Dopamine Receptor Ligand Selectivity—An In Silico/In Vitro Insight

Different dopamine receptor (DR) subtypes are involved in pathophysiological conditions such as Parkinson’s Disease (PD), schizophrenia and depression. While many DR-targeting drugs have been approved by the U.S. Food and Drug Administration (FDA), only a very small number are truly selective for one of the DR subtypes. Additionally, most of them show promiscuous activity at related G-protein coupled receptors, thus suffering from diverse side-effect profiles. Multiple studies have shown that combined in silico/in vitro approaches are a valuable contribution to drug discovery processes. They can also be applied to divulge the mechanisms behind ligand selectivity. In this study, novel DR ligands were investigated in vitro to assess binding affinities at different DR subtypes. Thus, nine D2R/D3R-selective ligands (micro- to nanomolar binding affinities, D3R-selective profile) were successfully identified. The most promising ligand exerted nanomolar D3R activity (Ki = 2.3 nM) with 263.7-fold D2R/D3R selectivity. Subsequently, ligand selectivity was rationalized in silico based on ligand interaction with a secondary binding pocket, supporting the selectivity data determined in vitro. The developed workflow and identified ligands could aid in the further understanding of the structural motifs responsible for DR subtype selectivity, thus benefitting drug development in D2R/D3R-associated pathologies such as PD.


Introduction
G-protein coupled receptors (GPCRs) are one of the most prominent protein families targeted in drug research. Currently, they are represented by 475 approved (by the FDA) drugs acting on 108 different GPCRs [1]. Sixty-five of them target an essential sub-group of GPCRs, the dopamine receptor (DR) family, consisting of the subtypes 1, 2, 3, 4 and 5 (D 1 R, D 2 R, D 3 R, D 4 R and D 5 R, respectively) [2]. The DR family is divided into D 1 like-(D 1 R and D 5 R) and D 2 like receptors (D 2 R, D 3 R and D 4 R) and plays a crucial role in physiological processes such a motoric function, cognition, sleep and memory [3]. However, it is also involved in many devastating diseases of the central nervous system (CNS) such as Parkinson's Disease (PD), schizophrenia and bipolar disorders. DR-targeting drugs act in different ways acting as, e g., agonists in PD by activating the receptor, antagonists in schizophrenia by blocking the receptor or partial agonists used in treating bipolar disorders or addiction [3][4][5].
While all of the listed diseases are connected to the dopaminergic system, they are also characterized by a distinct dysfunctionality of different dopaminergic projection pathways [6]. In PD, the degeneration of dopaminergic neurons in the substantia nigra leads to reduced dopamine levels, thus reducing activation of the D 2 R [4]. In contrast, schizophrenia is defined by hyperproductive, presynaptical dopaminergic neurons in the mesolimbic region, thus overactivating the D 2 R. At the same time, dopaminergic neurons in the prefrontal cortex are hypofunctional, resulting in insufficient activation of the D 1 R due to a lack of dopamine [3]. On the one hand, aberrant signalling involving the D 3 R has been implicated in diseases such as PD, restless leg syndrome and depression, where agonists are used to treat motor dysbalances. On the other hand, D 3 R antagonists have been shown to be useful as antipsychotics [7,8].
For most of those conditions, DR-targeting drugs have been approved by the FDA, successfully ameliorating major symptoms [2]. At the same time, they suffer from major drawbacks due to promiscuous activity at the DR subtypes other than the intended one, as well as closely related GPCRs [9]. Levodopa (L-DOPA), the gold standard in treating PD, successfully reduces the major motoric symptoms, such as bradykinesia and tremor, after biotransformation to dopamine, subsequently activating the D 2 R. Problematically, L-DOPA (and dopamine, respectively) is also known to induce dyskinesia (L-DOPA-induced dyskinesia) due to the promiscuous activation of the D 1 R in long-term treatment conditions [10]. While D 2 R agonists play an important role in treating PD, antagonists act as potent antipsychotics in different psychiatric disorders associated with the DR family. Those antipsychotics are also tightly connected to serious adverse drug events such as extrapyramidal syndrome and neuroleptic malignant syndrome [11,12]. D 2 R-selective drugs are clearly beneficial in treating PD and psychiatric disorders by alleviating the mentioned off-target effects. However, DR-subtype selectivity should not only be seen as a tool to counteract side effects but also to open up novel therapeutic avenues. Selective D 3 R agonists have been shown to be effective in vivo by mitigating the cell death of dopaminergic neurons and improving behavioural performances in mouse models of PD [13,14]. Interesting results have also been obtained in clinical studies establishing pramipexole (a D 3 R-preferring ligand) as an effective dopamine substitute in patients not responding to L-DOPA treatment, simultaneously delaying dyskinesia [15]. Another in vivo study indicated the capability of D 3 R-preferring agonists to reverse motivational deficits related to PD [16]. D 3 R-selective antagonists present a promising opportunity in the treatment of schizophrenia. They appear to be completely devoid of the D 2 R-associated side effects described earlier and also treat negative symptoms, which are not covered by conventional antipsychotics [17,18]. Selective D 1 R agonists are particularly interesting in treating cognitive deficits affecting patients suffering from schizophrenia by targeting the prefrontal cortex. Since the clinical relevance of D 1 R agonists was recognized early on, several selective compounds with diverse chemical scaffolds have been designed throughout the years [19]. While many selective ligands are considered a success, they also suffer from limited oral bioavailability and poor blood-brain barrier (BBB) permeability, thus exposing them to rapid peripheral metabolism. This has mainly been attributed to the presence of catechol functionalities in many of the ligands [20]. Different agonists have shown promising results in improving cognitive impairments and working memory in schizophrenia [21][22][23]. Unfortunately, other studies have provided evidence for D 1 R agonists being responsible for inducing seizures [24,25]. While the seizure-inducing mechanisms and the involvement of structure-activity relationships (SAR) are still not fully understood, the development of novel, potentially non-catechol agonists continues [19,26]. All of these findings clearly indicate the benefits of DR-subtype selective drugs. Moreover, they highlight the necessity of better understanding the molecular mechanisms involved in DR-ligand interactions to rationalize the SARs responsible for specific effects.
Drug research in the field of GPCRs has been benefitting from the 'golden age of GPCR structural biology' in the discipline of cheminformatics [27]. Different studies have been utilizing computer-assisted drug-design (CADD) methods to investigate GPCRs and also different DR subtypes [28][29][30]. A particularly interesting study by Bueschbell et al. investigated the selectivity of several known DR ligands (e.g., apomorphine and bromocriptine) with homology modelling and molecular docking approaches [31]. The ever-increasing availability of X-ray or cryo-EM structures of the discussed DR subtypes, D 1 R, D 2 R and D 3 R, aids our ability to comprehend DR ligand selectivity. In total, twelve three-dimensional (3D) protein structures of the D 1 R, five D 2 R structures and three D 3 R structures are accessible in the Protein Data Bank (PDB) database as of March 2023. The advent of cryo-EM technologies enabled the high-resolution depiction of the complex DR subtype structures at ≤3 Å, potentially improving molecular docking approaches that investigate DR ligand selectivity.
The conserved amino acids that create the orthosteric binding pockets (OBPs) of virtually all DR subtypes are well known and described [31,32]. Asp 3.32 in transmembrane (TM) 3 is responsible for ligand recognition forming a salt bridge with the positively charged amine function of ligands. The serine triade consisting of Ser 5.42 , Ser 5.43 and Ser 5.46 positioned in TM5 is important in orienting the respective ligand (especially if a catechol functional group is involved) and considering the ligands' binding affinity. An aromatic microdomain in TM6 includes Trp 6.48 , Phe 6.51 and Phe 6.52 as well as His/Asn 6.55 and is involved in activating the receptor upon interaction with an agonist. Agonist binding induces the so-called 'rotamer toggle switch', a domino-like cascade along TM6 reorienting the named amino acids, eventually triggering receptor activation. Less is known about DR sub-domains or structural elements originating from ligands responsible for selectivity. The D 1 R, although belonging to the D 1 -like DR family, is phylogenetically closest to the β-adrenergic receptors (βARs) [33]. Consequently, it features distinct motifs, responsible for selectivity. A study by Zhuang et al. suggested the involvement of the extracellular loop (ECL) 2, more specifically Ser188, which enables the D 1 R to accommodate bulkier ligands such as SKF81297 and SKF83959 [34]. In comparison, the same ligands would sterically clash with the corresponding amino acid Ile184 in the ECL2 of the D 2 R, consequently resulting in D 1 R-selectivity over the D 2 -like DR family. Considering selectivity between D 2 R and D 3 R, work by Newman et al. revealed a secondary binding pocket (SBP), consisting of multiple amino acids such as Val 2.61 , Leu 2.64 , Phe 3.28 and conserved Gly and Cys residues located in ECL1 and ECL2, respectively [35]. In more detail, Michino and colleagues have suggested the Gly residue in ECL1 to be the critical selectivity determinant [36]. Additionally, studies have shown that the D 3 R possesses an intrinsically higher affinity towards ligands such as dopamine and quinpirole. Robinson and colleagues have shown that the intracellular loop (ICL) 3 might be responsible for this behaviour. Generating D 2 R hybrids containing the D 3 R-ICL3 motif could increase ligand affinity 10-to 20-fold compared with the wild-type D 2 R. A D 3 R-D 2 R-ICL3 hybrid showed inverse effects [37]. An overview of the described SBP and the different domains involved in DR subtype selectivity is shown in Figure 1.
A great deal of effort has been invested in CADD-approaches to investigate and discover potential DR subtype-selective ligands, thus benefitting drug development in, e.g., neurodegenerative diseases such as PD [28,30,31,34,38,39]. However, due to the complexity of DR selectivity, in silico approaches require in vitro validation. In vitro binding affinities at different DR subtypes can be investigated using, e.g., homogenous timeresolved fluorescence (HTRF) assays, which are standardizable, commercially available and also semi-high-throughput compatible [40,41].
Therefore, the aim of this study was to develop a combined in silico/in vitro approach to assess the selectivity of novel DR ligands at different receptor subtypes using a cell-based HTRF assay as well as a molecular docking approach. Discovering DR-selective ligands as well as providing more detailed insights into their binding behaviour would contribute to better pharmacological tools and new starting points in drug development.

Figure 1.
Overview of DR sub-domains relevant in DR subtype selectivity. Red dots highlight the highly conserved amino acids Val 2.61 , Leu 2.64 , Phe 3.28 , Gly ECL1 and Leu ECL2 in the SBP of D2R and D3R. Zoomed in box of the conserved SBP shows the 3D arrangement. Partial primary sequences (amino acid positions are shown in the index) of ICL3 are shown for both D2R and D3R. (Created with Bio-Render.com (accessed on 13 April 2023)).
A great deal of effort has been invested in CADD-approaches to investigate and discover potential DR subtype-selective ligands, thus benefitting drug development in, e.g., neurodegenerative diseases such as PD [28,30,31,34,38,39]. However, due to the complexity of DR selectivity, in silico approaches require in vitro validation. In vitro binding affinities at different DR subtypes can be investigated using, e.g., homogenous time-resolved fluorescence (HTRF) assays, which are standardizable, commercially available and also semi-high-throughput compatible [40,41].
Therefore, the aim of this study was to develop a combined in silico/in vitro approach to assess the selectivity of novel DR ligands at different receptor subtypes using a cellbased HTRF assay as well as a molecular docking approach. Discovering DR-selective ligands as well as providing more detailed insights into their binding behaviour would contribute to better pharmacological tools and new starting points in drug development.

Ligand Selection for Combined In Silico/In Vitro Approach
Compounds selected for combined in silico/in vitro investigations were chosen based on two main criteria. First, compounds identified as active D 2 R ligands with the previously developed workflow shown in Zell et al. [42] were selected for further in vitro investigations. Second, other compounds from this study showing normalized decreased fluorescence (NDF) values ≥2-fold increased (during in vitro activity screening, Section 2.10) at any of the investigated DR subtypes compared with the other two DR subtypes were included in further investigations.

Similarity Assessment-Tanimoto Scoring (TS) Matrix
Canonical SMILES codes of all the compounds of interest were imported to Canvas version 3.8 (Canvas, Schrödinger Inc., New York, NY, USA). In Canvas version 3.8, radial fingerprints (Extended Connectivity Fingerprint (ECFP4) [43,44]) of all the molecules (based on 2D structures) were calculated followed by an automated calculation of a TS [45] for each compared pair. TS matrices were exported to Excel (Microsoft, Redmond, WA, USA) as csv files and imported to GraphPad Prism version 8 (GraphPad Software, San Diego, CA, USA) to display heatmaps, color-coding the structural similarities. An increasing coefficient indicated an increasing structural similarity. (Dis-)similarities considering chemical scaffolds were further used to assess observed in silico/in vitro phenomena.

Molecular Docking Workflow
Docking was performed using GOLD version 5.8.0 (CCDC, Cambridge, United Kingdom) [120]. Protein structures were not energetically minimized during the docking process. Hydrogens were added to all protein structures. CHEMPLP was used as a scoring function, not allowing early termination. For defining the binding site, all atoms within 6 Å of the bound ligand (depending on the cryo-EM structure) were chosen. The number of GA runs was set to 30.
2.6.1. Molecular Docking-D 1 R Molecular docking into the D 1 R was performed using the apomorphine-bound cryo-EM structure of the PDB entry 7jvq [34]. Specific settings for the D 1 R structure used during docking are shown in Table 1.  Figure S1 and Tables S5-S8). A detailed description of the MDS calculation is given in Tables S5-S8.

Molecular Docking D 2 R
Molecular docking into the D 2 R ligand binding site was performed using the MDSmodified (see Section 2.6.2) cryo-EM structure of the PDB entry 7jvr [34]. During docking, only ASP114R was specified as flexible (1 rotamer (free)).

Molecular Docking D 3 R
Molecular Docking into the D 3 R was performed using the PD128907-bound cryo-EM structure of the PDB entry 7cmv [121]. Specific settings for the D 3 R structure used during docking are shown in Table 2.

DR Subtypes-BLASTP Alignment
To identify the analogous amino acids of the SBP of D 1 R and D 2 R in respect to D 3 R (defined in [35,36]) a BLASTP alignment was performed (https://blast.ncbi.nlm.nih.gov (accessed on 10 March 2023)) [122]. The relevant amino acids in regard to the SBP of D 3 R are shown in Table 3. Table 3. Overview of the amino acids forming the SBP in different DR subtypes. D 2 like subtypes include D 2 R and D 3 R.

DR Subtype
The respective amino acids were used during the in silico selectivity assessment during the validation process and the analysis of the novel DR ligands.

Validation of the Molecular Docking Approach-ChEMBL Dataset(s)
Molecular docking results for each ChEMBL dataset (containing 30 poses for each compound) were uploaded to Pipeline Pilot Client version 9.1 (Accelrys, San Diego, CA, USA) [123]. Duplicates from each molecular docking output were removed. Only topranked poses (based on fitness score) of each docked compound were retained in the datasets used for further evaluation. The subsequent docking analysis (top-ranked poses) was performed using DiscoveryStudio (DS) 2018 Client (Accelrys, San Diego, CA, USA). The (modified) DR subtype protein structures were loaded into DS. The conserved Gly residues (shown in Table 3) were marked; centroids were calculated and checked as center of mass (COM). Subsequently, the docked DR subtype-selective output files were loaded into the respective DR protein structure. All molecules were marked and COM was calculated with respect to the Gly residues. Finally, distances between COM (Gly residue) and COM (docked ligands) were calculated in Å.

Docking Analysis-Novel DR Ligands
The docking analysis of the novel ligands was performed using LigandScout version 4.4.4 (Inte:Ligand GmbH, Vienna, Austria). Docked ligands (sd files) were loaded into the different DR protein structures (D 1 R into 7jvq [34]; D 2 R into the MDS-modified D 2 R 7jvr [34]; and D 3 R into 7cmv [121], respectively). All 30 poses of each ligand were individually superimposed and the most frequent pose was assessed visually. Subsequently, DR protein structures as well as molecular docking output files, were loaded in PyMOL (Schrödinger Inc., New York, NY, USA) for each DR subtype individually. The protein including the most frequent respective ligand pose (taking the highest-ranking according to fitness score) was extracted as a pdb file. The resulting pdb files were loaded into DS for calculating distances [Å], as shown in Section 2.8.

HTRF-Based Receptor Binding Studies
All HTRF assays were performed using an HTRF-compatible plate reader (model Tecan Spark (Tecan Group, Männedorf, Switzerland)). The respective settings were specifically modified and optimized for the determination of D 2 R ligand-binding affinities. Binding affinities were determined using the same settings for measurements with D 1 R and D 3 R carrier cells. Experiments were performed using two different emission wavelengths at 620 (control) and 665 (D 2/3 R)/510 (D 1 R) nm, respectively. Fluorophores were excited at 320 nm. A dichroic 510 mirror was used, while lag and integration times of 100 and 400 µs were applied, respectively. Flashes were set to 75. Electronic gain was automatically optimized, while the z-position was optimized based on the well with the highest expected signal. Experiments described in Sections 2.11, 2.12 and 2.14 required the use of two 96-well plates. The first plate was used to determine the gain and the z-position. Subsequently, the determined values were set manually for the second plate to enable direct comparison between the different plates.

Characterization of DR Carrier Cells (D 1 R and D 3 R)-Kd Determination
The cells used for the subsequent screening and detailed investigation of D 1 R and D 3 R ligands were acquired from PerkinElmer/cisbio (Waltham, MA, USA; Tag-lite Dopamine D1 or D3a-labeled Cells, ready-to-use (transformed and labeled), 200 tests; C1TT1D1 and C1TT1D3A, respectively). The cells were stored in liquid nitrogen until further use. Fluorescent-labelled ligands (Dopamine D2 Receptor red antagonist Fluorescent Ligand (L0002RED), stored at −20 • C and Dopamine D1 Receptor green antagonist (L0031GRE), stored at −20 • C), assay buffer (5Xconcentrate Tag-lite Buffer (TLB), 100 mL, stored at +4 • C; LABMED), and 96-well plates (HTRF 96-well low-volume white plate; 66PL96005) required for the in vitro assay were also acquired from PerkinElmer/cisbio. The assay was conducted according to the standard operation protocol (SOP) available from PerkinElmer/cisbio. The 96-well plates were incubated at room temperature for 2 h. The 96-well plates were read as described in Section 2.10. The respective concentrations of the dilution series were performed in triplicates. In total, Kd determination was performed twice.
The characterization of the D 2 R carrier cells is detailed in [42].

In Vitro Screening-Assessment of Compound Activity
Materials described in Sections 2.1 and 2.11 were also used during ligand screening. TLB (1X was prepared diluting 5Xconcentrate TLB in water. For ligand screening, compounds were prepared at a working solution concentration of 40 µM in 1XTLB. Compound 1, apomorphine, was used as the positive control at the same concentration. The assay was conducted in duplicates following the SOP available from PerkinElmer/cisbio and as described in Section 2.10.

Ligand Selection for Ki Determination
Ligand selection was based on NDF values detailed in Table S9. Novel D 2 R ligands from our previous study (compounds 2, 3, 5, 6, 7 and 9) [42] were selected for selectivity assessment. Additionally, compounds 4, 8 and 10 were investigated due to an NDF folddifference ≥2 of any of the three DR subtypes compared with the other two.

K I Determination for Selected Ligands
The materials described in 2.1 and 2.11 were also used for Ki determination of the ligands selected after screening. The selected ligands (compounds 1-10) were diluted in 1x TLB. Compounds 1, 2 and 4-9 were diluted to an initial working solution concentration of 4 × 10 −4 M. Compounds 3 and 10 were diluted to an initial working solution concentration of 1 × 10 −4 M. Different concentrations were chosen due to differences in aqueous solubility of the compounds. The Ki was determined in duplicates following the SOP available from PerkinElmer/cisbio and as described in Section 2.10.

Data Processing, Representation and Analysis
Saturation binding curves were processed and visualized in GraphPad Prism 8 (Nonlinear regression (curve fit), One site-Fit logIC 50 was performed using GraphPad Prism version 8.2.1 for Windows, GraphPad Software, San Diego. CA, USA). Molecular docking was performed in GOLD 5.8.0 (CCDC, Cambridge, United Kingdom) [120]. Docking analysis was performed in LigandScout version 4.4.5 (Inte:Ligand GmbH, Vienna, Austria) [124]. MDS and calculations for distance-based in silico approach were performed in DS Client 2018 (DiscoveryStudio, Accelrys Inc., San Diego, CA, USA). Twodimensional structures of all shown compounds were generated using ChemDraw version 19.0 (PerkinElmer, Waltham, MA, USA). SD files used for similarity assessment were generated using PipelinePilot Client 9.1 (Dassault Systems, BIOVIA Discovery Studio, San Diego, CA, USA, 2018). Similarity assessment was performed using Canvas 3.8 (Canvas, Schrödinger, LLC, New York, NY, USA, 2021). Docking alignments and visualization were performed in PyMOL (PyMOL, Schrödinger, LLC, New York, NY, USA, 2021).

Structural Summary of the Investigated Ligands
All compounds investigated in silico and in vitro within this study are shown in Figure 2.
The novel ligands (compounds 2-10) investigated with the combined approach within the scope of this study were structurally compared with each other using a Tanimoto scoring (TS) matrix. Therefore, the observed in silico and/or in vitro phenomena could be potentially correlated to structural (dis-)similarities. The TS matrix is shown in Figure 3, ranging from 0 (green) to 1 (red), corresponding to structurally unrelated and identical compounds, respectively. LLC, New York, NY, USA, 2021). Docking alignments and visualization were performed in PyMOL (PyMOL, Schrödinger, LLC, New York, NY, USA, 2021).

Structural Summary of the Investigated Ligands
All compounds investigated in silico and in vitro within this study are shown in  The novel ligands (compounds 2-10) investigated with the combined approach within the scope of this study were structurally compared with each other using a Tanimoto scoring (TS) matrix. Therefore, the observed in silico and/or in vitro phenomena could be potentially correlated to structural (dis-)similarities. The TS matrix is shown in Figure 3, ranging from 0 (green) to 1 (red), corresponding to structurally unrelated and identical compounds, respectively.   Thirty-three out of thirty-six pairs scored between 0.03 and 0.15, thus representing a structurally diverse compound collection. Only three pairs, i.e., compounds 5 and 6 (TS 0.28), 6 and 10 (TS 0.21) and 9 and 10 (TS 0.21), were characterized by a similarity score of >0.21, reflecting a higher degree of similarity (considering the use of radial fingerprints).

In Vitro Compound Screening-An Assessment of DR Subtype Selectivity
The investigated compounds were taken from a previous pharmacophore-based virtual screening study described in Zell et al. [42]. All 2D structures (compounds 2-10 and SC285-SC365) and respective NDF values for all DR subtypes are shown in Table S9 and Figures S2-S10. The activities of all compounds were investigated via a competitive bind- Thirty-three out of thirty-six pairs scored between 0.03 and 0.15, thus representing a structurally diverse compound collection. Only three pairs, i.e., compounds 5 and 6 (TS 0.28), 6 and 10 (TS 0.21) and 9 and 10 (TS 0.21), were characterized by a similarity score of >0.21, reflecting a higher degree of similarity (considering the use of radial fingerprints).

In Vitro Compound Screening-An Assessment of DR Subtype Selectivity
The investigated compounds were taken from a previous pharmacophore-based virtual screening study described in Zell et al. [42]. All 2D structures (compounds 2-10 and SC285-SC365) and respective NDF values for all DR subtypes are shown in Table S9 and Figures S2-S10. The activities of all compounds were investigated via a competitive binding (in comparison with a fluorescence-labeled ligand) of the respective compounds at D 1 R/D 3 R, utilizing an HTRF assay using a screening concentration of 10 µM. NDF values of compounds chosen for further evaluation are shown in Table 4. Table 4. Summary of the in vitro screening of known and potential DR ligands considered selective for one of the three investigated subtypes. All measurements were conducted at a concentration of 10 µM (n = 4). Fluorescence decrease was normalized to the control. Cpd., compound.
Based on the resulting NDF values, all compounds but 4 and 8 showed promiscuous receptor activities suggesting diverse selectivity profiles. Only compounds 4 and 8 showed NDF values close to 1 at both D 1 R and D 2 R, suggesting inactivity at those DR subtypes and, respectively, selectivity for the D 3 R.

Ki Determination-Of the Selected Compounds at DR Subtypes
The selected compounds were investigated in vitro to determine their binding affinities (Ki values) at the three different DR subtypes, D 1 R, D 2 R and D 3 R. In Figure 4a,b, compounds 2 and 10 are shown as examples. Compound 2 represents a non-selective ligand while compound 10 is characterized by the highest selectivity (for D 3 R). The remaining binding curves are shown in Figure S11a-h.
The Ki values of all investigated ligands, as well as the calculated fold-differences for each receptor pair, are shown in Table 5.
All the compounds investigated in vitro, except compound 1, showed a clear D 2 like selectivity with preferences for D 3 R (D 1 R/D 3 R fold differences ranging from 3.06 to 1031.4, D 2 R/D 3 R fold-differences ranging from 1.66 to 263.7, respectively). While compounds 2, 5, 6, 9 and 10 showed higher affinities for D 1 R compared with the D 2 like DR subtypes, the affinities of compounds 3, 4 and 8 were not determinable for D 1 R, and were thus considered inactive. Compounds 4 and 8 were also inactive at D 2 R, and were thus considered D 3 Rselective. Only compound 1 was characterized by the lowest binding affinity for D 2 R. Interestingly, all compounds showed the highest affinity at the D 3 R. The Ki values of all investigated ligands, as well as the calculated fold-differences for each receptor pair, are shown in Table 5.

Dataset Assembly-In Silico Assessment
To validate the molecular docking approach utilized to assess compound selectivity in silico, DR ligands with different selectivities for the DR subtypes D 1 R, D 2 R and D 3 R with known biological activities were extracted from the ChEMBL database. Compounds were only included in the final datasets if (I) their binding affinities were determined in vitro at all three DR subtypes and (II) their in vitro measurements included a valid control to assess assay functionality. The curated ChEMBL entries were divided into D 1 R-, D 2 R-, D 2 like-and D 3 R-selective subsets. D 1 R-, D 2 R-or D 3 R-selectivity was assumed for molecules with binding affinities ≤1000 nM at the respective subtype and ≥1000 nM at the others. D 2 like-selective compounds showed binding affinities ≤500 nM at D 2 R and D 3 R. The final datasets consisting of 29 (SC1-SC29 [46][47][48][49][50][51][52][53][54][55][56][57][58][59][60] Table S1 to Table S4.

Validation of Molecular Docking
The utilized molecular docking approach was based on the work of Michino and colleagues [36]. Therefore, different ChEMBL datasets, previously defined as DR subtypeselective (see Section 3.4), were docked into the 3D protein structures of D 1 R, D 2 R and D 3 R (molecular docking workflow described in Sections 2.6.1, 2.6.3 and 2.6.4). Due to the high number of investigated ligands (>300), only the top-ranked poses (considering the fitness score) were considered during further analysis. The COM for all ligands included in each specific DR-selective subset was calculated using DS. Distances of each COM with respect to each DRs conserved Gly residue (shown in Table 3) were calculated in [Å]. The calculated fold-differences for each subset, comparing different DRs with each other, are shown in Figure 5 (absolute distances determined in DS are given in Table S10). 6, 9 and 10 showed higher affinities for D1R compared with the D2like DR subtypes, the affinities of compounds 3, 4 and 8 were not determinable for D1R, and were thus considered inactive. Compounds 4 and 8 were also inactive at D2R, and were thus considered D3R-selective. Only compound 1 was characterized by the lowest binding affinity for D2R. Interestingly, all compounds showed the highest affinity at the D3R.

Validation of Molecular Docking
The utilized molecular docking approach was based on the work of Michino and colleagues [36]. Therefore, different ChEMBL datasets, previously defined as DR subtypeselective (see Section 3.4), were docked into the 3D protein structures of D1R, D2R and D3R (molecular docking workflow described in Sections 2.6.1, 2.6.3 and 2.6.4). Due to the high number of investigated ligands (>300), only the top-ranked poses (considering the fitness score) were considered during further analysis. The COM for all ligands included in each specific DR-selective subset was calculated using DS. Distances of each COM with respect to each DRs conserved Gly residue (shown in Table 3) were calculated in [Å]. The calculated fold-differences for each subset, comparing different DRs with each other, are shown in Figure 5 (absolute distances determined in DS are given in Table S10).  The dashed red line shown in Figure 6 indicates a fold-difference of 1.0, which indicates the same distance between the ligands collective COM and the conserved Gly residue after docking into the respective DR structures. All investigated datasets show a fold-difference close to 1.0 considering the D 2 R/D 3 R comparison. This means that they showed an almost identical distance between COM and the Gly residue. In contrast, all datasets showed an increased fold-difference >1.0, when comparing D 1 R with D 2 R or D 3 R, respectively. Details considering all datasets are shown in Table 6.
cates the same distance between the ligands collective COM and the conserved Gly residue after docking into the respective DR structures. All investigated datasets show a folddifference close to 1.0 considering the D2R/D3R comparison. This means that they showed an almost identical distance between COM and the Gly residue. In contrast, all datasets showed an increased fold-difference >1.0, when comparing D1R with D2R or D3R, respectively. Details considering all datasets are shown in Table 6.  Clearly, the approach was incapable of distinguishing D2R-and D3R-selectivity from each other based on the COM-Gly distance. However, the utilized molecular docking approach was capable of identifying D2like-selective ligands based on their position within the respective DRs OBP.

In Silico Assessment of DR Selectivity-Interaction with the SBP
For the in silico assessment of the selected compounds 1-10, the most frequent poses after docking were used. After calculating the fold-differences based on the distances between each ligand's individual COM and the respective Gly residue (within each of the  Clearly, the approach was incapable of distinguishing D 2 R-and D 3 R-selectivity from each other based on the COM-Gly distance. However, the utilized molecular docking approach was capable of identifying D 2 like-selective ligands based on their position within the respective DRs OBP.

In Silico Assessment of DR Selectivity-Interaction with the SBP
For the in silico assessment of the selected compounds 1-10, the most frequent poses after docking were used. After calculating the fold-differences based on the distances between each ligand's individual COM and the respective Gly residue (within each of the three DR subtypes SBP, shown in Table S11), they were plotted against the fold-differences based on the DR-specific Ki values (shown in Table 5) determined in vitro. The resulting scatter plot is shown in Figure 6.
In addition to the individual data points, Figure 6 shows regression curves for all DR pairs. The D 1 R/D 2 R curve (dots) was characterized by the steepest slope suggesting the capability of the in silico approach in discriminating D 2 R-selective ligands. While the slope for the D 1 R/D 3 R curve (squares) was less steep, the calculated fold-differences (based on distance, y-axis) was already higher at lower Ki-based fold-differences (x-axis), indicating a similar capability to discriminate D 3 R-selective ligands. The D 2 R/D 3 R curve (triangles) was flatter, with individual values scattered around 1.0. Consequently, this reflected the results shown in Figure 5, where D 2 R/D 3 R-selectivity could not be discriminated based on the selected approach. In summary, the developed distance-based in silico approach was highly capable in identifying D 2 like-selectivity. This was also indicated by the R 2 values (shown in Figure 6) regarding the D 1 R/D 2 R and the D 1 R/D 3 R comparison showing a positive correlation between increasing binding affinities with increasing selectivity.

Retrospective Analyis of the In Silico/In Vitro Correlation
To get a more detailed insight into the binding mode of each of the investigated ligands at the respective DR subtype, the most frequent docking poses of each compound (Figures 7-9 and S12-S17) were visualized in the different binding pockets using PyMOL. Figures 7 and 8 show the different binding poses of the non-selective compound 2 and compound 10, which had the highest D 3 R-selectivity.
In Figure 7, the tertiary amine functionality (contained in the piperazine motif) of compound 2 is clearly oriented towards the OBP, allowing the formation of the salt-bridge with Asp 3.32 (described as the crucial interaction to define a DR ligand). While the position of compound 2 was flipped in D 2 R in comparison with D 1 R and D 3 R (highlighted by the orientation of the chlorine, green), the overall positioning of compound 2 was similar in each DR subtype. Consequently, there was no distinct orientation of any of the poses towards the SBP, resulting in the non-selective binding with Ki fold-differences between 1.2 and 2.2 (see Table 5).
Biomedicines 2023, 11, x FOR PEER REVIEW 15 of 27 three DR subtypes SBP, shown in Table S11), they were plotted against the fold-differences based on the DR-specific Ki values (shown in Table 5) determined in vitro. The resulting scatter plot is shown in Figure 6. In addition to the individual data points, Figure 6 shows regression curves for all DR pairs. The D1R/D2R curve (dots) was characterized by the steepest slope suggesting the capability of the in silico approach in discriminating D2R-selective ligands. While the slope for the D1R/D3R curve (squares) was less steep, the calculated fold-differences (based on distance, y-axis) was already higher at lower Ki-based fold-differences (x-axis), indicating a similar capability to discriminate D3R-selective ligands. The D2R/D3R curve (triangles) was flatter, with individual values scattered around 1.0. Consequently, this reflected the results shown in Figure 5, where D2R/D3R-selectivity could not be discriminated based on the selected approach. In summary, the developed distance-based in silico approach was highly capable in identifying D2like-selectivity. This was also indicated by the R 2 values (shown in Figure 6) regarding the D1R/D2R and the D1R/D3R comparison showing a positive correlation between increasing binding affinities with increasing selectivity.

Retrospective Analyis of the In Silico/In Vitro Correlation
To get a more detailed insight into the binding mode of each of the investigated ligands at the respective DR subtype, the most frequent docking poses of each compound (Figures 7-9 and S12-S17) were visualized in the different binding pockets using PyMOL. Figures 7 and 8 show the different binding poses of the non-selective compound 2 and compound 10, which had the highest D3R-selectivity.  Table 3), respectively. Ki values determined in vitro are shown for each DR subtype. Two-dimensional structure of compound 2 is shown. Amine functional group involved in formation of the saltbridge is highlighted in red.  Table 3), respectively. Ki values determined in vitro are shown for each DR subtype. Two-dimensional structure of compound 2 is shown. Amine functional group involved in formation of the salt-bridge is highlighted in red.
In Figure 8, the tertiary amine functionality (contained in the piperazine motif) of compound 10 was again oriented towards the OBP. Thus, the salt-bridge formation with Asp 3.32 was possible. In contrast to compound 2, the binding poses of compound 10 were distinctly different in the respective DR subtypes. Comparing the poses in D 1 R and D 3 R, the D 3 R pose (green) was shifted slightly to the right towards the SBP. The D 2 R binding pose (orange) was clearly different from both D 1 R and D 3 R, with the chlorosubstituted ring clearly oriented towards the SBP. While this explained the observed D 1 R/D 2 R fold difference of 3.7, it did not correlate with the D 2 R/D 3 R fold-difference of 331.8. However, the detailed analysis of the binding poses of compound 10 correlated with the observed D 2 like selectivity determined in vitro. Additionally, it also partially confirmed the retrospective results of the distance-based approach shown Figure 6, highlighting the capability of the developed approach to identify D 2 like-selectivity.  Table 3), respectively. Ki values determined in vitro are shown for each DR subtype. Two-dimensional structure of compound 10 is shown. Amine functional group involved in formation of the salt-bridge is highlighted in red.  Table 3), respectively. Ki values determined in vitro are shown for each DR subtype. Two-dimensional structure of compound 4 is shown. Amine functional group involved in formation of the saltbridge is highlighted in red. n.d., not determinable.   These findings were also supported by the in-depth analyses of compounds 3, 5, 6, 7 and 9 (PyMOL alignments shown in Figures S12-S15 and S17), where D 2 R and D 3 R poses were distinctly oriented towards the SBP. However, in agreement with the findings considering compound 10, the D 2 R-and D 3 R binding poses did not correlate with the higher D 3 R binding affinities found in vitro. Again, the results allowed for the confirmation of D 2 like-selectivity of the investigated ligands.
Compounds 4 ( Figure 9) and 8 ( Figure S16) were the only compounds with no determinable binding affinity at D 1 R and D 2 R, additionally showing slightly increased distancebased fold-differences ( Figure 6 and Table S11), regarding D 2 R/D 3 R-selectivity, of 1.10 and 1.04, respectively.
This was also reflected in the binding pose of compound 4, where the D 3 R pose (green) was oriented closer to the SBP. Similar results were observed in the binding pocket comparison of compound 8.

Discussion
The characterized DR ligands showed different selectivity profiles. Interestingly, all ten compounds investigated by the developed in silico/in vitro approach (including the novel compounds 2-10) showed either D 3 R-preferences or clear D 3 R-selectivity. Compound 2, for example, showed fold-differences of 3.23 and 1.66 for D 1 R/D 3 R and D 2 R/D 3 R, respectively, thus exerting D 3 R-preferences. Compounds 4 and 8 were characterized by no determinable binding affinities at D 1 R and D 2 R, consequently they were categorized as D 3 R-selective. While compound 10 showed low to intermediate binding affinities at D 1 R (2.38 µM) and D 2 R (0.61 µM), it also exerted the highest quantifiable selectivity fold-differences with values of 1031.4 and 263.7 for D 1 R/D 3 R and D 2 R/D 3 R, respectively. Additionally, all investigated compounds but 1 were D 2 like-selective.
The rather promiscuous behavior of compound 2 is attributed to its structural similarity to clozapine, the prototypical representative of atypical antipsychotics (a drug class belonging to the atypical antipsychotics). While clozapine is characterized by its potent antipsychotic effect, it is also known as a so-called 'dirty drug' due to its promiscuous activity at a variety of aminergic GPCRs (including dopaminergic, serotonergic and adrenergic receptor families) [125]. Thus, a similar pharmacological profile of compound 2 was expected. This was not only confirmed by the in vitro data but also by the developed in silico approach, correlating the positioning of the ligand within the OBP and SBP with its respective DR subtype selectivity. Even though the binding behavior of compound 2 appeared non-selective, the in silico approach was capable of detecting the slight D 2 likepreference resulting in distance-based fold-differences of 1.33 and 1.28 for D 1 R/D 2 R and D 1 R/D 3 R, respectively. Moreover, the compound could be active at other GPCRs which were not investigated within this study. In general, the investigated compounds could be biologically active at other PCRs. For example, a study by Garcia-Romero and colleagues identified several antiparkinsonian molecules with polypharmacological profiles. Those molecules were also biologically active at other GPCRs such as muscarinic acetyl choline receptors and adenosine receptors but also at the norepinephrine transporter [126]. This study highlights the necessity of investigating the identified ligands in even more detail to potentially exploit potential polypharmacological aspects and, even more importantly, to identify possible off-target activities. However, this study deliberately focused on isolated ligand-receptor interactions to generate reliable in silico/in vitro correlation, thus elaborating upon DR selectivity mechanisms.
The comparison of compounds 6 and 10 allowed for very interesting insights into the DR subtype-selectivity profile of structurally similar ligands differing mainly regarding linker lengths. Compounds 6 and 10 are both characterized by two terminal aromatic rings and a linker region consisting of a piperazine motif, an amide functionality and an alkyl chain (see Figure 2). Moreover, they also share binding preferences at the different DR subtypes following D 1 R > D 2 R > D 3 R. Both compounds showed comparable Ki-based fold-differences of 5.77 (compound 6) and 3.91 (compound 10) for D 1 R/D 2 R. However, the D 1 R/D 3 R and D 2 R/D 3 R fold-differences increased drastically for compound 10 (1031.4 and 263.7, respectively) compared with compound 6 (20.7 and 3.59, respectively). Michino and colleagues showed similar phenomena in their study investigating the impact of the linker length in analogues of the highly D 3 R-selective compound R22 ([(R)-N-(4-(4-(2,3-dichlorophenyl)piperazin-1-yl)-3-hydroxybutyl)-1H-indole-2-carboxamide]) [36,79]. The investigated R-22 analogues included C3-to C5-linker regions. The C3-linker length resulted in non-selective binding behavior at D 2 R and D 3 R. The C5-linker length markedly reduced D 2 R/D 3 R selectivity. Only the C4 analogue retained a significant D 2 R/D 3 Rselectivity with a 45.7 fold-difference. Compound 6, including a C2-linker region, showed a comparably reduced D 2 R/D 3 R-selectivity of 3.59. In contrast, compound 10, including a C4-linker region, exerted a fold-difference of 263.7. While compounds 6 and 10 are only partially related (similarities shown in red) to the R22-analogues (see Figure 10), the observed in vitro effects are potentially attributable to the length of the linker region.
The comparison of compounds 6 and 10 allowed for very interesting insights into the DR subtype-selectivity profile of structurally similar ligands differing mainly regarding linker lengths. Compounds 6 and 10 are both characterized by two terminal aromatic rings and a linker region consisting of a piperazine motif, an amide functionality and an alkyl chain (see Figure 2). Moreover, they also share binding preferences at the different DR subtypes following D1R > D2R > D3R. Both compounds showed comparable Ki-based folddifferences of 5.77 (compound 6) and 3.91 (compound 10) for D1R/D2R. However, the D1R/D3R and D2R/D3R fold-differences increased drastically for compound 10 (1031.4 and 263.7, respectively) compared with compound 6 (20.7 and 3.59, respectively). Michino and colleagues showed similar phenomena in their study investigating the impact of the linker length in analogues of the highly D3R-selective compound R22 ([(R)-N-(4-(4-(2,3-dichlorophenyl)piperazin-1-yl)-3-hydroxybutyl)-1H-indole-2-carboxamide]) [36,79]. The investigated R-22 analogues included C3-to C5-linker regions. The C3-linker length resulted in non-selective binding behavior at D2R and D3R. The C5-linker length markedly reduced D2R/D3R selectivity. Only the C4 analogue retained a significant D2R/D3R-selectivity with a 45.7 fold-difference. Compound 6, including a C2-linker region, showed a comparably reduced D2R/D3R-selectivity of 3.59. In contrast, compound 10, including a C4-linker region, exerted a fold-difference of 263.7. While compounds 6 and 10 are only partially related (similarities shown in red) to the R22-analogues (see Figure 10), the observed in vitro effects are potentially attributable to the length of the linker region. Compounds 2, 3, 5, 6, 7 and 9 were already reviewed in our earlier publication investigating their novelty, as was the DR-associated effects of their closest structural relatives [42]. While none of the investigated structures yielded exact structural matches, the most similar structures were associated with different DR-related effects. Structurally similar compounds to 5 and 6 were associated with D4R-selectivity but no defined mode of action (agonism or antagonism) [127]. A compound similar to 2 was associated with D4R antagonism, while structurally similar ligands regarding compounds 3 and 7 were investigated considering D2R antagonism [128][129][130][131]. Only a compound structurally similar to 9 was Figure 10. Comparison of the chemical scaffolds of compounds 6, 10 and the R22-analogue. Structural elements highlighted in red show similarities between the different compounds also indicating the differences in linker length. Compounds 2, 3, 5, 6, 7 and 9 were already reviewed in our earlier publication investigating their novelty, as was the DR-associated effects of their closest structural relatives [42]. While none of the investigated structures yielded exact structural matches, the most similar structures were associated with different DR-related effects. Structurally similar compounds to 5 and 6 were associated with D 4 R-selectivity but no defined mode of action (agonism or antagonism) [127]. A compound similar to 2 was associated with D 4 R antagonism, while structurally similar ligands regarding compounds 3 and 7 were investigated considering D 2 R antagonism [128][129][130][131]. Only a compound structurally similar to 9 was associated with D 3 R-selectivity and D 2 R antagonism [132]. The novel ligands included within this study were compared with the literature using SwissTargetPrediction (http://www.swisstargetprediction.ch/ (accessed on 5 March 2023)) and SwissSimilarity (http://www.swisssimilarity.ch/ (accessed on 5 March 2023)) [133][134][135]. Compounds 4 and 8 yielded low scores in SwissTargetPrediction where the identified similar compounds (ChEMBL IDs 59603 and 592377) had been investigated considering D 1 R-and D 2 R activity but not D 3 R selectivity [136,137]. ChEMBL entry 4081151 was structurally closely related to compound 4 but had only been investigated for kappa opioid receptors [138]. A SwissSimilarity match for compound 8 (ChEMBL ID 1094101) was investigated for its binding affinity at serotonergic receptors and aminergic GPCR family members, but not in respect to DRs [139]. Thus, compounds 4 and 8 open up novel insights into D 3 R-selectivity. Compound 10 resulted in exact structural matches and closely related matches in both SwissTargetPredicition and SwissSimilarity investigating D 3 R-selectivity. Still, the comparison between compounds 6 and 10 contributes to a better understanding of the role of the linker length on the DR subtype selectivity of structurally related, but not identical, chemical scaffolds.
As mentioned earlier, all novel ligands exerted their highest binding affinities at the D 3 R. This is partially in accordance with the scientific literature, where the D 3 R shows a high intrinsic binding affinity for agonists such as dopamine (420-fold increased affinity) and quinpirole [37,140]. While this is attributed to intracellular loop 3 in D 3 R, this characteristic has only been shown for agonists. However, the known characteristics of the structurally related compounds of the novel compounds described above suggest a low probability that all investigated ligands are actually agonists. Thus, the increased D 3 R affinity of compounds 2-10 presumably originates from a distinct interaction with the described SBP [36]. The developed in silico approach proposes a workflow to identify D 2 like-selectivity. However, the static nature of the molecular docking approach does not allow for discrimination of D 2 R/D 3 R-selectivity. This limitation can be attributed to the very dynamic nature of the EL structural motifs of the D 2 like DRs responsible for subtype selectivity. Different studies propose MDS approaches to circumvent the shortcomings of molecular docking approaches and to account for protein flexibility [36,141].
Thus, the developed in silico/in vitro workflow clearly demonstrated its potential use in preclinical drug research by enabling the identification of D 2 like-selective ligands independently of chemical scaffolds. This could be especially important in diseases of the CNS, where D 1 R activation has been associated with induction of seizures. In addition, the D 3 R is a fast emerging molecular target of interest in treating PD. Thus, the accurate prediction of D 2 like-selectivity could act as an important starting point in developing truly D 3 R-selective compounds and also providing pharmacological tools to aid in the understanding of D 2 like DR subtype selectivity.

Conclusions
In this study, ten compounds were investigated for their DR subtype selectivity. A combined in silico/in vitro approach was developed to correlate the positioning within the receptor binding pocket with the biological activity. With the workflow, we observed a correlation between the distance of the ligand to the conserved glycine residue within the secondary binding pocket and the DR subtype selectivity. Most prominently, the workflow was able to identify D 2 like-selectivity but could not explain D 2 R/D 3 R selectivity observed in vitro. The most selective compound, 10, was characterized by a low nano-molar activity at D 3 R (Ki = 2.3 nM) showing a distinct selectivity over D 1 R and D 2 R with fold-differences of 1031.4 and 263.7, respectively. This study provides a valuable tool in further understanding DR subtype selectivity mechanisms, thus aiding the development of more selective DR ligands.
Supplementary Materials: The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/biomedicines11051468/s1, Figure S1: Summary of the MDS approach for the modification of the D 2 R structure; Figures S2-S10: Summary of the in vitro screening results of all investigated compounds; Figure S11: Overview of the in vitro determined binding affinities of selected compounds (1 and 3-9); Figures S12-S17: Detailed insights into the docking poses and binding pockets of selected compounds (3, 5, 6, 7, 8 and 9); Table S1: Dataset of D 1 R-selective compounds (ChEMBL); Table S2: Dataset of D 2 R-selective compounds (ChEMBL); Table S3: Dataset of D 2 like-selective compounds (ChEMBL); Table S4: Dataset of D 3 R-selective compounds (ChEMBL); Table S5: Summary of the Minimization Process for the D 2 R structure; Table S6: Detailed overview of the settings used during the Minimization Process of the D 2 R structure; Table  S7: Summary of the Standard Dynamics Cascade Process for the D 2 R structure; Table S8: Detailed overview of the settings used during the Standard Dynamics Cascade Process of the D 2 R structure; Table S9: Overview of 2D structures and NDF values; Table S10: Validation of in silico approach using ChEMBL; Table S11: Summary of the retrospective analysis of investigated compounds.