Angiotensin II type 1/adenosine A2A receptor oligomers: a novel target for tardive dyskinesia

Tardive dyskinesia (TD) is a serious motor side effect that may appear after long-term treatment with neuroleptics and mostly mediated by dopamine D2 receptors (D2Rs). Striatal D2R functioning may be finely regulated by either adenosine A2A receptor (A2AR) or angiotensin receptor type 1 (AT1R) through putative receptor heteromers. Here, we examined whether A2AR and AT1R may oligomerize in the striatum to synergistically modulate dopaminergic transmission. First, by using bioluminescence resonance energy transfer, we demonstrated a physical AT1R-A2AR interaction in cultured cells. Interestingly, by protein-protein docking and molecular dynamics simulations, we described that a stable heterotetrameric interaction may exist between AT1R and A2AR bound to antagonists (i.e. losartan and istradefylline, respectively). Accordingly, we subsequently ascertained the existence of AT1R/A2AR heteromers in the striatum by proximity ligation in situ assay. Finally, we took advantage of a TD animal model, namely the reserpine-induced vacuous chewing movement (VCM), to evaluate a novel multimodal pharmacological TD treatment approach based on targeting the AT1R/A2AR complex. Thus, reserpinized mice were co-treated with sub-effective losartan and istradefylline doses, which prompted a synergistic reduction in VCM. Overall, our results demonstrated the existence of striatal AT1R/A2AR oligomers with potential usefulness for the therapeutic management of TD.

postulated that brain RAS may be involved in dopaminergic degeneration, especially when the dopaminergic system is impaired, thus contributing to the pathogenesis and progression of dopaminergic-related pathologies such as Parkinson's disease (PD).
The concept that cell surface receptors may physically interact forming oligomers appeared early in the eighties, while characterizing G protein-coupled receptors (GPCRs) for neurotransmitters 11,12 . Notably, striatal dopaminergic receptors in general, and the dopamine D 2 receptor (D 2 R) in particular, constitute the archetypal GPCR capable of forming receptor-receptor complexes. Indeed, the potential impact of these oligomers in pathophysiological conditions involving dopaminergic dysfunction has been extensively studied. Interestingly, the D 2 R has been shown to oligomerize with several GPCRs 13 , including the adenosine A 2A receptor (A 2A R) 14 . The D 2 R-A 2A R heteromer is expressed in GABAergic striatopallidal neurons and a reciprocal negative allosteric receptor-receptor interaction is defined as its "biochemical fingerprint" 15 . Noteworthy, the D 2 R-A 2A R heteromer has been defined as a potential pharmacological target for pathologies associated with dysfunctional dopaminergic signaling, such as PD and schizophrenia. Indeed, A 2A R antagonists (i.e. istradefylline) are currently used for PD treatment in Japan 16 . On the other hand, the D 2 R has also been shown to oligomerize with the AT 1 R in the striatum 17 , thus the potential use of AT 1 R ligands to modulate dopaminergic signaling has been postulated. Interestingly, early studies also indicated interactions between the adenosinergic and the angiotensinergic systems, for instance the antinociceptive effect of AII was related to that produced by adenosine A 1 receptor agonists 18 . In addition, an A 2A R-and AT 1 R-mediated synergistic interaction in the peripheral RAS was described 19,20 . Thus, while adenosine was able to reverse the stimulatory effect of AII on Na + -ATPase activity in the renal proximal tubules via A 2A R activation 21 , A 2A R blockers reduced AII-mediated ROS formation via Nox2 (NADPH complex enzyme) in endothelial cells 20 . Conversely, AII potentiated the adenosine-induced contraction of afferent arterioles 22 , while losartan-mediated AT 1 R blockade abolished the adenosine-mediated reflex sympatho-excitatory response in the brachial artery 23 . Altogether, the aforementioned evidence highlights the need for a better understanding of the adenosinergic system-RAS interaction. Furthermore, this interaction may be relevant not only in the periphery but also in the brain, where a functional interplay with the dopaminergic system may occur.
Here, we study the possible existence, both in cultured cells and in mouse striatum, of a physical AT 1 R-A 2A R interaction, which may be a potential target for managing dopaminergic-related disorders (i.e. tardive dyskinesia, TD). Also, we seek to characterize the most likely heteromeric receptor arrangement through protein-protein docking and long-timescale molecular dynamics (MD) simulations. Finally, we propose a novel multimodal treatment for TD based on the use of AT 1 R and A 2A R antagonists at sub-effective doses, and test it in a mouse TD model, namely the reserpine-induced vacuous chewing movement (VCM).

AT 1 R and A 2A R form heteromers in cultured cells.
Based on the existence of AT 1 R/D 2 R heteromers 17 , we aimed to elucidate whether AT 1 R is also able to oligomerize with the A 2A R, a well-known D 2 R partner 24 . To this end, we first assessed the co-distribution of AT 1 R and A 2A R in cultured cells through the fluorescence detection of CFP/YFP tagged receptors. Thus, by means of confocal microscopy analysis of HEK-293T cells transiently expressing AT 1 R CFP and A 2A R YFP , a high overlapping in the distribution of the former receptors was observed (Fig. 1a). Next, we examined the possible physical interaction of AT 1 R and A 2A R in living cells by means of the BRET approach. Thus, cells were transiently transfected with receptor constructs carrying the appropriate fluorophore pairs (A 2A R Rluc and AT 1 R YFP ). A positive and saturable BRET signal was observed in cells co-transfected with a constant concentration of the A 2A R Rluc and increasing concentrations of AT 1 R YFP (Fig. 1b). Of note, as the control pair GABA B2 R Rluc and AT 1 R YFP led to a low and linear distribution, the specificity of the saturation (hyperbolic) assay for the A 2A R Rluc and AT 1 R YFP pair could be established (Fig. 1b). Overall, these results demonstrate that AT 1 R and A 2A R form heteromers in living HEK-293T cells.
Structure of AT 1 R/A 2A R heteromer. Computational modeling, protein-protein docking, and MD simulations were used to probe the interaction between AT 1 R and A 2A R, and determine their most likely heteromeric arrangement. Initially, AT 1 R and A 2A R antagonists (losartan and istradefylline, respectively), were docked into their respective inactive-state receptor crystal structure using Autodock4.2 25 . The corresponding best docked AT 1 R-losartan and A 2A R-istradefylline complexes were then embedded in lipid bilayer membranes and subjected to MD simulations of 250 ns and 500 ns, respectively, where both bound antagonists were observed to stabilize. In particular, in AT 1 R, ARG167, located on extracellular loop 2 (ECL2) above the orthosteric pocket, was observed to make H-bonds with losartan at both ends of the ligand (see SI Fig. 1) in a similar manner to that observed in the AT 1 R crystal structure containing bound olmesartan 26 . Likewise, in A 2A R, ASN253 (ASN 6.55 in Ballesteros-Weinstein numbering 27 ) made an H-bond with istradefylline in a similar manner to other co-crystallized A 2A R xanthine antagonists 28 (Fig. S1).
As both AT 1 R and A 2A R are thought to form functional homodimers at the cell surface [29][30][31][32][33][34][35] , we investigated the likely structure and behavior of these respective homodimers with bound antagonists, prior to investigating heteromeric interactions. In order to do this we utilized the A 2A R homodimer crystal structure with co-crystallized antagonist 36 as a structural guide for initializing AT 1 R-losartan and A 2A R-istradefylline homodimer models. This dimeric crystal structure is observed to contain an interface between TM4 and TM5 helices of each monomer, with TM4 of one monomer interacting with TM5 of the other, and vice versa 36 . Initial AT 1 R and A 2A R homodimer models were refined with protein-protein docking using the ROSIE webserver 37 , each consisting of two antagonist-bound receptors in the same MD-generated conformation (see above). Following protein-protein docking, the A 2A R and AT 1 R homodimers were subjected to further MD simulations of 1.5 μs and 750 ns, respectively. During these simulations, both AT 1 R and A 2A R homodimers were seen to form significant interactions via their TM4 and TM5 helices, respectively, with considerable contact between monomers, indicative of energetically stable dimers (Fig. S2). In addition, the respective bound antagonists remained stably bound in Scientific RepoRts | 7: 1857 | DOI:10.1038/s41598-017-02037-z each participating monomer, with all receptor subunits maintaining an inactive state. From these results, it was inferred that the antagonist-bound homodimeric states of AT 1 R and A 2A R are stable in silico, and likely form the minimum constituents that participate in cross-receptor heteromeric interaction.
As other described heteromeric interactions involving A 2A R fit a heterotetramer model 38,39 , and as MD simulations of AT 1 R and A 2A R homodimers suggest their respective stability, we investigated heterotetrameric interactions between the two receptor homodimers. As there is no crystal structure for GPCRs in tetrameric formation, we performed extensive protein-protein docking with ROSIE to identify the highest possible scoring interaction of AT 1 R and A 2A R homodimers (see Methods). The "best" conformation identified a tetramer with cross-receptor interfaces involving TM5 and TM6 of one receptor with TM1 and TM2 of the other, and vice versa (Fig. 2). In order to assess the stability of the proposed interaction, the heterotetramer complex was subjected to an MD simulation in a membrane for 2 µs. Results show the receptors progressively stabilized (RMSD curve in Fig. S3) and enhanced their interaction, whilst maintaining the original tetrameric configuration (Fig. 2). Furthermore, the respective AT 1 R and A 2A R homodimers remained stable and unperturbed within the tetramer, maintaining their respective inactive states. In conclusion, stable heterotetrameric interaction between AT 1 R and A 2A R is plausible at a molecular level and compatible with bound antagonists, losartan and istradefylline.
Functional consequences of the AT 1 R and A 2A R oligomerization. The formation of AT 1 R-A 2A R complexes in transfected cells suggests that there might exist a functional coupling between these two receptors. Thus, we assessed the impact of A 2A R expression on AT 1 R-mediated intracellular Ca 2+ mobilization from internal stores by means of Fluo4 determinations. Thus, in Fluo4 loaded cells expressing AT 1 R alone, the activation with angiotensin II increased intracellular Ca 2+ (Fig. 3a, red trace), as expected. Interestingly, in cells co-expressing AT 1 R and A 2A R, the angiotensin II-mediated intracellular Ca 2+ mobilization was boosted (Fig. 3a, blue trace). Indeed, in cells expressing only A 2A R, a residual and not significant effect of angiotensin II was observed, probably because of the endogenous expression of AT 1 R in HEK-293T cells (Fig. 3a, black trace). Quantification of the results (Fig. 3c) demonstrated a significant [F (2,6) = 8.40 (P < 0.05)] difference between the experimental groups assessed, thus a significant (P < 0.05) increase in the AT 1 R-mediated intracellular calcium accumulation in AT 1 R-A 2A R cells was observed (Fig. 3c). These results suggest that a functional interplay between AT 1 R and A 2A R might exist upon expression in heterologous cells.

AT 1 R and A 2A R heteromers are expressed in mouse striatum. Once demonstrated that AT 1 R and
A 2A R assemble into functionally interacting complexes in living cells, we aimed to determine the existence of AT 1 R/A 2A R heteromers in native tissue, namely the striatum. To this end, we first conducted immunofluorescence experiments to assess the expression levels and distribution of both AT 1 R and A 2A R in mouse striatum. Interestingly, both receptors showed a high degree of co-distribution throughout the striatal neuropil (Fig. 4a, upper panels) and eventually within the medium spiny neurons (MSN) cell bodies (Fig. 4a, lower panels). Importantly, the myelinated fiber bundles that penetrate the striatum were visible as dark (not stained) structures within the stained neuropil (Fig. 4a, upper panel). These results give rise to the possibility that these two receptors might be forming heteromers under native conditions. Subsequently, to confirm the existence of AT 1 R/A 2A R heteromers in the striatum we implemented the P-LISA approach, a well described technique providing enough  sensitivity to evaluate receptor's close proximity within a named GPCR oligomer in native conditions 40 . Thus, by using proper antibody combinations, the AT 1 R/A 2A R heteromer expression in mouse striatum was addressed by P-LISA assays. Indeed, red dots reflecting a positive P-LISA signal was observed in the striatum of wild-type mice (Fig. 4b), thus allowing the visualization of the AT 1 R/A 2A R receptor-receptor interaction. Interestingly, in striatal slices from the A 2A R-KO mice the P-LISA signal was negligible (Fig. 4b), thus reinforcing the specificity of our P-LISA assay. Indeed, when the P-LISA signal was quantified the wild-type animal showed 4 ± 0.5 dots/nuclei while the A 2A R-KO displayed only 1 ± 0.2 dots/nuclei under the same experimental conditions. Thus, a marked and significant (P < 0.005) reduction in the P-LISA signal was observed in the A 2A R-KO striatal slices. Taken together, data gathered from our P-LISA experiments strongly support the existence of AT 1 R/A 2A R heteromers in the mouse striatum.  and pathological conditions 42,43 . However, although A 2A R has been linked to neuroleptic-induced TD 44,45 , its impact on this syndrome is still ambiguous 46 . Consequently, we sought to investigate whether the AT 1 R/A 2A R heteromer might play a role in TD. We took advantage of the vacuous chewing movement (VCM) model of TD in mice. Interestingly, administration of the AT 1 R antagonist losartan dose-dependently reduced reserpine-induced VCM (Fig. 5a). Similarly, administration of the A 2A R antagonist istradefylline dose-dependently reduced reserpine-induced VCM (Fig. 5b). Subsequently, we investigated whether co-treatment at sub-effective low doses of AT 1 R and A 2A R antagonists would elicit a significant reduction of VCM in our reserpine-induced TD animal model. Therefore, for combination treatment, 0.05 mg/kg of losartan and 0.03 mg/kg of istradefylline were selected as they were not effective in reducing VCM. Noteworthy, the combined treatment produced a significant (P < 0.05) reduction in VCM (Fig. 5c), thus demonstrating a synergistic interaction between both drugs. Overall, these results suggest that co-treatment with AT 1 R and A 2A R antagonists at sub-effective low doses is a useful therapeutic approach for TD management.
Finally, in an attempt to ascertain the role of AT 1 R/A 2A R oligomers in the synergistic effect observed upon receptor antagonist co-treatment, we assessed the efficacy of the VCM sub-effective losartan dose in mice lacking the A 2A R (i.e. A 2A R-KO mice). Interestingly, the low dose of losartan (0.05 mg/kg) was able to significantly (P < 0.05) reduce the number of VCM in the A 2A R-KO mice (Fig. 5d). Hence, in the absence of A 2A R the efficacy of losartan was higher, thus indicating that AT 1 R/A 2A R heteromers are crucial for finely modulating TD. Collectively, these results suggest that AT 1 R and A 2A R functionally interact in vivo and that this functional interplay may be provided by the existence of AT 1 R/A 2A R oligomers.

Discussion
TD is a serious motor side effect associated to long-term treatment with neuroleptics 47 . Notably, D 2 R occupancy and its transience to occupation have been identified as a potential mechanistic substrate to develop antipsychotic-induced TD 48 . Indeed, D 2 R-mediated control of motor function has been related to the ability of this receptor to oligomerize with other GPCRs in general 49 and with the A 2A R in particular 42,43,50 . Also, in the brain, dopaminergic neurotransmission can be modulated by AII through AT 1 R. Thus, AT 1 R blocking precludes AII-mediated dopamine release 51,52 . Furthermore, a functional interaction between angiotensin and dopamine receptors in the striatum and substantia nigra 53,54 , together with the formation of D 2 R and AT 1 R heteromers in the striatum has been described 17 . Based on these data, we decided to explore a possible direct interaction between AT 1 R and A 2A R, and revealed for the first time the existence of AT 1 R/A 2A R oligomers in the striatum and its implications in TD.
Our experimental data shows that AT 1 R and A 2A R form heteromers both in co-transfected cells and in mouse striatum. This feature is especially strengthened by our in-silico analysis, which has predicted a heterotetrameric receptor arrangement that was stable during 2 μs of MD simulation. The "best" receptor-receptor interface identified for the AT 1 R/A 2A R heterotetramer involves TM5 and TM6 of one receptor with TM1 and TM2 of the other, and vice versa, while in the respective homodimers the TM4 of one monomer interactac with TM5 of the other, and vice versa. Interestingly, the D 2 R/A 2A R heterodimeric interface has been postulated to be formed by the TM4 and TM5 of D 2 R interacting with TM4 and TM5 of the A 2A R 49,55 . Therefore, when considering a putative AT 1 R/D 2 R/A 2A R oligomer new in-silico analysis will be needed to accurately determine TM-TM contacts and receptor rearrangement defining AT 1 R/D 2 R/A 2A R oligomer stoichiometry. Overall, this information will be extremely valuable when assessing potential multimodal TD pharmacotherapeutic interventions based on drugs targeting these receptors.
The AT 1 R/A 2A R oligomerization was shown to elicit functional consequences, since co-expression with A 2A R boosted AT 1 R signaling. This AT 1 R gain of function may most likely result from an A 2A R-mediated AT 1 R increased cell surface targeting, as was previously reported 56 . Alternatively, an A 2A R-mediated direct trans-activation of AT 1 R could not be excluded, as has been described for other A 2A R-containing oligomers 41 . Thus, further work is needed to elucidate the precise molecular mechanism behind this AT 1 R/A 2A R oligomer-dependent AT 1 R gain of function. Nevertheless, our main purpose consisted of ascertaining the in vivo implications of the AT 1 R/A 2A R oligomer formation, which is the cornerstone when describing a new GPCR oligomer 57 . Indeed, our P-LISA data strongly supported the existence of AT 1 R/A 2A R heteromers in the mouse striatum, thus warranting the need to assess the impact of this oligomer in behaving animals. Accordingly, we demonstrated an unprecedented synergism of AT 1 R and A 2A R antagonists on the control of involuntary mandibular movements induced by reserpine in an animal model of TD. Thus, co-treatment with AT 1 R and A 2A R antagonists at sub-effective low doses robustly (>60%) reduced reserpine-mediated VCM. Certainly, this makes this multimodal pharmacological approach an attractive solution for TD management.
The striatum is considered a pivotal brain region, since it receives projections from other basal ganglia areas and from many other brain regions involved in motor and non-motor functions, such as the motor cortex, the prefrontal cortex and the hippocampus 58,59 . Indeed, both the renin-angiotensin and the adenosinergic systems play an important role in controlling the striatal function. Thus, the ability of AT 1 R and A 2A R to heteromerize in the striatum might constitute a way of fine-tuning multiple receptor-signaling pathways harmonizing dopaminergic neurotransmission. Therefore, the AT 1 R/A 2A R oligomer could be envisaged as a potential drug target for striatum-related adverse motor dysfunctions associated to therapy, including TD and L-DOPA induced dyskinesia (LID). Indeed, A 2A R antagonists have been postulated and licensed as antiparkinsonian drugs 60 and eventually studied in the management of LID 61 . Furthermore, A 2A R has been linked to neuroleptic-induced TD 44,45 , although with some debate 46 .
Similarly, preclinical studies have demonstrated that blockade of AT 1 R reduces LID 62 . It is assumed that these A 2A Rand AT 1 R-mediated anti-LID effects are related to their ability to heteromerize with D 2 R 17, 24 and thus controlling dopaminergic neurotransmission. However, it could be speculated that AT 1 R and A 2A R might control D 2 R function through functional AT 1 R/D 2 R/A 2A R-containing complexes in GABAergic striatopallidal neurons. A number of facts support this last statement: i) the high and selective co-expression of AT 1 R, D 2 R and A 2A R in these particular cells; ii) the demonstration of A 2A R/D 2 R, AT 1 R/D 2 R and AT 1 R/A 2A R heteromers; and iii) the existence of strong multiple interactions between the three receptors. In conclusion, the demonstration of their simultaneous physical interaction may constitute a novel and very attractive target for developing new drugs in the management of pathologies in which these receptors play a key role, such as TD. Cell culture. Human embryonic kidney (HEK)-293T cells were grown in Dulbecco's modified Eagle's medium (DMEM) (Sigma-Aldrich) supplemented with 1 mM sodium pyruvate, 2 mM L-glutamine, 100 U/mL streptomycin, 100 mg/mL penicillin and 5% (v/v) fetal bovine serum at 37 °C and in an atmosphere of 5% CO 2 . HEK-293T cells growing in 25 cm 2 flasks or six-well plates containing 18 mm coverslips were used for western blot and fluorescence imaging, respectively. They were transiently transfected with the cDNA encoding the specified proteins using Polyethylenimine (Polysciences, Inc. Warrington, PA, USA). Fixed brain tissue preparation. Mice were anesthetized and perfused intracardially with 100-200 ml ice-cold 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS; 8.07 mM Na 2 HPO 4 , 1.47 mM KH 2 PO 4 , 137 mM NaCl, 0.27 mM KCl, pH 7.2). Brains were post-fixed in the same solution of PFA at 4 °C during 12 h. Coronal sections (25 μm) were processed using a vibratome (Leica Lasertechnik GmbH, Heidelberg, Germany). Slices were collected in Walter's Antifreezing solution (30% glycerol, 30% ethylene glycol in PBS, pH 7.2) and kept at −20 °C until processing. Bioluminescence resonance energy transfer measurements. Bioluminescence resonance energy transfer (BRET) experiments in HEK-293T cells were performed as previously described 66  Immunohistochemistry. Previously collected slices were washed three times in PBS, permeabilized with 0.3% Triton X-100 in PBS for 2 hours and rinsed back three times more with wash solution (0.05% Triton X-100 in PBS). The slices were then incubated with blocking solution (10% NDS in wash solution; Jackson ImmunoResearch Laboratories, Inc., West Grove, PA, USA) for 2 h at R.T. and subsequently incubated with the primary antibodies overnight at 4 °C. After two rinses (10 min each) with 1% NDS in wash solution, sections were incubated for 2 h at R.T. with the appropriate secondary antibodies conjugated with Alexa dyes (Invitrogen, Carlsbad, CA, USA), then washed (10 min each) two times with 1% NDS in wash solution and two more times with PBS, and mounted on slides. Fluorescence striatal images were obtained using a Leica TCS 4D confocal scanning laser microscope (Leica Lasertechnik GmbH).

Reagents
Proximity ligation in situ assay. Duolink in situ PLA detection Kit (Olink Bioscience, Uppsala, Sweden) was performed in a similar manner as immunohistochemistry explained above until the secondary antibody incubation step. The following steps were performed following the manufacturer's protocol, as previously described 24,40 . Fluorescence images were acquired on a Leica TCS 4D confocal scanning laser microscope (Leica Lasertechnik GmbH) using a 60x N.A. =1.42 oil objective from the selected area. High-resolution images were acquired as a z-stack with a 0.2 μm z-interval with a total thick of 5 μm. Nonspecific nuclear signal was eliminated from PLA images by substracting DAPI labeling. Analyze particle function from Image J (NIH) was used to count particles larger than 0.3 μm 2 for PLA signal and larger than 100 μm 2 to discriminate neuronal from glia nuclei 68 . For each image a number of oligomer particles and neuron nuclei was obtained and ratio among them was calculated.
Computational modeling. Ligand docking. Crystal structures of AT 1 R (PDB id: 4ZUD) and A 2A R (PDB id: 4EIY) were converted into apo wt forms by removing co-crystallized ligands and non-native fusion proteins i.e. cytochrome b562, building missing intracellular and extracellular loop sections with MODELLER v9.14 69 , and energy minimizing in the AMBER14SB force-field 70 with CHIMERA v1.10.2 71 . The AT 1 R antagonist losartan and A 2A R antagonist istradefylline were downloaded from PubChem 72 , energy-minimized in AMBER14SB force-field with CHIMERA and docked into respective receptor structures with Autodock4.2 25 using default parameters. Grid points were generated to cover total orthosteric pocket volumes. The final docking conformations of losartan and istradefylline represented top hits identified by best predicted affinity (nM). These were checked to be consistent with previously reported binding modes of relevant co-crystallized antagonists 26,28 . In particular, losartan was docked to interact with Arg167 and istradefylline was docked to interact with Asn253. Subsequent energy minimization of docked structures was performed with CHIMERA in the AMBER-14SB force-field.
Protein-protein docking. For generating homodimers of respective receptors: AT 1 R and A 2A R with bound antagonists, two molecular dynamics (MD)-generated receptor-ligand monomers (see MD methods) of either AT 1 R or A 2A R, in each case, were superimposed onto the A 2A R homodimer crystal structure (PDB id: 4EIY), yielding an initial homodimer model, which was then submitted to the ROSIE webserver 37 for protein-protein docking. For both AT 1 R and A 2A R, the best docked homodimer was identified by three factors: best possible ROSETTA interface score (I_sc), lowest possible RMSD in relation to initial model, and acceptable membrane-compatible orientation. For construction of an AT 1 R-A 2A R heterotetramer, two initial tetrameric arrangements were manually generated by combining respective MD-generated AT 1 R and A 2A R homodimers (see MD section) in alternative ways: (i) where homodimers are arranged side-to-side in a rectangular-like configuration, where each homodimer subunit interacts with a subunit of the other homodimer (by respective TM1/2-5/6 helices), (ii) where homodimers are partially displaced with respect to one another creating a parallelogram-like configuration, where both subunits of one homodimer interact with a single subunit of the other homodimer (by respective TM4/5 helices). Both these alternative configurations were submitted to the ROSIE webserver for identification of the best possible tetrameric arrangement according to the same criteria implemented previously. For all protein-protein docking runs executed on the ROSIE webserver, default local parameters were used, i.e. perturbation of 3 Å between proteins, 8° of tilt, and 360° rotation around protein centers, with generation of 1000 docking solutions per case.
Molecular dynamics system setup. Five different systems were generated using the CHARMM-GUI web-based interface 73 , each in a POPC membrane and solvated with TIP3P water molecules: AT 1 R monomer with bound losartan, A 2A R monomer with bound istradefylline, AT 1 R homodimer with bound losartan, A 2A R homodimer with bound istradefylline, and AT 1 R-A 2A R heterotetramer with bound antagonists. All receptor structures were orientated according to the OPM database 74 entry: 4eiy. Charge neutralizing ions (0.15 M KCl) were introduced to each system. Parameters of membrane, water and protein were automatically generated by CHARMM-GUI 73 according to CHARMM36 force-field 75 with ligand parameters automatically generated according to CHARMM36 General Force Field [76][77][78] .
Molecular dynamics simulations. Molecular dynamics (MD) simulations of AT 1 R and A 2A R were performed using the CHARMM36 force-field 75 with ACEMD 79 on specialized GPU-computer hardware, totaling 5 μs across systems. In detail, monomer AT 1 R/A 2A R systems were equilibrated for 20 ns at 300 K and 1 atm, while AT 1 R/A 2A R homodimers and heterotetramer systems were equilibrated for 50 ns under same conditions. During equilibration, positional harmonic restraints on protein and antagonist heavy atoms were progressively released over the first 8 ns and then continued without constraints. After equilibration, AT 1 R and A 2A R monomers were subjected to unbiased production runs of 250 ns and 500 ns under same conditions, respectively. Likewise, AT 1 R and A 2A R homodimers were subjected to unbiased production runs of 750 ns and 1.5 μs, respectively. The AT 1 R/A 2A R heterotetramer was subjected to an unbiased production run of 2 μs. Simulation trajectories were analyzed using VMD software v1.9.2 80 .
Reserpine-induced vacuous chewing movements. The VCM model of TD 48 was induced in mice through two subcutaneous (s.c.) reserpine injections (1 mg/kg) administered with an interval of 48 h. Twenty-four hours after the last reserpine administration, mice were treated by intraperitoneal (i.p.) route with losartan (0.05-50 mg/kg) and/or istradefylline (0.03-0.06 mg/kg). VCM parameters were evaluated as previously described 81 but with some modifications. Thus, the evaluation of VCM frequency consists of a manual counting of continuous single mouth openings in a vertical plane, not directed to a physical material. Mirrors were placed on the table and behind the glass cylinder (Ø 19 cm and 22 cm height) to allow observation of the orofacial movements when mice were not facing the observer. The evaluation of this parameter during 10 min was performed by a blind observer, 30 min after the pharmacological treatments administered 24 h after the second reserpine injection 81 .