Structure-guided development of heterodimer-selective GPCR ligands

Crystal structures of G protein-coupled receptor (GPCR) ligand complexes allow a rational design of novel molecular probes and drugs. Here we report the structure-guided design, chemical synthesis and biological investigations of bivalent ligands for dopamine D2 receptor/neurotensin NTS1 receptor (D2R/NTS1R) heterodimers. The compounds of types 1–3 consist of three different D2R pharmacophores bound to an affinity-generating lipophilic appendage, a polyethylene glycol-based linker and the NTS1R agonist NT(8-13). The bivalent ligands show binding affinity in the picomolar range for cells coexpressing both GPCRs and unprecedented selectivity (up to three orders of magnitude), compared with cells that only express D2Rs. A functional switch is observed for the bivalent ligands 3b,c inhibiting cAMP formation in cells singly expressing D2Rs but stimulating cAMP accumulation in D2R/NTS1R-coexpressing cells. Moreover, the newly synthesized bivalent ligands show a strong, predominantly NTS1R-mediated β-arrestin-2 recruitment at the D2R/NTS1R-coexpressing cells.


Supplementary Figure 3. Attachment points of phenylpiperazine and aminoindane-type scaffolds determined by docking studies.
Ribbons and surface of D 2 R, the structures of the phenylpiperazine-(top) and the aminoindane-type scaffold (bottom) are shown in grey, yellow and cyan, respectively. Positions of the two scaffolds were achieved by docking both compounds into our recently described D 2 R homology model (left) 1 . Docking studies were performed as described in the methods section. In both cases, the basic nitrogen (highlighted in pink in all representations) is accessible from the extracellular side of the receptor and therefore was selected as attachment point for the lipophilic appendage for compounds of type 2 and 3.

Supplementary Figure 5:
Competition binding experiments evaluating D 2 R affinities for compounds 2e-g, 3e-g. Dopamine receptor binding of newly sythesized ligands was investigated by radioligand displacement with membranes from HEK 293T cells expressing only D 2 R (filled grey circles) or coexpressing D 2 R/NTS 1 R (open blue circles). (a-d) Bivalent compounds comprising a peptoid-peptide hybrid instead of NT (8)(9)(10)(11)(12)(13) (2e/f, 3e/f, m=1 and 2, n=4 for all experiments) do not exhibit bivalent binding modes as indicated by the absence of biphasic competition binding curves. (e,f) Typical monophasic binding curves are observed for the monovalent ligands 2g/3g (n=3 for D 2 R, n=6 for D 2 R/NTS 1 R). Data points represent mean ± s.e.m. of n independent experiments, each performed in triplicate. Figure 11: cAMP accumulation assay with ligands 2b, 3b and 2g, 3g under different conditions. Activation profiles of the bivalent ligands 2b, 3b and their monovalent congeners 2g, 3g were determined at different ratios of D 2 R/NTS 1 R expression or NTS 1 R monoexpression in presence (a,b) or absence (c-e) of 10 µM forskolin (+/-FSK). (a) Although potent activation for the D 2 R mediated inhibition of cAMP accumulation is seen for the monovalent ligand 3g (open blue circles) (n=5) and quinpirole (n=11), the bivalent analog 3b (filled blue circles) leads to an NTS 1 R mediated increase of the intracellular cAMP concentration (n=4). For ligand 2g (open red circles) (n=3) no significant influence on cAMP levels was observed, while the bivalent congener 2b (filled red circles) (n=5) also induces an increase of cAMP comparable to the effect of NT(8-13) (n=11). (b) At higher expression levels of NTS 1 R, the Gα s -stimulated response is increased (E max 148 % vs 124 % for NT (8)(9)(10)(11)(12)(13), n=5). However, activation profiles of ligands 2b,g (n=3) and 3b,g (n=4) remain unchanged relative to NT (8)(9)(10)(11)(12)(13) and quinpirole (n=5). (c) In the absence of forskolin (n=4 for all compounds), D 2 R activation is only weakly detectable. Quinpirole and the monovalent D 2 R-agonist 3g reduce cAMP levels by -25 % and -31 % respectively. The phenylpiperazine 2g has a marginal effect, it leads to a slight decrease in basal cAMP (-10 %) However the bivalent ligands 2b, 3b behave similar to NT (8)(9)(10)(11)(12)(13), increasing the cAMP levels to 83 % and 75 % of the maximum effect. (d) Under conditions with increased NTS 1 R expression (n=3 for all compounds), cAMP responses are comparable to the effects seen in (c). (e) In cells expressing NTS 1 R only, the monovalent ligands quinpirole, 2g, 3g do not elicit any response. Bivalent ligands 2b and 3b are able to increase cAMP levels in a manner similar to NT(8-13) (n=4 for all compounds) in the absence of FSK. In presence of FSK data was normalized to vehicle (0 %) and the effect obtained by FSK (10 µM, 100 %), in absence of FSK data was normalized to vehicle (0 %) and the maximum effect obtained with NT(8-13) (100 %). Ratios indicate the relative amount of cDNAs used for transfection. Data represent mean ± s.e.m. of n independent experiments, each performed in triplicate. Figure 12: Binding profile and activation properties for the coexpression of NTS 1 R with a signalling deficient D 2 R mutant. (a) Displacement of [ 3 H]spiperone from membranes of HEK 293T cells expressing NTS 1 R and a signalling incompetent D 2 R mutant (D80A) show biphasic binding behavior for the bivalent ligand 3b (K i high 0.014 ± 0.02 nM, K i low 310 ± 30 nM) (b) When a signalling incompetent D 2 R (D80A) mutant is coexpressed with NTS 1 R, quinpirole and 3g are unable to inhibit cAMP accumulation. However, the activation profiles of NT (8)(9)(10)(11)(12)(13) and the bivalent ligand 3b remain unchanged compared to cells coexpressing wild type D 2 R/NTS 1 R. Data represent mean ± s.e.m. of 3 independent experiments, each performed in triplicate. Figure 13: β-Arrestin-2 recruitment at D 2 R and D 2 R/NTS 1 R induced by the bivalent ligands 2c and 3c and an equimolar combination of quinpirole and NT (8)(9)(10)(11)(12)(13). β-Arrestin recruitment was determined employing an assay based on enzyme complementation (DiscoveRx Pathhunter). HEK 293 cells stably expressing β-arrestin-2 tagged with the enzyme acceptor (EA) were transfected with ProLink-tagged D 2 R with (open blue circles) or without (filled grey circles) cotransfection of NTS 1 R. (a) When applied at equimolar concentration, quinpirole and NT (8)(9)(10)(11)(12)(13) are able to stimulate β-arrestin-2 recruitment in D 2 R (n=4) and D 2 R/NTS 1 R (n=6) expressing cells. Coexpression of NTS 1 R markedly increases the observed potency but also slightly the efficacy. (b) The bivalent ligand 2c bearing a phenylpiperazine scaffold elicits β-arrestin-2 recruitment only in cells coexpressing NTS 1 R (n=3). The bell-shaped dose-response curve indicates a switch from bivalent to monovalent binding at concentrations >1 µM. (c) The aminoindane-derived bivalent ligand 3c induces β-arrestin-2 recruitment in both cell types (n=4 for D 2 R, n=3 for D 2 R/NTS 1 R). Coexpression of NTS 1 R potentiates the effect and leads to a bell-shaped dose-response curve with a maximum effect at 300 nM. Data represent mean ± s.e.m. of n independent experiments, each performed in duplicate. Results were normalized to the maximum effect of quinpirole (100 % for D 2 R and D 2 R/NTS 1 R). Figure 14: β-Arrestin-2 recruitment in HEK 293 cells coexpressing a signalling incompetent D 2 R mutant (D80A) and NTS 1 R. β-Arrestin-2 recruitment was determined employing an assay based on enzyme complementation (DiscoveRx Pathhunter). HEK 293 cells stably expressing βarrestin-2 tagged with the enzyme acceptor (EA) were transfected with a signalling incompetent ProLink-tagged D 2 R mutant (D80A) and NTS 1 R (black circles). (a) Quinpirole is not able to induce significant recruitment of β-arrestin-2 when D 2 R is signalling incompetent (n=6). (b) NT(8-13) elicits βarrestin-2 recruitment at a similar potency compared to coexpression of wild type D 2 R/NTS 1 R (EC 50 5.9 ± 1.2 nM, n=9). (c) The bivalent ligand 2a with a 22-atom spacer displays a sigmoid dose-response curve with a maximum effect of 98 ± 4 % (EC 50 67 ± 23 nM, n=4). (d,e) Increasing the spacer length to 44 or 66 atoms converts the sigmoid curves into bell-shaped dose-response relationships, where the maximum effect is observed at a concentration of 300 nM. Maximum effects are significantly higher as observed for NT (8)(9)(10)(11)(12)(13), indicating a higher degree of heterodimerization (n=4 for 2b and 2c). (f) Similar to compound 2a, ligand 3a shows β-arrestin-2 recruitment to a similar extend as NT (8)(9)(10)(11)(12)(13), albeit at lower concentration (EC 50 190 ± 40 nM, n=4). (g, h) The bivalent ligands 3b and 3c bearing a D 2 R agonist pharmacophore and a spacer length of 44 and 66 atoms, respectively, display bell-shaped doseresponse curves (n=3 and 4 for 3b and 3c) similar to the type 2 ligands 2b and 2c. Data represent mean ± s.e.m. of n independent experiments, each performed in duplicate. Responses were normalized to the maximum effect obtained with NT(8-13) (100 %). Figure 15: β-arrestin-2 recruitment in PAR 2 /NTS 1 R coexpressing HEK 293 cells. β-Arrestin-2 recruitment was determined employing the DiscoveRx Pathhunter assay. HEK 293 cells stably expressing β-arrestin-2 tagged with the enzyme acceptor (EA) were transfected with ProLinktagged PAR 2 and wild type NTS 1 R. (a) The dopamine receptor agonist quinpirole (orange circles) does not cause any β-arrestin-2 recruitment compared to the PAR 2 agonist f-LIGRLO-NH 2 (open black circles). (b-d) NT (8)(9)(10)(11)(12)(13)    A summary of crystal structures of GPCRs that have been solved as dimers. * Heterodimer models were built as described in the methods section. Instead of the β1-AR dimer, the crystal structure in question was used as template for structural alignment. Constructed dimer models were assessed after following three criteria for its suitability to function as template for modelling of a D2R and NTS1R dimer. † Protomer clashes were rated by excess of clashes between protomer 1 and protomer 2. Clash of side-chain atoms (-), clash of side-chain with backbone-atoms (--), clash of backbone atoms (---). ‡ For the distance rating between both protomers we looked at interactions along the dimer interface, interactions along less than half of interface (-), interaction mainly mediated by membrane lipids or cholesterol (--). § The parallelism of both protomers was used to rate the possibility of the dimer model to exist in a membrane environment in this orientation. Therefore we looked at the parallelism of both helix bundles as well as the individual receptor position relative to a potential membrane by measuring the angles between both protomers (> 30° (-)) and between helices eight and a theoretical membrane plane (25° < x ≤ 35° (-), > 35° (--)).  (2) 24 ± 4 (2)  (2) Receptor binding affinities were determined by radioligand displacement studies performed with the indicated radioligands as described in the supplementary methods section. a Mean values ± standard deviation (s.d.) are derived from (n) individual experiments, each performed in triplicate. b radioligand [ 3 H]NT(8-13).

Supplementary
a Effect = Emax relative to the effect of 10 µM forskolin (=100 %) and the unstimulated effect by buffer (= 0 %) subtracted by the effect of 10 µM forskolin. Negative values reflect the activation of a Gi/o system leading to the reduction of cAMP concentration. Positive numbers reflect the activation of a Gs system leading to an additional increase of cAMP. Data represent mean ± s.e.m. derived from (n) individual experiments each performed in triplicate. n.d.: no effect determined.
Supplementary Note 1: Selection of the dimeric β 1 -AR crystal structure as a scaffold for the

generation of the D 2 R/NTS 1 R heterodimer models
Aiming to build a reliable dopamine D 2 receptor/neurotensin receptor 1 (D 2 R/NTS 1 R) heterodimer model, we planned to arrange both receptors in a way that is compatible with experimental studies on direct interactions of the individual receptors. As, to our knowledge, experimentally derived information providing a molecular basis for the architecture of a D 2 R/NTS 1 R heterodimer has not been described, we focused on studies of D 2 Rs in homodimers and in heterodimers with other GPCRs (Supplementary  Table 2). We created models of D 2 R/NTS 1 R heterodimers by structurally aligning the NTS 1 R crystal structure (PDB-ID 4BUO) 16 and our recently described D 2 R homology model 1 with the crystal structures of all 18 GPCR dimers. Models were not considered further if they showed substantial clashes between the two receptors, as well as models revealing a high distance between the protomers or showing a low parallelism of the two protomers.
Taking into account this evaluation and showing relatively high sequence similarity with D 2 R, a crystal structure of the β 1 adrenergic receptor (PDB-ID 4GPO) 17 was used as a template, featuring the above described interface involving TM1 and helix 8. This architecture can also be considered biologically relevant because the interface was verified by in vitro cross-linking studies. The crystal displayed a second crystallographic asymmetric unit featuring a dimer with an interface between TM4 and 5. The respective dimer, however, did not allow formation of a stable GPCR-G protein complex, when the β 2 -AR active-state crystal structure was used as a template 17 . Taking further into account our previous studies on D 2 R homodimers 7 , in which the interface was suggested to be formed via TM1 and helix 8, we selected the heterodimer that showed an interface between TM1 and H8 of each protomer. Thus, superimposing our D 2 R homology model and the NTS 1 R crystal structure with the β 1 -AR dimer template yielded a D 2 R/NTS 1 R heterodimer model (Supplementary Fig. 1) that we used for the structure guideddesign of heterobivalent ligands.

Supplementary Note 2: Structure-guided design
Our heterobivalent ligands comprise a D 2 R pharmacophore (eticlopride, a phenylpiperazine or an aminoindane-type scaffold) coupled to an affinity-generating lipophilic appendage (consisting of a biphenyltriazole moiety) 18,19 . The appendage adopting an extension of the binding pocket directs to the extracellular region. The appendage is connected to the NTS 1 R agonist NT(8-13) via a flexible spacer of varying length to enable bridging of the binding pockets of D 2 R and NTS 1 R. As a spacer element, we used ω-amino-acid functionalized PEG-units. Possible attachment points of eticlopride and NT (8)(9)(10)(11)(12)(13) were identified using the crystal structures of D 3 R (which shows high homology to D 2 R) and NTS 1  of about 55 Å. We concluded that at least two spacer units should be necessary to enable a bivalent binding-mode, while a structure with only one spacer unit should lack the ability to bridge both binding sites ( Supplementary Fig. 2b). To check for an optimum spacer length, we considered compounds 1b, 1c and 1d bearing two, three and four units of the functionalized PEG-spacer, respectively. 1a with one linker unit should be used as a control compound.
Attachment points for the phenylpiperazine and aminoindane-type scaffolds were identified based on docking studies ( Supplementary Fig. 3). In both cases the basic nitrogen was selected to link the pharmacophore with the lipophilic appendage. In analogy to the structure-based strategy above, we designed compounds 2a-d (phenylpiperazine-type) and 3a-d (aminoindane-type) with one, two, three and four PEG-units, respectively. complexes were submitted to energy minimization using the SANDER module of AMBER10 as described 1 . The all-atom force field ff99SB 21 was used for standard amino-acid residues. For the bivalent ligands 2b and 3b, we used ff99SB for the NT(8-13)-part of the ligand and the general AMBER force field (GAFF) 22 for the linker and D 2 R pharmacophores. Charges for the non-peptidic parts were calculated using Gaussian09 23 at HF/6-31G(d) level and a manually conducted RESP 24  Eticlopride carries an aromatic chlorine atom, which shows an anisotropic charge distribution resulting in a σ-hole, a small positive area along the C-Cl axis and opposite of C 25 . Similar to hydrogens in a hydrogen bond, this area can interact with negative interaction sites 26 . As shown by Rendine et al. 27 , it is advantageous to consider halogen bonding for Molecular Dynamics simulations with halogenated molecules to obtain correct binding positions. To include the σ-hole in our MD calculations on compound 1b, we added a single point charge (a pseudo atom) along the C-Cl axis of the eticlopride pharmacophore and included the pseudo atom in the RESP fit procedure, in analogy to a method previously suggested by several groups [27][28][29] . Subsequently we applied the same procedures as described for compounds 2b and 3b.
For the membrane simulations, the AMBER topology and coordinate files for the minimized complex were converted into GROMACS 30, 31 input files. The ligand−bound dimer complex was inserted into a pre-equilibrated membrane of dioleoylphosphatidylcholine (DOPC) lipids by means of the GROMACS tool g_membed 32 . A pre-equilibrated system bearing a hydrated membrane with 72 DOPC lipids 33 was used as a starting point, which had to be enlarged in the x, y and z dimensions as described earlier 34 to fully surround the dimer model. water molecules) for the system with compound 3b. Within the membrane simulations, GAFF was used for DOPC molecules and the force field ff99SB for the protein residues. Parameters for compounds 1b, 2b and 3b were used as described above. The SPC/E water model 35

Determination of binding affinities for a panel of D 2 -like receptors and the receptor subtypes NTS 1 R and NTS 2 R
Receptor binding studies were carried out as described previously 37

General materials and methods for organic synthesis
Reagents and dry solvents were obtained from commercial sources and were used as received.

Syntheses of bivalent ligands
The synthesis was performed according to standard protocols as described below and was performed concentrating the combined organic phases in vacuo yielded a semi-solid crude which was extracted with CH 2 Cl 2 under sonication (3x). Filtration and concentration of the combined extracts yielded a crude oil that was used in the next reaction without further purification.