Dibenzazecine compounds with a novel dopamine/5HT2A receptor profile and 3D-QSAR analysis

Background Antipsychotics are divided into typical and atypical compounds based on clinical efficacy and side effects. The purpose of this study was to characterize in vitro a series of novel azecine-type compounds at human dopamine D1-D5 and 5HT2A receptors and to assign them to different classes according to their dopamine/5HT2A receptor profile. Results Regardless of using affinity data (pKi values at D1-D5 and 5HT2A) or selectivity data (15 log (Ki ratios)), principal component analysis with azecine-type compounds, haloperidol, and clozapine revealed three groups of dopamine/5HT2A ligands: 1) haloperidol; 2) clozapine plus four azecine-type compounds; 3) two hydroxylated dibenzazecines. Reducing the number of Ki ratios used for principal component analysis from 15 to two (the D1/D2 and D2/5HT2A Ki ratios) obtained the same three groups of compounds. The most potent dibenzazecine clustering in the same group as clozapine was the non-hydroxylated LE410 which shows a slightly different D2-like receptor profile (D2L > D3 > D4.4) than clozapine (D4.4 > D2L > D3). The monohydroxylated dibenzacezine LE404 clusters in a separate group from clozapine/LE410 and from haloperidol and shows increased D1 selectivity. Conclusion In conclusion, two compounds with a novel dopamine/5HT2A receptor profile, LE404 and LE410, with some differences in their respective D1/D2 receptor affinities including a validated pharmacophore-based 3D-QSAR model for D1 antagonists are presented.


Background
Dopamine is an important neurotransmitter in the mammalian CNS which has influence on physiological, behavioural and neuroendocrine functions, mediated through receptors on the cell surface. Five different dopamine receptor subtypes have been cloned and characterized. They belong to the super-family of G protein-coupled receptors (GPCR) and can be divided into two subfamilies, D 1 -like (D 1 , D 5 ) and D 2 -like (D 2 , D 3 , D 4 ) receptors, according to their sequence homologies, biochemical properties, and pharmacologic profiles [1]. D 1 -like receptor stimulation activates adenylyl cyclase (AC) via coupling to stimulatory G protein Gα s /Gα olf subunits leading to an increase in intracellular cAMP concen-trations. In contrast, D 2 -like receptors are Gα i /Gα o linked and inhibit AC activity [2]. Dopamine receptors are clinically important drug targets for the treatment of disorders such as Parkinson's disease and schizophrenia [3]. Blockade of dopamine D 2 receptors is the main feature of antipsychotic action. Typical antipsychotics like the first generation D 2 receptor antagonists haloperidol or chlorpromazine can cause therapy-limiting extrapyramidalmotor side effects (EPS). Second generation (atypical) antipsychotics are serotonin/dopamine antagonists with no or low EPS at doses showing antipsychotic activity and have significantly greater affinity for 5HT 2A than for D 2 receptors [4]. This serotonin-dopamine ratio could contribute to atypicality [5][6][7] but further investigations are needed to define the precise mechanism of atypical antipsychotics. However, antipsychotic activity is not only the result of D 2 and 5HT 2A receptor blockade but an inhibitory/modulating effect on various dopamine and serotonin (D 1 , D 2 , D 3 , D 4 , 5HT 1A , 5HT 1D , 5HT 2A , 5HT 2C ) and further receptors [8]. Within the heterogeneous group of atypical antipsychotics, only clozapine exhibits effects against treatment-resistant schizophrenia [9]. Responsible for this net effect among atypical antipsychotics may be the moderate affinity of clozapine at various receptor subtypes, especially at D 1 -receptors. A dysfunction in D 1receptor modulation in the prefrontal cortex contributes to the negative symptoms and cognitive deficits observed in schizophrenia. However, selective D 1 antagonism alone has not turned out as an effective antipsychotic principle [9,10].
LE300, an indolobenzacezine (figure 1) has previously been characterized [11] and shows a binding profile similar to that of clozapine, however with a greater affinity for D 1 -than D 2 -like receptors. A series of LE300-derived compounds was recently synthesized and screened at dopamine D 1 , D 2L , and D 5 receptors by a previously published functional calcium assay [12,13]. The aim of the current study was to investigate the comprehensive binding and functional receptor profile of the most active of the dibenzazecine derivatives of LE300 (LE400, LE401, LE403, LE404, LE410, and LE420, figure 1) at all human dopamine and 5HT 2A receptors, to test whether data from the calcium assay initially used for screening of LE300derived compounds [13] correlate with other assays measuring functional activation of GPCRs (cAMP, [ 35 S]-GTPγS), and to establish a 3D-QSAR pharmacophore model of these ligands. Heterologous competition binding experiments were carried out at recombinantly expressed human dopamine and 5HT 2A receptors, and obtained data were compared with functional data from intracellular [cAMP] and [Ca 2+ ] measurements and [ 35 S]-GTPγS-binding. Indeed, dibenzazecine compounds with a previously not available receptor profile (increased antagonist activity at D 1 -like and 5HT 2A receptors) were found.
3D-QSAR studies were performed resulting in QSAR models allowing further rational ligand design at a molecular level.

Receptor expression and characterization
Homologous radioligand competition binding experiments were performed to determine the receptor expression levels (B max ) and binding affinities (K d ) of the used radioligands. Average B max and K d values for each receptor are shown in table 1. All K d values except for 5HT 2A receptors were 3-6-fold higher than those found in the literature (table 1, [14][15][16][17]). This effect could be attributed to the use of isotonic Krebs-HEPES-buffer pH 7.4 in this study instead of the widely used TRIS-HCl buffer pH 7.4 in the literature. Figure 2 shows as an example the bufferdependent inhibition by LE300 of [ 3 H]SCH23390 binding to D 1 receptor membranes. Using Krebs-HEPES instead of TRIS-HCl buffer yielded ~4-fold higher K i values of LE300 (figure 2) but allowed a better comparison of functional and binding data. A buffer-dependent change of affinity was also observed with the test compounds. However, the K d ratios among the receptor subtypes using Krebs-HEPES buffer were equal to literature data using TRIS-HCl (not shown).
All compounds showed similar affinities at hD1 and hD5 receptors. The mono-hydroxylated LE404 turned out as the most potent compound at hD1/hD5 receptors with pKi values of 8.47 and 8.53, respectively, followed by the bis-hydroxylated LE403 which is 3-10-fold less potent than LE404. Replacement of the hydroxy-by methoxysubstituents resulting in LE400 dramatically decreased the affinity at all tested receptors. An increase of the size of the nitrogen substituent (allyl group of LE401) further decreased the affinity at all tested receptors. Except LE400 and LE401, all other compounds possessed up to 33fold (LE403, LE404) higher affinities for D1-like than for D2-like receptors (table 2). Among D2-like receptors, all compounds -except LE404 -showed the highest affinity at hD2L and lower affinities at hD3 and hD4.4 receptors similar to the profile of haloperidol at D2-like receptors. However, different to haloperidol which shows a strong D2 over D1 selectivity, LE compounds (except LE400 and LE401) show selectivity for D1 over D2. Removal of the hydroxy-group of LE404 yielding LE410 resulted in a dramatic loss of D1 over D2 selectivity, and left LE410 as the most potent compound at hD2L and hD3 receptors with pKi values of 7.54 and 6.86, respectively. Bioisosteric replacement of one benzene residue in LE410 by thiophene gave LE420 showing a similar receptor profile as LE410 but with reduced affinity at all tested receptors. Structural formulas of the indolobenzazecine LE300, SCH23390, and a series of ten derived compounds Figure 1 Structural formulas of the indolobenzazecine LE300, SCH23390, and a series of ten derived compounds. In contrast to clozapine which appeared ~ equipotent at D1/D2 receptors in all of our test systems, LE404 shows 25fold selectivity for D1 over D2. LE404 displayed higher affinities than LE300 at all dopamine receptors except hD2L where both compounds are ~ equipotent. All compounds except LE401 showed the highest affinities among all tested receptors at 5HT2A. The most potent compound at 5HT2A was LE300 with an affinity in the subnanomolar range followed by LE404 in the low nanomolar range. LE300, LE400, LE403, LE404, LE410, and LE420 achieved Ki-D2i-5HT2A /K selectivity ratios > 7.

Functional studies (cAMP, Ca 2+ and [ 35 S]-GTPγS binding) at hD 1 and hD 2L receptors
For functional studies, hD 1 and hD 2L receptors were chosen as characteristic representatives of each of the two dopamine receptor subtype groups allowing a comparison of functional and binding data.  Buffer-dependent differences in hD 1 receptor potencies of LE300 in competition binding Figure 2 Buffer-dependent differences in hD 1 receptor potencies of LE300 in competition binding.

Statistical comparison of functional and binding data at D 1 and D 2L receptors
The multiple intercorrelation and thus the equality of the results obtained by binding and the three functional assays at D 1 and D 2L receptors, respectively, was determined by principal component analysis (PCA). Results of Table 3: Characterization of AHAD11, B157, LERU301, and SH3 at hD 1 receptors used for 3D-QSAR analysis.

Nature of antagonism of LE compounds at D 1 and D 2 receptors
Next, the nature of antagonism of LE compounds at D 1 and D 2L receptors was tested by Clark analysis [20]. Since LE404 was the most potent compound at D 1 and LE410 the most potent at D 2L receptors (binding, table 2), LE404 and LE410 were chosen as representatives to undergo functional analysis for competitive antagonism. In the presence of increasing concentrations of LE404 and LE410, parallel rightward shifts of the agonist concentration-effect curves in the Ca 2+ assay were observed without loss of maximum efficacy at hD 1 and hD 2L receptors (data not shown). The rightward shift of the concentrationeffect curves of the agonist was analyzed with non-linear regression analysis according to Lew and Angus [20]. Data were fitted to equations (1) and (2)  pK b values were calculated as: hD 1 : pK b LE404 = 8.09 ± 0.15; pK b LE410 = 7.69 ± 0.13; hD 2L : pK b LE404 = 7.61 ± 0.10; pK b LE410 = 8.05 ± 0.11. pK b values of LE404 and LE410 derived from non-linear Clark analysis show no significant difference to those derived from Schild analysis [21] (data not shown). Both functional analyses (Schild, Clark) give thus evidence for a competitive antagonistic behaviour of LE404 and LE410 at D 1 and D 2L receptors.

Statistical analysis of binding affinities and selectivities at dopamine and 5HT 2A receptors
In order to perform a statistically valid test for the discovery of ligands with differing affinity profiles at dopamine D 1 -D 5 and 5HT 2A receptors among the examined compounds, multiple intercorrelations of binding affinity values (pK i ,     The resulting log (K i ratio) data matrix contains selectivity information for each of the compounds. Results of this second ("selectivity") PCA were basically identical to results from the first ("affinity") PCA (table  7). The first extracted PC explained 74.8% of the total variance among the eight variables (log (K i ratio) values), and the second PC extracted 15.4% of the total variance. The second ("selectivity") PCA discovered the same three subgroups of dopamine/5HT 2A ligands as did the first PCA: 1) clozapine, LE300, LE400, LE410, and LE420 with factor loadings contributing to the first PC of > 0.780 (table 7); 2) haloperidol in the second PC with a factor loading of -0.901; 3) LE403 and LE404 in the second PC with opposite direction to haloperidol (factor loadings 0.933 and 0.893). Thus, regardless of using affinity information (pK i ) or selectivity information (log (K i ratio)) for PCA, the same three subgroups of dopamine/5HT 2A ligands were discriminated. The agreeing results from both PCA's underline that the statistical analysis of binding affinities and selectivities at dopamine and 5HT 2A receptors did not create chance correlations.

3D-QSAR (CoMFA/CoMSIA studies)
Since the main feature of the LE compounds is their D 1 selectivity, a 3D-QSAR pharmacophore model for the D 1 receptor was establish using the 12 compounds shown in figure 1 and their D 1 -pK i values from table 2 and 3. For a successful CoMFA/CoMSIA study, it is crucial to find an appropriate alignment of the examined compounds. It is not necessary that all compounds possess the bioactive conformation but it is useful that the compounds adopt a relative conformation and position to each other as they would bind to the receptor. The D 1 /D 5 selective antagonist (-)-2b-SCH39166 (ecopipam) was taken as a pharmacophore template. (-)-2b-SCH39166 is a benzonaphthazepine, a rigid analogue of SCH23390, thus limiting the number of possible conformations (figure 7) [24]. Unfortunately, (-)-2b-SCH39166 was not available to us for testing, and was thus not used for the final QSAR-analysis. However, due to its rigid nature, it was helpful to find a good starting point for selecting conformations and alignments of the 12 compounds from figure 1. Essential pharmacophore features of (-)-2b-SCH39166 are the two aromatic rings and the basic nitrogen (hydrogen acceptor) while the hydroxyl group served as an optional H-donor/ acceptor feature ( figure 7). Results of the alignment of the final models of the LE compounds are shown in figure 8. The aromatic residues and basic nitrogen atoms remain the main pharmacophore features. Crossvalidation results (leave-one-out) for the final models for CoMFA and CoM-SIA both using steric and electrostatic fields are displayed in table 8, and show crossvalidation parameters q 2 of 0.82 for CoMFA and 0.88 for CoMSIA. To prove that these models were not a result of a chance correlation, a stability test was performed using the random groups PLS  method ("leave-many-out"). The test showed a high stability of the models presented in figure 8 with a mean q 2 of 0.76 (SD 0.10) for the combined steric and electrostatic field in CoMFA and a mean q 2 of 0.81 (SD 0.12) in CoM-SIA. The distribution of the q 2 values for this validation is shown in figure 9.

Discussion
Among a group of new azecine compounds, this study has revealed two dibenzacezines (LE404 and LE410) with potent activity at dopamine and 5HT 2A receptors displaying a novel receptor profile at D 1 -D 5  ] messenger generation). A comparison of pK i values of one compound in the four different assays thus leads to differences, e.g., K i ratios of haloperidol at D 1 /D 2L receptors are ~1200 in cAMP, 2500 in Ca 2+ , and ~100 in [ 35 S]-GTPγS and binding studies but the rank order of potency remains almost unchanged (tables 2 and 5). Mottola et al. [25] have introduced the term "functional selectivity" to propose that depending on the experimental (buffer, equilibrium) and cellular conditions regarding receptor and G protein expression, a mixture of agonist/partial agonist and/or antagonist actions are likely. The ~2-fold difference in D 1 and D 2L receptor expression in this study (table 1) may thus contribute to differences in pK i values observed in functional and binding studies. The same reasons may serve as an explanation for differences in the K d values of SCH23390 and spiperone in this study and in the literature (table 1) and for the ~1.4-5.5-fold differences in the affinity of LE300 in this and a previous study [11]. Further, affinities in this study were tested at recombinantly expressed receptors in HEK293 cell membranes in Krebs-HEPES-buffer whereas the previous study used CHO cell membranes in a Tris-Mg 2+ -buffer [11]. As was shown in figure 2, different buffers can result in significantly different affinity of a ligand.
LE404 and LE410 are competitive antagonists as was shown by Clark analysis (figure 6). pK b values of LE404 and LE410 derived from these functional analyses are in accordance with pK i values derived from inhibition curves (tables 5 and 2). Statistical analysis (PCA) of binding affinity data (pK i values, table 2) and binding selectivity data [log (K i ratio) values, calculated from table 2] Functional analysis of the antagonist effect of LE404 at hD 1 and hD 2L receptors Figure 6 Functional analysis of the antagonist effect of LE404 at hD 1 and hD 2L receptors. The analysis was carried out by measuring the attenuation by LE404 of the agonistinduced increase in intracellular [Ca 2+ ] in HEK293 cells recombinantly expressing hD 1 and hD 2L receptors, respectively. Dashed lines show 95% confidence intervals. SKF38393 was used as agonist at hD 1 receptors, quinpirole at hD 2L . All slopes were not significantly different from unity. Presented data are means ± SEM from at least three independent experiments each with at least threereplicates. A, B. Clark analysis of LE404 at hD 1 (A) and hD 2L receptors (B). Inserts show Clark plots. resulted in three groups of ligands: first: haloperidol; second: clozapine, LE300, LE400, LE410, and LE420; andinterestingly -a third group: containing LE403 and LE404 (table 7). The most potent compounds in group 2 and group 3 are LE410 and LE404. LE410 has a similar affinity profile as clozapine except the lower potency of LE410 at the hD 4.4 receptor (table 2). In contrast, LE404 has a 25fold selectivity for D 1 over D 2L receptors and thus a novel dopamine/5HT 2A receptor profile. Interestingly, if instead of all K i ratio values which have been used for the PCA in table 7 only the D 1 /D 2L and D 2L /5HT 2A ratios of all compounds were used for clustering, the same three groups were found: 1) haloperidol, 2) clozapine, LE400, LE410, LE420, LE300, and 3) LE403 and LE404 (table 9). Thus, instead of six receptors and 15 K i ratios, a reduction to three receptors (D 1 , D 2L , 5HT 2A ) and two K i ratios is sufficient to obtain the same clustering of compounds.
Meltzer et al. suggested the use of D 1 /D 2L and D 2L /5HT 2A ratios to allow a clustering of antipsychotics into typical and atypical compounds [5][6][7]. However, instead of Meltzer et al. who calculated pK i ratio values which are imprecise in defining selectivity (same selectivity may result in different pK i ratios depending on the potency), K i ratios (table 9) or log (K i ratio) values (for PCA in table 7) were calculated in this study. K i ratios recalculated from data of Meltzer et al. [5] and K i ratios from this study were no more different than 3-fold (table 9). LE300, LE403, LE404, LE410, and LE420 achieved K i-D2 /K i-5HT2A selectivity ratios > 7 which may suggest an atypical behaviour of these compounds according to Meltzer et al. [5]. However, so far there are no in vivo behavioural studies underlying an antipsychotic effect of the LE compounds. The third group of ligands, LE403 and LE404, differ from LE410 by a 15-20-fold increase in D 1 selectivity (table 9). RMI-81582 has very similar D 1 /D 2 and D 2 /5HT 2A K i ratios as LE403 and LE404 (table 9) and was shown to exert antipsychotic effects [26]. A further increase in D 1 selectivity over D 2 , e.g., compound SCH23390 (table 9), results in a complete loss of antipsychotic activity [5,9,10]. Therefore, LE403 and LE404 might display an antipsychotic effect which however needs to be proven in in vivo studies. Only in vivo studies take into account the complexity of neuropsychiatric diseases including expression, distribution, and regulation of multiple receptors as well as adaptive processes.
This study confirmed recent findings that an increase in the size of the residue of the azecine nitrogen is detrimental to the affinity at dopamine/5HT 2A receptors (table 2) [11]. Hydroxylated versus non-hydroxylated dibenzacezines differ in their affinity and selectivity profiles (LE410, LE404, table 2) and define 2 separate groups. Monohydroxylation (LE404) results in higher potency than bis-hydroxylated compounds (LE403). Abolishing the H-donor properties by exchanging hydroxyl by methoxy groups was detrimental to the potency (LE400 versus LE403). Binding data of all compounds in figure 1 have been used to establish a valid 3D-QSAR pharmacophore model for D 1 receptors ( figure 8). The resulting model shows excellent q 2 values for crossvalidation results and random groups PLS tests for both, CoMFA and CoMSIA (figure 9) excluding a chance correlation. The pharmacophore model is thus a solid basis for further improvement of dopamine receptor ligands.

Conclusion
In conclusion, this study has revealed two compounds, the dibenzacezines LE410 and LE404 with a novel dopamine/5HT 2A receptor profile. LE404 and LE410 differ in their D 1 /D 2L selectivity. LE410 clusters in one group with the atypical antipsychotic clozapine but has a different D 2 -like receptor profile (hD 2L > hD 3 > hD 4.4 ) than clozapine (hD 4.4 > hD 2L > hD 3 ). LE404 clusters in a separate group from clozapine/LE410 and from haloperidol and shows increased D 1 selectivity similar to the experimental compound RMI-81582 which displayed antipsychotic activity [26]. An antipsychotic activity of LE404 and LE410 in in vivo studies still needs to be shown. Further, a validated 3D-QSAR pharmacophore model for D 1 antagonists is presented.

A B
using the DeBlasi equation [31]. Heterologous competition binding experiments were performed in Krebs-HEPES buffer in a final volume of 1.1 ml at 26°C for 2 h (D 1 -like receptors) or 3 h (D 2 -like receptors and 5HT 2A receptors) as described previously [11]. Cell membranes (total protein amount ~90 μg/tube) were incubated with 0.

Measurement of changes in intracellular [Ca 2+ ] in HEK293 cells
Measurement of changes in intracellular [Ca 2+ ] was performed as previously described using a NOVOstar microplate reader with a built-in pipetor (BMG LabTech, Offenburg, Germany) [11]. HEK293 cells expressing the respective dopamine receptor were loaded with 3 μM Oregon Green 488 BAPTA-1/AM (Molecular Probes, Eugene, OR) for 1 h at 25°C in Krebs-HEPES buffer containing 1% Pluronic F-127. Then, cells were rinsed three times with Krebs-HEPES buffer containing 0.5% bovine serum albumin, diluted, and evenly plated into 96 well plates (Greiner, Frickenhausen, Germany) at a density of 35,000 cells/well. Microplates were kept at 37°C. Fluorescence intensity was measured at 520 nm (bandwidth 35 nm) for 5 s at 0.4 s intervals to monitore baseline. Buffer alone or test compounds dissolved in buffer were then injected into separate wells, and fluorescence intensity was monitored at 520 nm (bandwidth 35 nm) for 25 s at 0.4 s intervals. Excitation wavelength was 485 nm (bandwidth 12 nm). Concentration-inhibition curves in the presence of the test compounds were obtained by preincubating the cells with the compounds for 30 min at 37°C prior to injection of agonist (hD 1 : 100 nM SKF38393; hD 2L : 30 nM quinpirole).

Measurement of changes in intracellular [cAMP] in HEK293 cells
Intracellular [cAMP] levels were estimated by using a cAMP reporter gene assay. pCRE-Luc Cis-Reporter plasmid (Path Detect ® CRE Cis-Reporting System, Stratagene, La Jolla, CA) coding for the firefly luciferase under the control of a cAMP response element was transiently transfected in HEK293 cells stably expressing the hD 1 or hD 2L receptor. 24 h after transfection, cells were reseeded in poly-L-lysine-coated (Biochrom, Berlin, Germany) white 96-well plates with clear bottom (Greiner, Frickenhausen, Germany) at a density of ~25,000 cells/well. Microplates were incubated for 48 h at 37°C and 5% CO 2 before using the cells for adenylyl cyclase stimulation or inhibition experiments. Cells were then exposed to increasing concentrations of test compounds dissolved in serum-free and phenol red-free medium and incubated for 3 h at 37°C and 5% CO 2 . In case of hD 2L , 10 μM forskolin was added. Antagonistic activity was tested by pre-incubation of test compounds for 30 min at 37°C and 5% CO 2 prior to the addition of agonist (hD 1 : 100 nM SKF38393; hD 2L : 100 nM quinpirole plus 10 μM forskolin) for 3 h. Incubation was terminated by adding 100 μl of cell lysis buffer (8 mM tricine, 1 mM dithiothreitol, 2 mM EDTA, 5 % Triton ® X-100, pH 7.8) for 20 min at 4°C. Luciferase activities were measured with the LUMIstar microplate reader (BMG LabTech, Offenburg, Germany). After monitoring the baseline for 0.3 s, 100 μl of luciferase assay reagent (30 mM tricine, 0.5 mM ATP, 10 mM MgSO 4 , 0.5 mM EDTA, 10 mM dithiothreitol, 0.5 mM coenzyme A, 0.5 mM Dluciferin, pH 7.8) was added and luminescence was measured at 25°C for 12.7 s at 0.1 s intervals. Luminescence was corrected by subtracting baseline levels.

Functional analysis of antagonism
Functional analysis of the antagonist effect of LE404 and LE410 was carried out by measuring the attenuation by LE404 or LE410 of the agonist-induced increase in intracellular [Ca 2+ ] in HEK293 cells recombinantly expressing hD 1 or hD 2L receptors. At least four antagonist concentrations were used. Functional data were used for nonlinear regression analysis according to Clark [20]. The pEC 50 values of the agonist curves were plotted against the concentration of test compounds LE404 or LE410 and analyzed by non-linear regression curve fitting using the following equations: (1) pEC 50 = -log ([B] n + 10 -pKb ) -log c (2) pEC 50 = -log ([B] + 10 -pKb ) -log c where [B] is the concentration of antagonist (LE404 or LE410), pK b is the negative decadic logarithm of the antagonist dissociation constant, n the Hill coefficient, and log c the difference between the pK b and the pEC 50 value of the agonist concentration-response curve in absence of the antagonist. Fits to equations (1) and (2) were compared by an F-test. where IC 50 is the inhibitory concentration of the antagonist to block by 50% the agonist effect, EC 50 is the effective concentration 50% of the used agonist (i.e., SKF38393 for hD 1 , and quinpirole for hD 2L receptors), and L is the molar concentration of the used agonist. Data (data points in figures and numbers in tables) are given as mean ± SEM of at least three independent experiments each performed with triplicates unless otherwise stated. Statistical analyses including principal component analysis were performed using SPSS (version 12.0.1 for Windows).

3D-QSAR (CoMFA/CoMSIA) studies
All calculations were carried out on an x86-compatible PC running SuSE-Linux 9.2. For molecular modelling, SYBYL 7.0 (Tripos Inc., St. Louis, Missouri, USA) and MOE 2004.03 for Linux (Chemical Computing Group Inc., Montreal, Quebec, Canada) were used. Conformational clustering was done using MATLAB Release 13 for Linux (The MathWorks Inc., Natick, MA, USA). Conformational analyses of all 12 compounds from figure 1 were done using a repeated molecular dynamics based simulated annealing approach as implemented in SYBYL 7.0. MMFF94 served as the force field with distance dependent electrostatics. A molecule was heated up to 1000 K within 2000 fs, held at this temperature for 2000 fs and annealed to 0 K for 10000 fs using an exponential annealing function. By applying this procedure, a total of 100 conformations were sampled during 100 cycles to account for conformational flexibility and to find the most likely conformations occurring most often in the resulting pool. This was done for both configurations of the protonated nitrogen atom because molecular mechanics is not able to switch configurations. All conformations were then optimized with the semi-empirical quantum mechanics method AM1 as implemented in MOPAC 6 from SYBYL and further compared using the SYBYL MATCH algorithm. Subsequently, a MATLAB clustering algorithm was used to extract the most divergent conformations using the root mean square (RMS) values of the comparison and the AM1 heat of formation. The most diverse and most often represented conformations of each compound were selected and overlaid with the pharmacophore resulting from the rigid ligand (-)-2b-SCH-39166 using the program MOE. The best 2-4 matched alignments per compound were selected for the CoMFA/CoMSIA study upon minimum RMS criteria and visual examination. These conformations were transferred to a SYBYL database and used as an initial alignment for the CoMFA/CoMSIA study. During an automated procedure, all possible combinations were tested on the CoMFA and CoMSIA combined steric/electrostatic fields with partial least squares analysis (PLS). In subsequent PLS analyses, the alignment was refined and the CoMFA/CoMSIA models were optimized. To prove that these models were not a result of a chance correlation, a stability test was performed using the random groups PLS method. Within this method, cross-validation was done with groups of more than one compound, which were excluded earlier during the model-building regression. Unlike the leave-one-out cross-validation, these groups are selected on a random basis and instead of twelve cross-validation groups, only five were used. Because of the random selection of the group members, this cross-validation was repeated a hundred times.

Authors' contributions
AH established recombinant cell lines, carried out functional measurements and radioligand binding studies, and performed data evaluation. Further, AH drafted the manuscript. MWeigt carried out the 3D-QSAR studies.
MWiese provided intellectual input and critical interpretation of the data. BH carried out calcium measurements. JL provided the LE compounds. MUK carried out principal component analyses and finalized the manuscript for publication. All authors read and approved the final manuscript.