. Bologna, Italy, September 18-22, 2000, p 422.
4. Feighner JP, Ehrensing RH, Kastin AJ, et al: J Affec Disord 61:119-126, 2000.
5. Feighner JP, Noble JF, Sverdlov L, et al: Am J Psych, in press.
aThe computational chemistry studies were done using the QSAR properties in HyperChem and the Pallas system from CompuDrug.
bThe author thanks Dr. Lev Sverdlov, Statistical Department, Innapharma, Inc., Park Ridge, NJ, for these determinations.
Table 2. Regression coefficients of physiochemical descriptors and antidepressant activity
Descriptor Measure of Antidepressant Activity
Z score Percentage of
Method of Analysis PCA PLS PCA PLS
Heat of formation 0.54 0.74 0.54 0.74
Ionization potential -0.86 -1.09 -0.79 -1.09
Dipole moment 0.35 -0.02 0.02 -0.02
Polarizabilityb 0.73 0.87 0.49 0.87
(second derivative term)
Molecular volume 0.70 1.01 0.62 1.01
Octanol/water partition coefficient 2.00 2.00 2.00 2.00
PKa dissociation constant -0.35 -0.53 -0.21 -0.53
(within the physiological range)
Regression coefficients of correlation are expressed in arbitrary units, adjusted to set that of log P equal to 2.0.
PCA = Principal component analysis.
PLS = Partial least squares analysis.
©2000-2013. All Rights Reserved. Veterinary Solutions LLC