(Aryloxy)propanolamines, and Tetrahydropyridylindoles to the 5-HT1A

126

J. Med. Chem. 1996, 39, 126-134

Binding of Arylpiperazines, (Aryloxy)propanolamines, and Tetrahydropyridylindoles to the 5-HT1A Receptor: Contribution of the Molecular Lipophilicity Potential to Three-Dimensional Quantitative Structure-Affinity Relationship Models Patrick Gaillard, Pierre-Alain Carrupt, Bernard Testa,* and Philippe Schambel† Institut de Chimie The´ rapeutique, Section de Pharmacie, Universite´ de Lausanne, BEP, CH-1015 Lausanne, Switzerland Received June 6, 1995X

A set of 280 5-HT1A receptor ligands were selected from available literature data according to predefined criteria and subjected to three-dimensional quantitative structure-affinity relationship analysis using comparative molecular field analysis. No model was obtained for serotonin analogues (19 compounds) and aminotetralins (60 compounds), despite a variety of alignment hypotheses being tried. In contrast, the steric, electrostatic, and lipophilicity fields alone and in combination yielded informative models for arylpiperazines (101 training compounds and 12 test compounds), (aryloxy)propanolamines (30 training compounds and four test compounds), and tetrahydropyridylindoles (54 training compounds) taken separately (models A, B, and C). Arylpiperazines and (aryloxy)propanolamines were then combined successfully to yield reasonably good models for 131 compounds (model D). In a last step, the three chemical classes (185 compounds) were combined, again successfully (model E). This stepwise procedure not only ascertains self-consistency in alignments but it also allows statistical signals (i.e., favorable or unfavorable regions around molecules) to emerge which cannot exist in a single chemical class. The models so obtained reveal a number of interaction sites between ligands and the 5-HT1A receptor, and extend the information gathered from a model based on homology modeling. Introduction Serotonin receptors are involved in a variety of functions and disorders of the central nervous system. Thus, the 5-HT1A subtype receptors are implicated in psychiatric disorders like depression and anxiety. A broad number of compounds of different chemical classes have a high affinity for 5-HT1A receptors and can act as agonists, partial agonists, or antagonists. The most useful and selective 5-HT1A agonist is 8-hydroxy2-(di-n-propylamino)tetralin (8-OH-DPAT, Figure 1), whose tritiated form is commonly used as a radioligand to label 5-HT1A sites. Arylpiperazines such as buspirone (Figure 1) and isaspirone are effective anxiolytic and antidepressant drugs, while (aryloxy)propanolamines such as pindolol (Figure 1) and propranolol can block the pharmacological action of agonists. The 5-HT1A receptor has been cloned (protein product of the genomic clone, G21, transiently expressed in monkey kidney cells) and is constituted by 421 amino acids with the transmembrane moiety arranged in seven helices.1,2 Kuipers et al.3 have proposed a model of the 5-HT1A receptor using bacteriorhodopsin as template. Two binding sites were thus defined, one for agonists like serotonin, aminotetralin and ergoline derivatives (some of which however were antagonists),4 and another for antagonists such as (aryloxy)propanolamines. The binding mode of agonists is postulated to involve an ionic bond between their basic nitrogen and Asp116 on helix III, and two hydrogen bonds linking their 5-OH substituent and indole nitrogen to Thr200 and Ser199, respectively, on helix V. For antagonists, three interactions were identified, namely the same ionic bond, and † Permanent address: Centre de Recherche Pierre Fabre, F-81106 Castres, France. X Abstract published in Advance ACS Abstracts, November 1, 1995.

0022-2623/96/1839-0126$12.00/0

Figure 1. Molecular structure of 8-OH-DPAT, buspirone, and pindolol.

two hydrogen bonds linking the hydroxy group of (aryloxy)propanolamines or the phenolic oxygen of pindolol with Asn386 on helix VII. Comparative molecular field analysis (CoMFA)5,6 is a widely used three-dimensional quantitative structureactivity relationship (3D-QSAR) method to correlate biological activities with three-dimensional structural properties described by a steric and an electrostatic molecular field. This approach, which allows binding modes to an active site to be postulated, is complementary to receptor modeling for characterizing the interactions between a ligand and a receptor. To the best of our knowledge, the CoMFA approach has been applied only four times to 5-HT1A receptors,7-10 but each of these studies focused on a single structural class of ligands. These four models are compared in detail with our results in the Discussion part of the paper. To date, 3D-QSAR investigations based on CoMFA use standard steric and electrostatic molecular fields to model receptor-ligand interactions.11-16 However, these two fields are not always sufficient to describe all recognition forces. Moreover, CoMFA describes only the © 1996 American Chemical Society

MLP in CoMFA of 5-HT1A Receptor Ligands

Journal of Medicinal Chemistry, 1996, Vol. 39, No. 1 127

Figure 3. The template molecule used in the present CoMFA work: 1-(2-methoxyphenyl)-4-[4-(2-phthalimido)butyl]piperazine.

Figure 2. Definition of the five classes of ligands investigated in this study.

enthalpic component in the formation of the ligandreceptor complex.17 As shown by us,18 introducing the molecular lipophilicity potential (MLP) as an additional field may significantly improve the correlative, predictive, and interpretative power of CoMFA in a number of cases. Indeed, the MLP encodes hydrogen bonds and hydrophobic interactions not adequately described by the steric and electrostatic fields, and also includes an entropy component. In this study, we examine the affinity of a large number of structurally heterogeneous ligands to the 5-HT1A receptor, using CoMFA based on steric and electrostatic fields and on the MLP. Models for each class were first derived, which could then be merged into more global models. The success of the analysis affords new insights into the topography and recognition forces characterizing the 5-HT1A receptor. Materials and Methods Selection of Ligands. 5-HT1A affinity values for 330 compounds found in the literature were compiled and critically evaluated. All values retained had been obtained in rat brain preparations with [3H]-8-OH-DPAT as the radioligand. IC50 values were converted to pKi values using the Cheng-Prusoff equation.19 The affinities of reference compounds (usually three or more) were used to compare one series with the others, and the series was retained if the differences were less then 0.3 pKi unit. It should also be noted that no distinction was made between agonists and antagonists since this information was often not available. As a result of this selection, 280 ligands were retained which belong to the 5 different structural classes defined in Figure 2. These ligands were serotonin derivatives (19 compounds in Table I, supporting information), aminotetralins (60 compounds in Table II, supporting information), arylpiperazines (113 compounds in Table III, supporting information), (aryloxy)propanolamines (34 compounds in Table IV, supporting information), and tetrahydropyridylindoles (54 compounds in Table V, supporting information).4,7-9,20-34 The selected compounds may be agonists or antagonists, but the information is seldom available. This problem is considered in the Discussion. In any CoMFA study, it is of importance to assess the predictive power of models by using a test set of compounds.

This was done here by setting aside ca. 10% of the compounds in an arbitrary fashion, except for regularly distributed pKi values. Technical Specifications. All molecular modeling calculations were performed with the SYBYL software versions 5.5 and 6.0435 (Tripos Associates, St. Louis, MO) running on workstations Sun Sparc 2.0 and Silicon Graphics Personal Iris 4D/35, Power Series 4D/320 and Indigo R4000. Energy minimization was performed with the Tripos force field including the electrostatic energy term calculated from Gasteiger and Marsili atomic charges.36 The method of Powell was used for minimizations, convergence being reached when the gradient decrease was smaller than 0.001 kcal/mol per Å. For the CoMFA studies, the QSAR module of SYBYL was used with the three molecular fields (steric, electrostatic and lipophilic). The lipophilicity field was calculated by the MLP.18 The specific parameters used for CoMFA include a grid size of 1.5 Å, atomic charges recalculated with the semiempirical package MOPAC using the Hamiltonian PM3,37 a dielectric function of 1/r and a dielectric constant ) 1. For partial least squares (PLS) analyses, the leave-one-out procedure was chosen for crossvalidation analyses and the component number (N) retained for final PLS analyses corresponded to the first local maximum of the graph q2 ) f(N). The other options were chosen according to published standards.38,39 Special attention must be given in CoMFA to the mixing of several molecular fields. Indeed, a strong correlation between CoMFA signals can occasionally be obtained in some congeneric series of compounds. As a result, the variation for all compounds of each molecular field which explains the variation in biological activity is nearly the same. This artifact may render difficult the interpretation of CoMFA models and precludes their utility in drug design. In the absence of a suitable tool to test this intercorrelation between statistical signals, we used a stepwise strategy. First, CoMFA models were calculated for a given set of ligands (designated by a capital letter A-E) with one (models 1, 2, and 3), two (models 4, 5, and 6), and then three fields (models 7). At each step, the statistics were evaluated and compared with previous steps. It is our experience that the most informative statistics in this stepwise evaluation are q2 and the number of components (N). In particular, signal correlation was suspected when all models for a given set had simultaneously comparable statistical and graphical results. Alignment of Ligands. Ligand alignments were performed using the optimized geometries modified by manual geometrical fitting. Although several alignment modes were tested, the 5-HT1A pharmacophore proposed in the literature22,23 was retained for the final analyses. Thus, all molecules were aligned on the template (1-(2-methoxyphenyl)-4[4-(2-phthalimido)butyl]piperazine) (compound 95 in Table III of the supporting information, pKi value of 9.22). The conformation of the template was as shown in Figure 3,26 i.e. (a) piperazine ring in a chair conformation and perpendicular to the aromatic moiety (free electron lone pair of the basic nitrogen parallel to the plane of the aromatic moiety), (b) distance of 5.7 Å between the basic nitrogen and the centroid of the aromatic moiety, in agreement with the published pharmacophore (where this distance is in the range of 5.25.6 Å, with the amino group 0.2-1.6 Å above the plane of the aromatic moiety),22,23 and (c) 4-phthalimidobutyl moiety in an extended form (linear alkyl chain). The alignment of ligands in all five classes was based on their aromatic moiety and basic nitrogen: (a) the aromatic moieties were aligned by their centroid and the normal to their plane and (b) the free electron lone pairs of the basic nitrogen were pointed in the same direction. Figure 4 shows an

128

Journal of Medicinal Chemistry, 1996, Vol. 39, No. 1

Gaillard et al. Table 7. CoMFA Models A on Arylpiperazines (101 Compounds)

Figure 4. Superposition of cyanopindolol and the template molecule. Table 6. Color Code of Graphical CoMFA Results field or component of field steric field electrostatic field (positive charge) electrostatic field (negative charge) lipophilic field (lipophilic component) lipophilic field (hydrophilic component)

increase affinity

decrease affinity

green white

red magenta

magenta

white

yellow

cyan

cyan

yellow

model

field type

q2 a

Nb

r2 c

sd

A1 A2 A3 A4 A5 A6 A7

steg eleg lipg ste and ele ste and lip ele and lip ste, ele, and lip

0.59 0.59 0.57 0.64 0.63 0.59 0.65

3 3 7 4 5 5 5

0.78 0.79 0.95 0.88 0.92 0.93 0.94

0.60 0.58 0.29 0.43 0.36 0.34 0.32

Fe

stef

elef

lipf

111 100 122 100 243 100 183 42 58 214 46 54 241 53 47 281 30 36 33

a Cross-validation correlation coefficient. b Number of components used in final PLS analyses corresponded to the first maximum of the fonction q2 ) f(N) in cross-validation analyses. c Correlation coefficient of the final PLS analysis. d Standard error of estimate, measure of the unexplained uncertainty. e F ratio, the higher the F ratio, the better the PLS analysis. f Relative contribution of the steric (ste), electrostatic (ele), and lipophilic (lip) field in the final PLS analysis. g Molecular field(s) used in CoMFA: ste ) steric field, ele ) electrostatic field, lip ) lipophilic field (calculated by MLP).

example of the chosen alignment, i.e. the superposition of cyanopindolol on the template molecule. Presentation of Results. The graphical results in Figures 5-9 represent the most relevant regions of space where the variations of the statistical field are the largest. The color code used to characterize the signals of each field is described in Table 6. Due to the duality of the electrostatic field, a white zone can mean a favorable influence of some electron deficiency or an unfavorable influence of some high electron density. The interpretation of the lipophilicity field is similarly ambiguous. Indeed, lipophilicity encodes two major structural contributions,40 namely a bulk term reflecting hydrophobic and dispersive forces and a polar term reflecting more directional electrostatic interactions and hydrogen bonds. The favorable influence of the hydrophobic and hydrophilic components is described by yellow or cyan colors, respectively. By examining the moieties and functionalities involved, a chemist should be able to rely on common sense to remove the ambiguity in interpreting the influence of the electrostatic and lipophilicity fields.

Results Serotonins and Aminotetralins. No PLS model with q2 > 0.4 was found for the serotonin and aminotetralin derivatives. The number of serotonin derivatives was too limited, and the structural variability in this series was too small to obtain a meaningful CoMFA model. For the aminotetralins, the best correlation was found with the steric field alone and one component (q2 ) 0.36, n ) 60, N ) 1, r2 ) 0.44, s ) 0.77, F ) 46). In this series, the structural variability was mainly due to the nitrogen substituents, but in the absence of other pharmacophoric elements the alignment was difficult to choose. Different alignment rules were tested but no coherent CoMFA model emerged. Arylpiperazines. The statistical results for this series are summarized in Table 7, the graphical results of the PLS models A1, A2, A3, and A7 being presented in Figure 5. All the PLS models were statistically significant. The salient results to emerge from Figure 5 are (a) a favorable steric region close to the aromatic ring, (b) a sterically forbidden region close to the basic nitrogen, (c) a favorable negative charge close to the ortho position of the aromatic ring, and (d) opposite lipophilic and electrostatic effects on the nitrogen substituent. (Aryloxy)propanolamines. The statistical results for the aryloxypropanolamines are summarized in Table

Figure 5. Graphical results of PLS models A1, A2, A3, and A7 for arylpiperazines with compound 101 displayed. Table 8. CoMFA Models B on (Aryloxy)propanolamines (30 Compounds) model

field type

q2

N

B1 B2 B3 B4 B5 B6 B7

ste ele lip ste and ele ste and lip ele and lip ste, ele, and lip

0.46 0.53 0.20 0.59 0.41 0.45 0.50

1 3 2 2 2 3 2

r2

s

F

ste

ele lip

0.66 0.74 55 100 0.93 0.36 108 100 0.87 0.78 0.92 0.82

0.47 0.61 0.38 0.55

90 48 98 63

44 51 32

56 49 55 45 37 31

8, and the graphical results of the PLS models B1, B2, and B4 are presented in Figure 6. The best PLS models were obtained without the lipophilicity field, implying that the variability in this class is due only to steric and



Figure 6. Graphical results of PLS models B1, B2, and B4 for (aryloxy)propanolamines with compound 131 displayed. Table 9. CoMFA Models C on Tetrahydropyridylindoles (54 Compounds) model C1 C2 C3 C4 C5 C6 C7

field type

q2

ste 0.41 ele 0.05 lip < 0.0 ste and ele 0.43 ste and lip 0.41 ele and lip 0.13 ste, ele, and lip 0.40

N

r2

s

F

ste

4 1

0.78

0.33

43

100

4 2 3 2

0.84 0.69

ele

lip

Figure 7. Graphical results of PLS models C1 and C4 for tetrahydropyridylindoles with compound 157 displayed. Table 10. CoMFA Models D on Arylpiperazines and (Aryloxy)propanolamines (131 Compounds)

0.73

0.28 0.39 0.36

65 56 70

68 73 49

32 27 31

20

electrostatic factors. The increase in affinity is controlled mainly by (a) favorable steric and electron density influences close to the ortho and meta positions of the aromatic ring, (b) steric hindrance and a favorable electron deficiency close to the para position of the aromatic ring, and (c) by a favorable electron deficiency close to the basic nitrogen. Tetrahydropyridylindoles. Table 9 shows the statistical results for the tetrahydropyridylindoles, with the graphical results of models C1 and C4 being presented in Figure 7. In this class of ligands like in the (aryloxy)propanolamines, the variability is explained mainly by steric properties, with a modest electrostatic contribution. A number of sterically unfavorable regions can be seen in models C1 and C4; they are located around the indole ring, above and below it, indicating forbidden positions. There is however a favorable influence of bulky and H-bond-donating 5-substituents on the indole ring. Models for Arylpiperazines and (Aryloxy)propanolamines Taken Together (131 Ligands). Table 10 shows the statistical results of the two classes taken together. Each molecular field individually yields acceptable models (D1, D2, and D3), and their combinations (D4, D5, D6, and D7) are also statistically valid.

model

field type

q2

N

r2

s

D1 D2 D3 D4 D5 D6 D7


0.61 0.60 0.53 0.66 0.63 0.61 0.66

3 3 3 3 4 3 3

0.80 0.79 0.77 0.83 0.87 0.82 0.85

0.64 0.64 0.68 0.58 0.52 0.59 0.55

F

ste

ele

lip

164 100 161 100 139 100 205 42 58 204 44 56 198 52 48 234 29 36 35

In fact, these models are better than those of the aryloxypropanolamines and comparable to those for the arylpiperazines. Figure 8 presents the graphical results of the PLS models D1, D2, D3, and D7. Model D3 for example reveals regions where hydrophobic (yellow) and hydrophilic (cyan) substituents increase affinity. Of particular interest is the fact that in the combined model D7, an additional signal due to the lipophilic field is apparent, namely the favorable influence of a hydrophilic substituent (cyan) in the region of the hydroxyl group of aryloxypropanolamines. General Model for Arylpiperazines, (Aryloxy)propanolamines, and Tetrahydropyridylindoles Taken Together (185 Ligands). The statistical results for the three classes taken together are summarized in Table 11. Clearly acceptable models are obtained for each molecular individually (models E1, E2, and E3), with their pairwise combinations yielding some improvement (models E4, E5, and E6). The best model combines all the three fields (E7) and is statistically fairly good. This model is comparable to model D7. The

130


Gaillard et al.

Figure 8. Graphical results of PLS models D1, D2, D3, and D7 for arylpiperazines and (aryloxy)propanolamines combined, with compound 101 displayed. Table 11. CoMFA Models E on Arylpiperazines, (Aryloxy)propanolamines, and Tetrahydropyridylindoles (185 Compounds) model

field type

q2

N

r2

s

E1 E2 E3 E4 E5 E6 E7


0.58 0.56 0.45 0.61 0.61 0.60 0.64

5 4 5 4 4 5 4

0.82 0.78 0.73 0.81 0.79 0.83 0.82

0.53 0.59 0.65 0.54 0.57 0.51 0.53

F

ste

ele

lip

160 100 155 100 95 100 193 42 58 167 50 50 179 58 42 205 30 39 31

graphical results of models E1, E2, E3 and E7 are presented in Figure 9. They are comparable to those for the arylpiperazines and (aryloxy)propanolamines, meaning that the tetrahydropyridylindoles can be merged successfully with the other two classes. However, the graphical representation of model E7 shows an additional signal due to the electrostatic field in the region of the 1-position of the indole ring of tetrahydropyridylindoles. The signal (a white region) indicates that an electron deficit in this region increases the 5-HT1A receptor affinity. Discussion Our results clearly demonstrate that a single CoMFA model can be built from all the arylpiperazines, (aryloxy)propanolamines, and tetrahydropyridylindoles examined in this study. Moreover the statistical improvement of model E7 over models E1-E6 (better q2 and number of components) illustrates the interest of adding the lipophilicity field to the standard steric and electrostatic CoMFA fields. The results of models E1-E7

Figure 9. Graphical results of PLS models E1, E2, E3, and E7 for arylpiperazines, (aryloxy)propanolamines, and tetrahydropyridylindoles combined, with compounds 95, 130, and 174 displayed.

further suggest that the alignment chosen is compatible with a common binding site and a common binding mode for arylpiperazines, (aryloxy)propanolamines, and tetrahydropyridylindoles derivatives. This result seems intriguing considering that some ligands may be agonists, other antagonists. It is well-known that the mode of binding of agonists and antagonists may differ at some receptors either because their relative positioning differs or because partly different anchoring sites may be involved. In the series investigated, information on pharmacological activity is seldom available to categorize the ligands as agonists and antagonists, and we can only note that statistically valid models indeed were obtained. Perhaps separate and different alignments for agonists and antagonists would have led to models of even better statistical quality, or the models generated here do not contain statistical signals in the receptor regions which discriminate between agonists and antagonists. Further studies are needed to solve this problem. Another open question is why no statistical model was generated for the two other series investigated, namely the serotonin and aminotetralin derivatives. This may be due to the limited structural variability in the


ligands, which does not allow meaningful statistical signals to be obtained. In addition, we may have failed to identify a proper alignment for the aminotetralins. The resolution of this problem must await larger and more diverse series of serotonin and aminotretralin ligands. Comparing the graphical results in Figures 5-9 is doubly informative. First, the graphical models reveal groups in the ligand molecules which interact with the 5-HT1A receptor. And second, the blending of different chemical classes of ligand into a single model is shown to uncover additional sites of interactions not apparent in separate models. Aromatic Region. The effects of aromatic substitutions on 5-HT1A receptor affinity is well-explored in the three series of ligands, leading to three important conclusions. First, substitution in the ortho position by a bulky group in the plane of the aromatic ring is sterically favorable to affinity (green region). However, the limited size of this green region in all models and the additional red regions (sterically unfavorable interactions) in models C1 and C4 indicates that ortho substituents cannot be larger than an optimal size. Second, meta and para substitutions are unfavorable for affinity, but their influence is lower than that of ortho substituents, as suggested by the disappearance of the corresponding signal when going from model B4 to model E7. Third, an electrostatic signal near the ortho position is common to the three ligand classes, indicating that affinity is increased by the presence of an electron excess and suggesting a strong electrostatic and/or hydrogen bond interaction in the ligand-receptor complex near the ortho position of the aromatic ring. The steric and electrostatic influences of substituents are compatible with the high affinity of o-methoxysubstituted ligands. Substituent on the Basic Nitrogen. A steric hindrance is noted in all PLS models close to the basic nitrogen. This translates the fact that N-monosubstitution is the pattern most favorable for affinity and that the accessibility of the nitrogen must be optimal. In other words, an interaction between the basic nitrogen and the receptor is evidenced as described by previous workers. The nitrogen substituent exerts steric, electrostatic, and lipophilic influences, but clear information about these influences cannot be obtained from our models. A major source of variability comes from the great conformational flexibility of the N-substituted moiety, allowing a large variety of possible alignments. To explore this region better, several alignments were tested, none of which improved markedly on the others. This suggests a relatively large pocket in the receptor allowing a considerable flexibility of the N-substituent. Additional Anchor Points. The appearance in model D7 of a well-defined signal for hydrophilicity close to the secondary alcoholic group of (aryloxy)propanolamines means that there is a specific polar interaction of this group with the receptor. This polar region is not apparent in CoMFA models of (aryloxy)propanolamines taken alone for the obvious reason that there is almost no variation at and around this group within the class of (aryloxy)propanolamines. But when (aryloxy)propanolamines are taken together with arylpiperazines, a variation (presence or absence of the hydroxy group) is created which generates a statistical signal. In model


Figure 10. Schematic results of the model E7: region 1, high electron density favorable; region 2, steric bulk favorable; region 3, steric hindrance; region 4, electrostatic and lipophilic influences; region 5, polar interaction; region 6, electron deficiency favorable.

E7, this polar signal combines with the more diffuse polar signal of model A7 to give a slightly shifted signal. Another new region becomes apparent when different chemical series are mixed, namely a region of electrostatic influence near the 1-position of the indole ring of tetrahydropyridylindoles. This means that a specific interaction occurs between the indole nitrogen of tetrahydropyridylindoles and the 5-HT1A receptor. In this case again, the signal does not exist when the ligands of this class are analyzed for themselves (model C4). The presence of these two additional anchor points underlines the interest in the lipophilicity field and the increase in information obtained when different classes of ligands are mixed. When indeed a given molecular variation is not large enough within a series of compounds to induce a statistical signal, this cannot be taken to imply that the molecular feature in question is not relevant for activity. 3D Pharmacophore Model of the 5-HT1A Receptor and Its Comparison with Previous Models. The results of our CoMFA models allow a detailed picture of the pharmacophoric elements around the aromatic moiety and the basic nitrogen and can be summarized as shown in Figure 10. As discussed below, there are no major discrepancies between model E7 and the four published CoMFA models,7-10 but the main advantage of model E7 is to be more informative by taking three different chemical classes into account. Agarwal et al.7 reported a CoMFA model (n ) 45, q2 ) 0.45, N ) 4, r2 ) 0.85, s ) 0.29, F ) 58) containing the steric (relative contribution 87%) and electrostatic (relative contribution 13%) fields of 45 tetrahydropyridylindoles where the main structural variability was at the 5-position of the indole ring. The main graphical signals corresponding to favorable influences on 5-HT1A affinity were (a) a steric region out of the plane of the indole ring and one in the plane (but not in close proximity of the indole nitrogen) and (b) a high electron density at the 5-position of the indole ring and an electron deficiency between the 4- and 5-positions of the indole ring. These results compare well with the signals of model C4, which is not astonishing considering that the 45 compounds in the study by Agarwal et al.7 were part of the 54 tetrahydropyridylindoles examined here. But although the CoMFA model of Agarwal et al.7 and model C4 both indicate an electrostatic contribution (13% and 32%, respectively), a careful examination of other models (i.e. C2 and C6) does not support the significance of electrostatic variations in this series of

132


compounds. The relative high contribution of a nonsignificant field could be an artifact of the PLS treatment easily resolved by the stepwise approach proposed in this paper, i.e. models with different combinations of fields. The difference between the relative contributions in model C4 and in literature models takes its origin in a different manner of calculating the electrostatic field.41 For a series of 50 arylpiperazines, El-Bermawy et al.8 found a CoMFA model including steric and electrostatic fields (n ) 50, q2 ) 0.68, N ) 3, r2 ) 0.89, s ) 0.3, relative contributions 47% and 53%, respectively). Only partial comparison with our models is possible since the graphical results were incompletely reported and for the steric field only. The allowed steric region close to the ortho position of the aromatic moiety is well reproduced in model A1. However, models A7 and E7 do not explicit a sterically allowed region near the γ carbon of arylpiperazines but attribute the variation of affinity near the γ carbon to the lipophilicity field. Here, variations in the lipophilicity field are due only to its hydrophobic component. Langlois et al.9 reported a CoMFA model for 17 (aryloxy)propanolamines, five of which were in fact inactive (pKi < 5.00). The statistical results were n ) 17, q2 ) ?, N ) 5, r2 ) 0.99, s ) 0.13, F ) 266, with the steric and electrostatic fields contributing 92% and 8%, respectively. This model is statistically poor since it needs five principal components for only 17 compounds. The high ratio between the number of principal components and the number of compounds leads to overfitting as depicted by the very high unrealistic correlation coefficient (r2 ) 0.99). The very weak contribution of the electrostatic field cannot be considered as a reliable result, while the graphical results of the steric field are comparable to those in Figure 6. For a series of 47 arylpiperazines, van Steen et al.10 found a CoMFA model with the steric field (n ) 47, q2 ) 0.86, N ) 3, r2 ) 0.95). The graphical results of the CoMFA model are localized around the N4-alkyl-substituent and are as follows: a prohibited steric region at 3.5 Å and a favorable steric region at 7.3 Å. This means that small N4-alkyl substituents decrease 5-HT1A affinity whereas large N4-alkyl substituents increase it. The CoMFA model of van Steen et al.10 and model E7 agree on the prohibited steric regions since the two red signals of model E7 are similarly located. Due to different selections of compounds, model E7 does not reveal a sterically favorable region around the N4-alkyl substituent, but the loss of significant CoMFA signals in this region may imply a large and fuzzy pocket in the receptor. Validations of Model E7. Whenever possible, the predictive power of any QSAR model should be assessed by its capacity to predict the activities of a test set as much as by its cross-validated correlation coefficient q2. Here, a test set of 16 arylpiperazines and (aryloxy)propanolamines was set aside at the beginning of the study. Figure 11 shows the experimental versus predicted 5-HT1A affinities using model E7. The linear regression has a correlation coefficient of r2 ) 0.72 in agreement with the q2 value of the model (q2 ) 0.64). The most deviant compounds are 134, 141, and 144 (too high prediction by >0.5) and 135, 143, 145, and 147 (too low prediction by

(Aryloxy)propanolamines, and Tetrahydropyridylindoles to the 5-HT1A

Recommend Documents