Inactive structures of the vasopressin V2 receptor reveal distinct binding modes for Tolvaptan and Mambaquaretin toxin

Abstract

Inhibitors of the arginine-vasopressin (AVP) V2 receptor (V2R) are key therapeutic compounds for treating hyponatremia or polycystic kidney diseases. Rational drug design based on experimental G protein-coupled receptor structures is a powerful avenue to develop better drugs. So far, the lack of inhibitor-bound V2R structures has impaired this strategy. Here we describe the cryo-electron microscopy structures of the V2R in complex with two selective inverse agonists, the non-peptide Tolvaptan (TVP) and the green mamba snake Mambaquaretin toxin (MQ1). Both ligands bind into the orthosteric binding site but with substantial differences. TVP binds deeper than MQ1, and directly contacts the toggle switch residue W284^6.48 in the transmembrane domain 6. The Kunitz-fold toxin displays extensive contacts with extracellular and transmembrane residues. As anticipated from TVP and MQ1 pharmacological properties, both structures represent inactive V2R conformations. Their comparison with those of the active AVP-bound V2R reveals the molecular mechanisms modulating receptor activity. The mini-protein MQ1-bound V2R structure suggests a new pharmacology approach for treating water homeostasis and renal diseases.

To help in the rational design of new compounds for treating hyponatremia and polycystic kidney disease, the authors solve cryo-EM structures of the G protein-coupled vasopressin type 2 receptor in its inactive state, bound to a small molecule or to a mini-protein from snake venom.

Introduction

The arginine-vasopressin (AVP) V2 receptor subtype (V2R) is a typical G protein-coupled receptor (GPCR) responsible for the antidiuretic physiological function of AVP. As such, V2R is a validated therapeutic target in water balance disorders and renal pathologies¹. In particular, V2R antagonists or inverse agonists are used for treating hyponatremia as a consequence of congestive heart failure, hypertension, hepatic cirrhosis, syndrome of inappropriate antidiuretic hormone secretion². The small non-peptide reference compound Tolvaptan (TVP)^3,4 is indicated in short-term treatment of clinically significant hypervolemic or euvolemic hyponatremia (serum sodium <135 mEq/L). It works as an aquaretic compound by competing with AVP and thus increasing the amount of urine volume. At the cellular level, TVP binds to V2R with high affinity and moderate selectivity to inhibit V2R-dependent signaling pathways such as Gs-induced cyclic adenosine monophosphate (cAMP) production, recruitment of β-arrestins, or MAP kinase phosphorylation^5,6. It is characterized as a V2R inverse agonist toward the Gs/cAMP pathway^7,8. In addition, TVP has also been available for several years as a therapy to retard progression of cyst development and renal insufficiency in autosomal dominant polycystic kidney disease (ADPKD)^9–11, the most frequent Mendelian inherited disorder affecting millions of people worldwide. However, chronic TVP treatment can cause unwanted side effects like serious and potentially fatal liver injury^12,13.

Conivaptan is a small non-peptide antagonist of the V2R used for treating euvolemic hyponatremia in hospitalized patients with underlying heart failure¹⁴, but it is not selective to V2R. It also displays a high affinity for the related AVP receptor V1a subtype (V1aR)^15,16 involved in many physiological functions of the hormone, such as control of blood pressure, platelet aggregation, memory, and social behavior. It is thus a dual V1aR/V2R antagonist.

Recently, we discovered a selective V2R cysteine-rich ligand in the venom of the green mamba snake¹⁷. This mini-protein toxin was named Mambaquaretin (MQ1) based on its pharmacological properties and its related in vivo aquaretic effect. It is made up of 57 amino acids and is characterized by a Kunitz-fold with three disulfide bridges. It is the most selective V2R inhibitor ever described, displays a nanomolar affinity, and does not interact with the other AVP receptor subtypes V1aR, V1bR, the closely-related oxytocin receptor (OTR), or with other tested GPCRs (150 in total)¹⁷. Like TVP, it is described as a V2R inverse agonist for the Gs/cAMP pathway¹⁸. As TVP, MQ1 inhibits the main V2R-associated signaling pathways, namely coupling to Gs protein and adenylyl cyclase, as well as recruitment of β-arrestins. Injected in mice, MQ1 increases urine outflow in a dose-dependent manner with concomitant reduction of urine osmolality, indicating an aquaretic effect. MQ1 is a potential therapeutic candidate to be used in all pathologies associated with an overactivity of the V2R, such as hyponatremia or the genetic pathology ADPKD. It is validated in vivo in murine models of hyponatremia and PKD^17,18. For instance, pcy mice, a juvenile model of PKD, treated daily with MQ1 for a hundred days, developed less abundant and smaller cysts than control mice, with no tachyphylaxis and no apparent toxicity¹⁷.

Despite long-term efforts to determine the molecular basis of antagonist or inverse agonist binding to V2R, mainly through a combination of molecular modeling and site-directed mutagenesis^19,20, the mechanism of V2R recognition at a near atomic level is still unknown. Here, we filled this gap of knowledge by revealing the structural basis of both TVP and MQ1 binding to V2R. In particular, we determined the three-dimensional (3D) structures of the V2R in complex with either TVP or a high-affinity version of MQ1 (MQ1^K39A) by cryo-electron microscopy (cryo-EM). In agreement with their competitive pharmacological profile, the nonpeptide small molecule and the toxin both bind into the orthosteric binding pocket of AVP. However, as expected from their differences in molecular weight (0.45 versus 6.4 kDa), in charge content (TVP is highly hydrophobic whereas the MQ1^K39A contains many charged residues) and in physico-chemical properties (a small poly-aromatic compound versus a disulfide-rich protein containing a two-stranded antiparallel beta sheet and an alpha helix), they present substantial differences in their contacts with the receptor. The structures also reveal the architecture of the inactive state of V2R and its comparison with AVP-bound V2R active states^21–24 uncovers the changes occurring upon receptor activation. Our study also provides an original structural aspect of GPCR ligand binding, since we elucidate how MQ1, a mini-protein ligand with a Kunitz-fold, acts as a GPCR ligand.

Results

Structure determination of TVP-V2R and MQ1^K39A-V2R complexes

Despite significant advances in cryo-EM single-particle analysis for high-resolution protein structure determination, small proteins (<100 kDa) remain elusive for reconstruction due to low signal-to-noise and lack of distinctive structural features. This is particularly true for purified GPCRs embedded in detergent micelles with no clear characteristic features protruding from the 7TM bundle, making particle alignment and classification challenging. Strategies have thus been developed to artificially increase molecular weight of the target GPCRs, using insertion of fusion tags and/or attaching scaffolding assemblies²⁵. Moreover, introducing a fusion tag into the GPCR structure or adding a specific antibody fragment (Fab) serves as a fiducial marker, breaking micelle roundness, introducing asymmetry, and facilitating accurate particle alignment, thereby enhancing overall reconstruction quality²⁶. To increase mass and help in particle alignment, we inserted different fusion proteins into the V2R sequence between A234^5.67 and A264^6.28 (superscripts indicate Ballesteros-Weinstein numbering²⁷), replacing the intracellular loop 3 (ICL3). We attempted (i) the bacteriophage T4 lysozyme (T4L), (ii) the circularized permutated green fluorescent protein (cpGFP), or (iii) the thermostabilized apocytochrome b562 (BRIL) (Supplementary Fig. 1)²⁸. The V2R-BRIL fusion protein was further complexed with an anti-BRIL Fab and anti-Fab nanobody²⁹.

While these synthetic proteins were successfully purified and suitable for cryo-EM data collection (Supplementary Fig. 1), they did not have the potential for high-resolution structural characterization due to high dynamics between the V2R and the fusion protein. To overcome this problem, we employed Alphafold3 predictions to guide the design of two rigid helical linkers from the A_2A adenosine receptor (ARRQL from TM5 and RARSTL from TM6)³⁰ that we added at the two extremities of the BRIL domain (Supplementary Fig. 2). This allowed a continuity in helicity between V2R TM5/6 and the BRIL protein (Supplementary Figs. 1 and 2), ensuring a proper BRIL orientation for Fab binding while avoiding steric hindrance between the components of the complexes (Supplementary Fig. 1). Given that GPCRs are highly dynamic, a more rigid BRIL fusion might help in dampening the conformational flexibility of the receptor while preserving its structural features, including TM6 positioning²⁵. Indeed, as a hallmark feature of GPCRs, TM6 is kept inward in inactive conformations, keeping the TM bundle densely packed³¹. For instance, based on the comparison between the inactive Frizzled (FZD)_4,7 (without BRIL rigid insertion) structures with the inactive FZD_1,3,5,6 (with BRIL rigid insertion) structures³², TM6 adopts a similar position in both cases, with only a slight effect on its intracellular region (less than 2 Å displacement). Moreover, when comparing the structures of adenosine A_2A receptor bound to caffeine (antagonist, no BRIL insertion) with that of LJ-4517 (antagonist, BRIL introduced in place of ICL3), TM6 position is almost not affected (again less than 2 Å displacement)^33,34. These data demonstrate that BRIL insertion is suitable for investigating inactive structures of GPCRs by cryo-EM.

Using the BRIL^A2A insertion, the TVP-V2R-BRIL-Fab-Nb and MQ1^K39A-V2R-BRIL-Fab-Nb complexes were purified, cryo-EM grids were prepared, and the samples characterized as described in the “Methods” section and Supplementary Fig. 3. Data collection and processing are summarized in Supplementary Tables 1 and 2. A combination of local refinements and particle subtractions was applied to address sample dynamics, resulting in maps with an overall resolution of 2.5 Å (FSC = 0.143) for both V2R-TVP and V2R-MQ1^K39A complexes, though the toxin density remained poor. Focusing specifically on MQ1^K39A through local refinement produced a map with an overall resolution of 3.8 Å (Fig. 1, Supplementary Figs. 4 and 6, Supplementary Tables 1 and 2).

Fig. 1. Structures of the inactive TVP-V2R and MQ1^K39A-V2R complexes.

a Cryo-EM density map of TVP-V2R-BRIL-Fab-Nb complex and b corresponding 3D structure as ribbon representation. TVP is colored in gold, V2R in light blue, BRIL in gray, anti-BRIL Fab in clear green, and Anti-Fab Nb in salmon. A close-up view of the inverse agonist-receptor interaction is shown. For a better view of TVP, TM7 has been hidden. c Cryo-EM density map of MQ1^K39A-V2R-BRIL-Fab-Nb complex and d corresponding 3D structure as ribbon representation. MQ1^K39A is shown in raspberry, V2R in green, BRIL, Fab, and Nb are colored as in (a, b). A close-up view of the toxin-receptor interaction is depicted.

Density maps were well defined at the TVP-V2R interface (Fig. 1a, b). At the MQ1^K39A-V2R interface, an initial model computed with Alphafold3 fitted well the data for toxin-receptor interactions at the extracellular space, and was subsequently refined in the cryo-EM map (Fig. 1c, d). In both maps, the TMs of V2R and the first half of the helix 8 (H8) are well defined, whereas parts of extracellular loops (ECL) 2 (from Q180 to T190), of ECL3 (from E299 to A305) and of the intracellular loop (ICL) 2 (from L146 to W156) are not visible and thus not included in the structures. Moreover, the N-terminus (up to D33) and the C-terminus (from S338 to S371) of V2R are not seen. As they are absent in the densities maps, several V2R residues were also removed (P34-L36, R68-R70, M145 in the TVP-V2R complex; Q34, P108, R181, A197-P199 in the MQ1^K39A-V2R complex), as well as several residue side-chains scattered all over the V2R structure. Again, to fit with the density map, some residues (R1-S3, S24-K29, G57) and several side-chains (Y22, S32, N43, Q49-V56) of the MQ1^K39A toxin were removed. Overall, both maps led to near atomic construction of the TVP-V2R and MQ1^K39A-V2R structures (Fig. 1, Supplementary Fig. 6).

TVP and MQ inverse agonists display distinct binding modes

In agreement with their competitive behavior, both TVP and MQ1^K39A occupy the AVP orthosteric binding pocket, but with substantial differences (Fig. 1). TVP interacts with the bottom of the pocket in a central position (perpendicular to the TM helical bundle) (Fig. 1a, b), whereas MQ1^K39A rather interacts with residues lining the binding cavity and located in the ECLs (Fig. 1c, d).

In more details, TVP interacts with 13 residues of V2R within a 4 Å distance (Fig. 2a, b, and Supplementary Fig. 7) involving polar and hydrophobic contacts (Supplementary Table 3): Q92^2.57 and V93^2.58 in TM2, K116^3.29, Q119^3.32 and M120^3.33 in TM3, Q174^4.60 and F178^4.64 in TM4, V206^5.39 in TM5, W284^6.48 (also defined as the activation toggle switch), F287^6.51 and Q291^6.55 in TM6, M311^7.39 and S315^7.43 in TM7. The interactions between the nonpeptide antagonist and the V2R are mainly hydrophobic, through the several aromatic rings of the ligand and of the receptor residues. Of note, TVP also creates polar contacts such as H-bonds with Q92^2.57, K116^3.29, and halogen bond with Q291^6.55. Observations seen in the experimental cryo-EM structure of the TVP-V2R complex strengthen the binding pose of TVP that was predicted recently based on molecular dynamics simulations. Using these approaches, we proposed that TVP stabilizes V2R at the bottom of the binding pocket, directly contacts with W284^6.48, and prevents outward movements of TM6³⁵.

Fig. 2. Binding poses of TVP and MQ1^K39A in the V2R orthosteric ligand pocket.

a, b Side and top view of the TVP-V2R binding interface. V2R is in light blue, TVP in gold. Receptor residues within a distance of 4 Å of the ligand are highlighted in sticks, and numbered following the Ballesteros-Weinstein nomenclature. The toggle switch W^6.48 is shown in orange. H-bonds and halogen bond are illustrated as dashed lines. c, d Side and top views of the MQ1^K39A-V2R binding interface. V2R is in green, and the toxin in raspberry. Residues within a 4 Å distance from the receptor and from the toxin are highlighted in sticks, and numbered as in (a, b). H-bonds are illustrated as dashed lines. e TVP-V2R and MQ1^K39A-V2R were aligned onto the receptor structures, color scheme is equivalent to that of (a–d). The partial overlap of TVP and MQ1^K39A is shown. f TVP-V2R structure was aligned onto that of the Retosiban-OTR structure (pdb-6tpk). OTR is in gray, V2R in light blue. The non-peptide antagonist Retosiban is in clear green. Overlap of TVP and Retosiban at the bottom of the binding pocket is shown. The toggle switch W^6.48 is illustrated in orange.

Regarding the MQ1^K39A, we identified 23 receptor residues participating in the binding of the toxin to V2R within a 4 Å distance (Fig. 2c, d, and Supplementary Fig. 7, Supplementary Table 3): Q96^2.61 and W99^2.64 in TM2, D103 and R104 in ECL1, K116^3.29 and Q119^3.32 in TM3, F178^4.64 in TM4, T190, D191, C192, W193, A194 and R202 in ECL2, V206^5.39 and I209^5.42 in TM5, F287^6.51, Q291^6.55, A294^6.58 in TM6, P298 in ECL3, P306^7.34, F307^7.35, V308^7.36, M311^7.39 in TM7. The number of contacts is probably underestimated taking into account that part of ECLs 2 and 3 are missing in the MQ1^K39A-V2R structure and that two-thirds of the toxin molecule stands out of the V2R TM bundle. The toxin, by occupying the extracellular and the side parts of the V2R binding pocket, seems to act as a plug that sterically blocks the entrance of the receptor binding site. Such a steric hindrance is absent in the case of TVP due to its smaller size (Fig. 1b, d). Due to the highly charged/polar nature of MQ1^K39A, many interactions in the toxin-receptor complex are polar. Indeed, H-bonds are seen between N15 from MQ1^K39A and Q119^3.32, F17 with Q291^6.55, T34 with A194 and R202 in ECL2, G37 with P306^7.34 and F307^7.35. A comparison of the experimental MQ1^K39A-V2R structure and the Alphafold3-generated 3D model of MQ1^K39A-V2R demonstrates the highly accurate computational prediction of the complex (Supplementary Fig. 8). This suggests that ECL2 and ECL3, as well as the N-terminus of the receptor (not visible in the map), might directly interact with the toxin through complementary contacts. For instance, D30, E184, E299, or E303, which are oriented toward the toxin, might establish ionic or polar interactions with MQ1^K39A (Supplementary Fig. 8).

There are seven V2R residues which are common to TVP and MQ1^K39A binding sites (Supplementary Table 4). These are K116^3.29, Q119^3.32, F178^4.64, V206^5.39, F287^6.51, Q291^6.55, and M311^7.39, all distributed in the TMs and on both sides (TM3-4 versus TM5-6-7) of the binding pocket. Although the peptide toxin and the small molecule display very different structures, the two antagonists share a common space inside the binding pocket: N15 from the MQ1^K39A overlaps the central methylphenyl ring of TVP, while F17 of MQ1^K39A overlaps the TVP benzazepine moiety (Fig. 2e and Supplementary Table 3). On the contrary, TVP binds deeper in the orthosteric pocket, its methylbenzamidine group being in contact with the toggle switch W284^6.48 (Supplementary Fig. 7), and also Q92, V93, and S315. Due to its large volume and to many interactions at the receptor surface, the deepest residues of the binding pocket that are in contact with MQ1^K39A are F287^6.51 (one helix turn above the toggle switch W284^6.48) and M311^7.39 (Fig. 2c, d and Supplementary Fig. 7).

The crystal structure of the closely-related OTR, in complex with the small molecule retosiban, has been solved³⁶. The human OTR and V2R share more than 42% sequence identity, most strikingly in the TM bundle, and exhibit a conserved AVP/OT binding pocket³⁷. Retosiban is a potent OTR-selective non-peptide antagonist developed as an oral drug for the prevention of preterm labor^38,39. Since this molecule is a competitive antagonist, it binds in the orthosteric binding pocket of OTR and more precisely interacts at the bottom of the crevice. If we align retosiban-OTR (pdb-6tpk) and TVP-V2R complexes, TVP and retosiban significantly overlap (Fig. 2f). The central 2,5-diketopiperazine core motif of retosiban occupies a similar position than the central methylphenyl ring of TVP, whereas the oxazol moiety and the butane-2-yl group of retosiban overlap the benzazepine moiety of TVP. In addition, the indanyl substituent of retosiban is located at the bottom of the OTR pocket like the methylbenzamidine of TVP in the V2R, both interacting with the toggle switch W^6.48 (Fig. 2f). Many conserved residues are common to retosiban and TVP binding sites from TM2 to TM6, Q^2.57, K^3.29, Q^3.32, Q^4.60, F^4.64, I/V^5.39, W^6.48, F^6.51, Q^6.55 (Supplementary Table 5). Consequently, their receptor selectivity is likely linked to other surrounding residues. Although these two small molecules are highly specific for their respective receptor, they seem to stabilize the two receptors with an equivalent mechanism involving direct contacts with the W^6.48. In addition, the orientation of the toggle switch W^6.48 side-chain is equivalent in both inactive structures (Fig. 2f).

Binding mode of ligands with distinct efficacies

Most of residues involved in TVP or MQ1^K39A contacts also participate in the binding of the endogenous hormone AVP (Fig. 3, Supplementary Table 4). Below we compare the ligand binding modes to discuss their potential impact on ligand efficacy (TVP and MQ1^K39A, both behave as inverse agonists, whereas AVP is the natural agonist). As described above, TVP is a small nonpeptide molecule whereas the toxin is a mini-protein with a Kunitz-fold, arranged as a twisted two-stranded antiparallel beta-sheet followed by an alpha-helix (Figs. 1 and 3). The natural hormone AVP is a small cyclic agonist peptide with hydrophobic and polar/charged residues. Nonetheless, the three ligands overlap in the V2R binding pocket (Fig. 3). TVP aligns with the most hydrophobic part of AVP, precisely Tyr2 and Phe3 residues (Fig. 3c). The chloro-benzazepine moiety of TVP overlaps with Phe3, the central methylphenyl group overlaps with the backbone of Tyr2 and part of its sidechain, and the methylbenzamidine of TVP dives deeper than the Tyr2 in the orthosteric pocket (Fig. 3a) to directly interact with W284^6.48 as discussed above (Fig. 2 and Supplementary Fig. 7). As already discussed, the natural hormone does not directly interact with this residue^21–24. MQ1^K39A mostly aligns with the backbone of the 6-residue cycle of AVP (Fig. 3b, d) in the center of the binding pocket. Again, MQ1^K39A does not contact the bottom of the orthosteric pocket but does contact F287^6.51 (Supplementary Fig. 7) close to W284^6.48, and the aromatic side chains of Tyr2 and Phe3 of AVP do not overlap with the toxin (Fig. 3d).

Fig. 3. Comparison of active AVP-bound and inactive inverse agonist-bound structures of the V2R.

a TVP-bound and AVP-bound (pdb-7dw9) structures were aligned (central panel) and major differences in conformations are shown (close-up views). Inactive V2R with TVP is in light blue, TVP is in gold, active V2R with AVP is in violet, AVP is in clear green. On the left, zoom images of AVP/TVP overlap (top), extracellular (middle) and intracellular (bottom) TM6 extremities are shown. On the right, zoom images of extracellular part of TM1 (top) and disruption of TM7 (bottom) are illustrated. Distances (in Å) between the position of reference residues in the active and inactive structures are shown as red dashed lines: A294^6.58, V266^6.30, R38^1.33, and A314^7.42 were chosen. b MQ1^K39A-bound and AVP-bound (pdb-7dw9) structures were aligned (central panel), and major differences in conformations are shown (close-up views). Inactive V2R with MQ1^K39A is in green, the toxin is in raspberry. The zoom images are equivalent to those shown in (a) the reference TM residues chosen for measuring distances (red dashed lines) are equivalent. c Overlap of AVP and TVP represented as spheres. d Overlay of AVP and MQ1^K39A displayed as spheres.

Structural changes occurring during V2R activation

Solving the inactive structures of the V2R enables a direct comparison with the previously published active structures of this GPCR in complex with AVP (Figs. 3 and 4). Several conformations have been defined, one in the presence of the heterotrimeric Gs protein partner^21,23,24, another one bound to β-arrestin1²².

Fig. 4. Structural comparison of conserved motifs in inactive and active conformations of V2R.

a Alignment of TVP-V2R (top) or of MQ1^K39A-V2R (bottom) with AVP-V2R (pdb-7dw9) is shown. The color scheme is equivalent to that of Fig. 3. Microswitches along the helix bundle are highlighted as sticks. b Close-up views of activation motifs of the V2R are presented: CWxP (toggle switch), PSY (transmission motif), NPxxY (tyrosine toggle switch), and DRH (ionic lock) from top to down. Residues participating in these motifs are highlighted as sticks and numbered as superscript following the Ballesteros-Weinstein nomenclature.

Large structural differences are seen in the extracellular part of transmembrane domains TM1 and TM6, which are more open in inverse agonist-bound structures than in the AVP-V2R-Gs protein complex (pdb-7dw9). This is striking in the case of the MQ1^K39A-V2R complex (Fig. 3b), where TM1 (at the level of residue R38^1.33) is moving 5.2 Å outward compared to its position in AVP-V2R complex. This difference is still present in the case of the TVP-V2R complex but is less pronounced (2 Å, Fig. 3a). The top of TM6 is shifted outward by 2.9 Å when considering A294^6.58 as a reference residue between MQ1^K39A-V2R and AVP-V2R (Fig. 3b). Again, this is less pronounced in the case of TVP-V2R (2.2 Å, Fig. 3a). With respect to the MQ1^K39A-V2R, the size and the volume of the toxin (Figs. 1b, 3b) probably explain such a wide opening of the receptor surface, not equivalent in TVP-V2R (Figs. 1a, 3a).

From the intracellular side of the transmembrane domains, MQ1^K39A-V2R and TVP-V2R complexes both display a typical architecture of inactive GPCRs (Fig. 3a, b). Indeed, TM6 is not open, and sits close to TM3, most likely preventing the binding of GRKs, G proteins, or β-arrestins. Taking V266^6.30 as a reference point, there is a 9.1 Å outward shift in the position of TM6 from the inactive to active V2R conformation (Fig. 3a, b), a hallmark of the GPCR activation process^40,41. Whereas disruption of TM7 was evident in the active structures of AVP-V2R-Gs complexes^23,24, due to a direct hydrogen bond between AVP Tyr2 residue with the main chain carbonyl group of L312^7.40, this unusual kink is not seen in the inverse agonist-bound structures of V2R (Fig. 3a, b and Supplementary Fig. 6). This TM7 distortion participates in the activation of the receptor upon binding of AVP, due to the close contact between A314^7.42 and W284^6.48. In both TVP-V2R and MQ1^K39A-V2R complexes, helicity of TM7 is stabilized and continuous. This probably favors the stabilization of the toggle switch and the transmission switch in their inactive conformation (see below). At the level of A314^7.42, the inward shift of the TM7 is 3.4 and 3.1 Å between inverse agonist-bound V2R (TVP and MQ1^K39A, respectively) and AVP-bound V2R structure (Fig. 3a, b).

Analyzing the global concerted movements between the two conformations reveals that in the inactive state (both TVP and MQ1^K39A), the extracellular surface of V2R is open whereas the intracellular pocket is closed (Supplementary Movies 1–4). The reverse situation is seen in the AVP-bound state of V2R, with a ligand binding pocket that is contracted and an intracellular pocket largely opened to accomodate a transducer. The transition from inactive to active states, and potential associated conformational movements, can be visualized (Supplementary Movies 1–4).

In TVP-V2R and MQ1^K39A-V2R structures, as compared to the AVP-V2R (pdb-7dw9) complex, the conserved motifs involved in receptor activation reflect the inactive state (Fig. 4a, b). First, the molecular motif CWxPF encompassing the toggle switch W284^6.48 seems to be stabilized by a direct interaction with ligands (W284^6.48 and F287^6.51 for TVP, as well as F287^6.51 for MQ1^K39A, distances between ligand-receptor residues are less than 4 Å) as already discussed above (Fig. 4b). Moreover, in both complexes, W284^6.48 is constrained in an inactive conformation, maintaining strong hydrophobic interactions with residues such as F214^5.47 and F287^6.51 (Fig. 4b). Of note, F287^6.51 is clearly involved in ligand interaction, for both inverse agonists (TVP or MQ1^K39A) and agonist AVP (its mutation leads to a decrease in AVP affinity²⁴, see also this study Fig. 5 and Supplementary Fig. 10), but the conformational rearrangement of its side chain seems not to be a prerequisite for receptor activation. Then, the PIF motif defined as a transmission motif in GPCRs, here represented in the V2R as a PSY motif (P217^5.50-S127^3.40-Y280^6.44), is also stabilized in the off-state (Fig. 4b). The NPxxY motif, also defined as the tyrosine toggle switch in the TM7 (N321^7.49-P322^7.50-xx-Y325^7.53 in the V2R), shows a constrained conformation where Y325^7.53 maintains its inter-helical contacts with V58^1.53 and V332^8.50 (Fig. 4b). Finally, the DRH motif identified as the ionic lock, is still in place in the inactive state: the ionic bridge between charged D136^3.49 and R137^3.50 is present (Fig. 4b).

Fig. 5. Functional role of key residues in V2R ligand binding.

a Affinities (Ki) of AVP, b TVP, and c MQ1^K39A for the wild-type (WT) V2R and the different receptor mutants were calculated from competition binding experiments using the benzazepine-red antagonist as a tracer (see “Methods”). Dashed lines indicate the mean Ki of each ligand for the wild-type V2R. Data are means ± SEM from 3 to 6 individual experiments, each performed in triplicates. For the WT V2R, n = 6 for all AVP, TVP and MQ1^K39A competition assays. For mutant W99A, n = 4 in all competition assays. For all other mutants, n = 3 in all competition assays. Statistical significance was assessed using one-way ANOVA, comparing all mutants to the wild-type receptor: ns, not significant p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001. Here are the precise values of p. In a E184A, p > 0.9999; R202A, p = 0.0327; I209A, p = 0.0918; all other mutants, p < 0.001. In b Q92A, p = 0.1272; Q96A, p = 0.0201; D103A, p = 0.0221; M120A, p = 0.0101; E184A, p = 0.4160; W193A, p = 0.9941; R202A, p = 0.2045; M311A, p = 0.0136, all other mutants, p < 0.001. In c Q96A, p = 0.9999; M120A, p = 0.0025; E184A, p = 0.0595; I209A, p = 0.9724; W284A, p = 0.1251; Q291A, 0.0985; M311A, p = 0.6222; S315A, p = 0.0030, all other mutants, p < 0.001. Source data are provided as a Source data file. d Graph representation of the effects of V2R mutations on ligand binding. The receptor mutants are classified according to their effect toward the 3 ligands AVP, TVP, and MQ1^K39A. To highlight the most significant effects, only a loss or a gain >5-fold as compared to the value for the WT V2R was considered. The effects of D103A, R104A, E184A, I209A, Q291A and S315 were less than 5-fold. Results from the competition assays and from saturation assays (benzazepine-red tracer) are illustrated on the left and on the right, respectively.

Functional role of key residues in V2R ligand binding

Based on the experimental structures of TVP-V2R and MQ1^K39A-V2R complexes, a site-directed mutagenesis approach was performed to analyze the functional role of receptor residues involved in ligand binding. Thus, we selected a set of residues distributed throughout the orthosteric binding site, from TM2 to TM7, and in ECLs for alanine mutagenesis. The expression levels of the alanine mutants were monitored using the SNAP-tag Lumi4-Tb labeling (see “Methods” section) and compared with that of the wild-type V2R (Supplementary Fig. 9a). Although expression levels of W284A, F287A and Q291A were lower than for wild-type and other mutants, this was not limiting for ligand binding studies (Fig. 5, Supplementary Figs. 9 and 10). Competitive TR-FRET binding assays were then performed to determine the affinities (Ki values) of AVP (the reference compound), TVP, and MQ1^K39A for these variants. To this end, we first measured the affinity (Kd constant) of the fluorescent antagonist, a benzazepine-red ligand used as a reference tracer in the competition binding assays⁴². The F178A mutation completely abolished benzazepine-red binding, but we were able to measure the affinity of the fluorescent tracer for all other receptor variants (Supplementary Fig. 9b, c). The most notable effects on Kd values were observed with Q174A, M120A, K116A, W193A, I209A (Fig. 5 and Supplementary Fig. 9b). Surprisingly, Q96A, D103A, and W284A mutations led to an increase in binding affinity. The calculated Kd values for all mutants (Supplementary Fig. 9c) were then used to calculate the affinity for AVP, TVP, and MQ1^K39A from competitive binding experiments (Fig. 5, Supplementary Fig. 10, and Supplementary Table 6).

In agreement with previous functional studies^23,24, Q96A, Q174A and F287A mutations induced a strong decrease (more than a 2-log shift in the Ki value, Supplementary Table 6) in AVP affinity (Fig. 5a). Interestingly, the W99A mutant also induced a striking decrease in AVP affinity (Fig. 5, Supplementary Fig. 10 and Supplementary Table 6), confirming that the residue W99 is crucial for AVP physiological function. Indeed, mutation W99R has been shown to induce congenital nephrogenic diabetes insipidus, impairing hormone binding and G protein-coupling properties of the V2R⁴³. In addition, Q174A produced a significant decrease in TVP affinity (Fig. 5b). With respect to MQ1^K39A (Fig. 5c), W99A, K116A, and W193A mutations led to a noteworthy decrease in binding affinity (again more than a 2-log shift in the Ki value, Supplementary Table 6). Some mutations provoked a moderate change in binding affinity: K100A, K116A, Q119A for AVP (Fig. 5a), K116A, F287A and Q291A for TVP as demonstrated previously²⁰, and Q174A for MQ1^K39A (Fig. 5c). Unexpectedly, a few mutations induced a significant gain in affinity, for instance for AVP (W284A), for TVP (Q119A) and for MQ1^K39A (R202A) (Fig. 5a–c and Supplementary Table 6). Although counterintuitive, a gain in affinity can be observed occasionally in mutagenesis experiments, depending on different factors such as overall folding of the receptor domains, physico-chemical properties of molecular interactions, decrease of steric hindrance. We do not have a clear explanation about the effect for these mutants, and more work will be needed to better understand this phenomenon. Finally, R104A, E184A, I209A, S315A mutations had very limited effect or were without effect toward AVP, TVP, or MQ1^K39A binding.

These binding data highlight a crucial role for several V2R residues in inverse agonist affinity. According to the structure of the TVP-V2R complex, F178^4.64 directly interacts with the benzazepine moiety of TVP (Supplementary Fig. 7); it is thus expected to interact with the benzazepine group of the fluorescent TVP-like tracer. The direct interaction between TVP and F178^4.64 is of hydrophobic nature (Supplementary Table 3), and probably contributes significantly to the binding process of the fluorescent tracer (Fig. 5 and Supplementary Fig. 9). In addition, the F to A mutation might probably destabilize the network of interaction with surrounding residues such as Q174^4.60. Based on binding measurements, Q174^4.60 plays a central role in TVP and MQ1^K39A affinity (Fig. 5 and Supplementary Fig. 10), like for AVP^21,23. This residue directly establishes contacts with the benzazepine moiety of TVP (Fig. 2 and Supplementary Fig. 7) and, by extension, with the benzazepine group of the fluorescent tracer. K116^3.29, located in a central position of the binding pocket (Fig. 2c, d), directly contacts N15 residue in the toxin through a hydrogen bond (a 3 Å distance was measured between the oxygen atom of N15 and nitrogen atom of K116^3.29), probably explaining the drastic mutational effect (Fig. 5c and Supplementary Fig. 10). The effect of K116A mutation was less pronounced regarding TVP affinity, but was significant (Fig. 5b). Based on the TVP-V2R model (Fig. 2a, b), the nitrogen atom of the K116^3.29 side-chain makes polar contact with the carbonyl oxygen atom (O18, Supplementary Table 3) of TVP (a 3.3 Å distance).

W193^ECL2 connects with the toxin through multiple hydrophobic interactions, in particular with V9, P11, and F33. Accordingly, a strong decrease in affinity was observed when this residue was mutated (Fig. 5c). Although we may expect a role for D103^ECL1 in the toxin affinity (very close to K10 residue in the experimental density map), its mutation in alanine had a relatively small effect (Fig. 5c). From a structural point of view, W99^2.64 localized at the top of the TM2, establishes multiple hydrophobic interactions with the toxin P13 residue and to a lesser extent to P11 residue. A drastic decrease in MQ1^K39A affinity (50-fold) was measured when W99^2.64 was mutated into an alanine (Fig. 5c and Supplementary Fig. 10). This multi-contact interface probably plays a crucial role in toxin binding.

Interestingly, W193^ECL2 and R202^ECL2 were identified as residues that discriminate the MQ1^K39A based on ligand binding studies, highlighting the role of ECLs in the affinity of the toxin antagonist (Fig. 5d). On the other hand, K116^3.29 and Q174^4.60 play a common central role in AVP, TVP and MQ1^K39A affinity (Fig. 5d).

From the toxin point of view, MQ1^K39A contacts the receptor with 17 residues (Supplementary Table 3), based on the experimental cryo-EM structure of the complex. Among these residues, N15 and F17 point toward the bottom of the binding pocket (Supplementary Figs. 7 and 11a, b), and principally interact with K116^3.29, Q119^3.32, F178^4.64, and F178^4.64, Q291^6.55, respectively. Interestingly, a double variant of the toxin N15K, G16A residues drastically reduced affinity of the ligand for the V2R (1,000-fold decrease)¹⁷. This result further validates the structural data and the effect of the V2R K116A mutation toward affinity (Fig. 5c, d). This situation is also seen with mutation of the toxin F17 residue into an alanine; again a 1,000-fold lower affinity was measured⁴⁴. This result further supports the critical role of F178^4.64 in the binding of V2R ligands including the MQ1^K39A (Fig. 5c, d and Supplementary Figs. 7 and 11b). Based on the structure of MQ1^K39A-V2R complex, the toxin V9 and K10 residues are located outside the transmembrane bundle of V2R and principally interact with D103^ECL1 and W193^ECL2 (Supplementary Table 3 and Supplementary Fig. 11). Again, mutation of either the V9 or the K10 residue of MQ1 into an alanine and a glutamic acid respectively, resulted in a lower affinity of the toxin (around a 2-log decrease)⁴⁴. These data are in agreement with those of mutations D103^ECL1 and W193^ECL2 within the V2R (Fig. 5c, d and Supplementary Fig. 10). In summary, the 3D structure, the site-directed mutagenesis and pharmacological data revealed critical pairs of toxin-receptor interacting residues responsible for the high-affinity MQ1 binding to V2R, such as P13-W99^2.64, N15-K116^3.29, F17-F178^4.64 and V9/K10-R103^ECL2/W193^ECL2 (Supplementary Fig. 11).

The orthosteric binding sites of V2R, V1aR, V1bR, and OTR are conserved^37,45, so we speculate that the binding specificity of MQ1 is encoded through the large contacts established with the extracellular space of the V2R (Supplementary Fig. 8).

To analyze the putative residues involved in the selective interaction of MQ1 with V2R, we carried out a sequence alignment by comparing the primary sequences of all AVP/OT receptors (V2R, V1aR, V1bR, OTR) (Supplementary Table 7). The extracellular residues (in the N-terminus part close to TM1, the 3 extracellular loops, and their adjacent residues at the transmembrane helices extremity) that are present in V2R and absent in the other receptor subtypes were selected. Information from the experimental cryo-EM structure of the toxin-receptor complex and from the corresponding predicted alphafold3 model was also taken into account (Supplementary Fig. 8). About fifteen residues emerge from this comparison: E26, R27, and D30 (although this one in also present in V1aR) from the N-terminus, D103 from the ECL1, E184, T190 and E198 from ECL2, R202 from the top of TM5, A294 from the top of TM6, P298, E299, L302 and G304 from ECL3, P306 and V308 from the top of TM7 (Supplementary Table 7). Many are charged residues and are probably involved in the interaction with charged residues from the toxin. Additional mutagenesis experiments will be necessary to confirm this hypothesis.

Discussion

Interactions of Kunitz versus three-finger toxins to GPCRs

Toxins from venoms are a rich source of drugs and pharmacological tools for a large range of targets, including GPCRs⁴⁶. There is increasing interest in these natural peptides, due to their high selectivity, and understanding the structural basis for their mode of action is crucial. However, before the MQ1^K39A-V2R complex described here, only one structure of a toxin in complex with a class A GPCR has been published, that of the muscarinic toxin 7 (MT7) bound to the muscarinic acetylcholine receptor 1 (M1R). MT7 and MQ1 are both peptides from the green mamba snake venom, of similar size (65 and 57 residues, respectively), displaying several disulfide bonds. MT7 is a three-finger toxin⁴⁷, whereas MQ1 is a Kunitz-fold toxin as discussed above (Fig. 6). MT7 is specific for the M1R⁴⁸, displaying allosteric modulator properties^49,50. The structure of the MT7-M1R complex, solved using X-ray crystallography⁵¹, revealed the mechanism by which the toxin binds to and regulates the receptor function. MT7 occupies an extracellular vestibule, with finger loop 2 blocking access to the orthosteric site of M1R. Due to a leaning position on the side of the receptor surface (Fig. 6), the interactions between MT7 and M1R occur predominantly with ECL2, and are limited to the top of TM4 and TM7, enabling co-binding with the atropine antagonist in the orthosteric pocket. This docking accounts for its allosteric properties. In contrast, MQ1^K39A occupies a central position at the extracellular surface of V2R, interacts with all ECLs in the receptor vestibule and with TM2 to TM7, enters the orthosteric site, and accordingly behaves as a competitive ligand (Figs. 1 and 6, Supplementary Fig. 8). Binding of both toxins leads to a wide opening of M1R and V2R, at the top of TM6 and TM7 for the M1R, at the top of TM6 for V2R (Fig. 6, top right). In both structures, TM1 is pushed away from the core transmembrane bundle and does not interact with the toxins. Interestingly, ionic bridges play a major role in MT7-M1R interactions, in particular with respect to MT7 finger loops 2 and 3. Indeed, R34, R40, R52 and K65 of MT7 directly bridge with M1R E170^ECL2, E397^7.32 and E401^7.36. Whether the MQ1 also establishes ionic bridges with the extracellular domains of V2R will have to be investigated. In both systems, a tight interhelical packing is observed in the cytoplasmic half of the transmembrane bundle, locking the receptor in an inactive state.

Fig. 6. Kunitz-fold versus three-finger toxin interactions to GPCRs.

Alignment of the MT7-muscarinic M1 receptor (M1R) structure (pdb-6wjc) with the inactive MQ1^K39A-V2R structure. The V2R is shown in green, the MQ1^K39A in raspberry, the M1R in salmon, and the MT7 in light blue. Orthogonal views are displayed from the side of the TM bundle (top panels) or from the extracellular space (bottom panels). The three-finger MT7 toxin and the MQ1^K39A Kunitz-fold toxin are represented as ribbons (top panels) and as transparent surfaces (bottom panels).

GPCR binding mode with a Kunitz-fold ligand and mechanism of V2R antagonism

As compared to TVP which binds V2R at the bottom of the orthosteric pocket and establishes many hydrophobic/aromatic contacts with residues V93^2.58, M120^3.33, V206^5.39, W284^6.48 and F287^6.51 (Figs. 2, 7a), the MQ1^K39A toxin does not directly interact with the toggle switch, and contacts very few residues in this aromatic network (Fig. 7). A limited number of contacts is observed between N15 and F17 from the toxin with Q119^3.32, I209^5.42 and F287^6.51 (Fig. 2) when compared to the exposed surface of the receptor (Fig. 1). We thus hypothesized that the receptor residues at the bottom of the binding pocket (Q119^3.32, I209^5.42 and F287^6.51) are not the main drivers of the MQ1 binding affinity. This hypothesis is supported by the mutagenesis data in which mutating Q119^3.32, I209^5.42, or F287^6.51 into an alanine did not significantly change affinity of the toxin for the V2R (Fig. 5 and Supplementary Fig. 10).

Fig. 7. GPCR binding mode with a Kunitz-fold ligand and mechanism of V2R inhibition.

a Aromatic interactions between TVP and V2R are shown. V2R is illustrated as ribbons and in light blue, TVP in gold. Only TM6 is not transparent. Direct aromatic/aromatic contacts between the different moieties of TVP and F287^6.51 and W284^6.48 are highlighted in red dashed lines. b Structure of MQ1^K39A-V2R complex is shown from the side of the TM bundle (top panel) or from the extracellular space (bottom panel). MQ1^K39A is depicted in raspberry as ribbons and transparent surface. The ring of receptor toxin-interacting residues from all TM excepted TM1 is shown in green. c Aromatic network between the MQ1^K39A and V2R is proposed. The scheme color is equivalent to that of (b). The pi-stacking (T-type) contact between the toxin F18 and F307^7.35 of V2R, as well as aromatic ring contacts between F307^7.35 in TM7, F287^6.51, and W284^6.48 in TM6 are illustrated as red dashed lines.

Obviously, taking into account that the activation motifs along the TM are all switched-off like in the TVP-V2R complex, MQ1^K39A stabilizes an inactive V2R conformation possibly through an indirect stabilization of the toggle switch through F287^6.51 (Fig. 7b, c).

Most of the toxin-V2R interactions are localized at the top of the TM domains. Apart from TM1, all other TM helices are involved in this network (Fig. 7b, c). It comprises Q96^2.61 and W99^2.64 from TM2, K116^3.29 and Q119^3.32 from TM3, Q174^4.60 and F178^4.64 from TM4, R202^5.35 and V206^5.39 from TM5, Q291^6.55 and A294^6.58 from TM6, P306^7.34, F307^7.35, V308^7.36, and M311^7.39 from TM7 (Fig. 2, and Supplementary Fig. 7), representing multiple types of contact. This network maintains the receptor widely open and probably blocks the flexibility of the TMs to help stabilizing an inactive conformation. Within this ring of interacting residues (Fig. 7b), mutation of K116^3.29 and Q174^4.60 significantly reduced affinity of the toxin (Fig. 5). Mutation of Q92^2.57 also led to a significant decrease in toxin affinity though this residue is not directly contacting the toxin. This indicates that its mutation to Ala probably reorganizes the network of interaction with the neighbor residues in direct contact with the toxins, such as Q119^3.32.

We propose an unprecedented mechanism of GPCR inhibition by a Kunitz-fold toxin, in which several steps in the toxin MQ1^K39A binding pathway might be necessary to ultimately prevent V2R activation. The inverse agonist might first interact with the extracellular vestibule of V2R, in particular with charged residues at the receptor surface. This initial contact with the GPCR vestibule was previously described for drug binding to the β2-adrenergic receptor⁵² and M2 and M3 muscarinic receptors⁵³ and is probably applied to many ligand and drugs for GPCRs. The role of N-terminus and ECLs in guiding/driving ligands to the orthosteric pocket was also proposed for agonist/antagonist binding to AVP receptors⁵⁴, as well as for TVP to V2R recently²⁰. With respect to MQ1^K39A, this first step would also induce opening of the binding pocket to correctly present the toxin. In a second step, MQ1^K39A would access the orthosteric binding site in agreement with its pharmacological competitive profile¹⁷. Due to its size and volume, MQ1^K39A could be stabilized in the receptor pocket through the ring network at the top of TM helices (Fig. 7b), without interacting with the bottom of the binding pocket. In addition, the outward movement of the top of TM6 could stabilize a linear aromatic network including the MQ1^K39A F18 with F307^7.35 (through a T-type stacking interaction) in TM7, F287^6.51 in TM6, directly blocking the toggle switch W284^6.48 (Fig. 7c).

In conclusion, this study describes the structures of V2R inactive conformations and allows comparison with previously published AVP-bound V2R active states. In addition, it also depicts a 3D structure of a GPCR in complex with a Kunitz-fold mini-protein. Structure-based drug design applied to AVP V2R is still a challenging endeavor but these findings will certainly help in translating GPCR structural knowledge into the discovery of novel compounds for clinical development without unwanted side effects and with better selectivity. With the rise of the machine-learning and AI methods, this could open the way to a whole new family of GPCR binders beyond the V2R and facilitate the structure-based drug design applied to GPCRs.

Methods

Data analysis and figure preparation

Figures and videos were created using the PyMOL 3.0.4 Molecular Graphics System (Schrödinger, LLC) and the UCSF Chimera X 1.7.1 package⁵⁵. Figures were prepared using a 7-color palette adapted for color blindness. Data were plotted with GraphPad Prism 9.1.1 (GraphPad Prism Software Inc.). Data processing, refinement, and analysis software were compiled and supported by the SBGrid Consortium⁵⁶.

V2R expression and purification

All constructs designed for receptor purification and complex formation are derived from the optimized sequence of the human V2R cloned into the pFastBac1 vector for baculovirus expression system. Briefly, we used a V2R construct that contains a hemagglutinin signal peptide, a first Flag-tag, a Twin-Strep-Tag, a human rhinovirus 3C protease cleavage site, and a second Flag-tag, all positioned at the N-terminus of V2R, as described previously²². Different fusion partners were inserted into the third intracellular loop of V2R to help with the orientation of the particles (Supplementary Fig. 1): we compared a T4L module (introduced between L236^5.69 and A260^6.24), a cpGFP partner (between A234^5.67 and A264^6.28), a BRIL fragment (again between A234^5.67 and A264^6.28) and a BRIL fragment delimited by two amino-acid hinges taken from the adenosine A2A receptor sequence (A204^5.65-RRQ-L208^5.69 and R220^6.22-ARST-L225^6.27 were added at N- and C-terminus of BRIL, respectively). All constructs were expressed in Sf9 insect cells grown in Ex-Cell 420 medium (Sigma-Aldrich) to a density of 4 × 10⁶ cells/ml and infected with the recombinant baculovirus at a multiplicity of infection of 2 to 3. Infections lasted from 48 to 54 h at 28 °C in the presence of the pharmacochaperone antagonist TVP (Sigma-Aldrich) at 1 μM to increase membrane receptor expression levels²². Cells were harvested by centrifugation, and pellets were stored at −80 °C until use. All the different steps to purify V2Rs (membrane extraction and solubilization, Streptactin affinity chromatography, and Anti-Flag M2 affinity chromatography) were performed as described previously²². During exchange of n-dodecyl-β-D-maltopyranoside (DDM, Anatrace) with Lauryl Maltose Neopentyl Glycol (LMNG, Anatrace) and glyco-diosgenin (GDN, Anatrace), TVP was maintained at 10 μM or was exchanged with MQ1^K39A at 10 μM to prepare the MQ1^K39A-bound V2R complex. After concentration using a 50-KDa molecular weight cut-off concentrator (Millipore), the antagonist-bound V2R was further purified by size exclusion chromatography (SEC) using a Superdex 200 increase column (10/300 GL, Cytiva) connected to an ÄKTA purifier system in 20 mM Hepes (pH7.5), 100 mM NaCl, 0.02% LMNG, 0.005% GDN, 0.002% cholesterol hemisuccinate (CHS, Sigma-Aldrich). Fractions corresponding to the pure monomeric receptor were pooled and concentrated to 50–100 μM supplemented with an excess of TVP (200 μM) or MQ1^K39A (250 μM). The receptor was used immediately and mixed with Fab anti-BRIL and Nb anti-Fab as described below.

Fab expression and purification

Anti-BRIL Fab heavy chain was cloned into pTarget, while light chain was cloned into pD2610^28,57. Light and heavy chain plasmids were transfected at 350 μg/L of culture into HEK Gnti cells (ATCC) grown in Freestyle media (ThermoFisher) with PEI (1500 µl/L of culture). After 20–24 h, the culture was supplemented with 6 mM valproic acid (Sigma-Aldrich) and 0.8% glucose final. Five days post transfection, cells were pelleted, and the supernatant was filtered and applied to a CaptureSelect IgG-CH1 resin (ThermoFisher). The resin was washed with 10 column volumes of 20 mM Hepes pH 7.4, 150 mM NaCl, and eluted with 100 mM glycine pH 2.5 directly into 1 M Hepes pH 8 (5:10 v/v ratio). The Fab was then dialyzed overnight into 20 mM Hepes pH 7.4, 150 mM NaCl, 10% glycerol. Finally, the Fab was concentrated to 50 µM, and aliquots were flash frozen until use.

Nb expression and purification

The anti-Fab nanobody was cloned into a pET26b vector with a N-terminal PelB signal sequence for periplasmic expression and a hexahistidine tag at its C-terminal. Plasmids were transformed into BL21(DE3) E. coli and grown in LB supplemented with 50 μg/mL kanamycin. Cultures were grown to an OD₆₀₀ of 0.7 at 37 °C before induction with 0.2 mM IPTG and incubated O/N at 17 °C. Cell pellets were resuspended in room temperature TS buffer (200 mM Tris pH 8, 500 mM sucrose) and stirred for one hour. Resuspended cells were diluted with two additional volumes of cold TS2 (200 mM Tris pH 8, 125 mM sucrose) and stirred for one hour. Cell debris was centrifuged (14,000 × g, 20 min). The supernatant was flowed over Ni-NTA resin, washed with 10 CV of phosphate buffer (50 mM Na₂HPO₄, 1 M NaCl, pH 7), and eluted in the phosphate buffer supplemented with 200 mM imidazole. Eluted fractions were then purified by size-exclusion chromatography using a Superdex S75 in a HN buffer (20 mM HEPES pH 7.5, 100 mM NaCl). Nanobodies were concentrated to 900 µM and flash frozen in liquid nitrogen.

MQ1^K39A synthesis

The MQ1^K39A toxin was produced on a Prelude Synthesizer (Protein Technologies®) by solid-phase chemical synthesis as previously described^17,44. Briefly, the solid-phase synthesis using a Fmoc strategy was done on 25 μmol of ChemMatrix®. The linear peptide was cleaved and purified before being folded in the presence of oxidized and reduced cysteine (1 and 0.1 mM, respectively) in 100 mM Hepes pH7.5 and guanidine 0.5 M for 24 h.

Formation and purification of the complexes

The purified inverse agonist-bound V2R (TVP or MQ1^K39A) was mixed with the purified anti-BRIL Fab and anti-Fab Nb at a 1:1.2:1.5 molar ratio in the presence of an excess of ligand. In a representative experiment, concentrations of the different components of the complex were as follow: 11 μM V2R-BRIL, 13.3 μM Fab, 17 μM Nb, 30 μM MQ1^K39A, or 30 μM TVP. The complex formation was allowed to occur for 2 h on ice, then concentrated on a 50-kDa molecular weight cut-off concentrator before injection onto a Superdex 200 increase column (10/300 GL, Cytiva) connected to an ÄKTA purifier system in 20 mM Hepes (pH7.5), 100 mM NaCl, 0.0011% LMNG, 0.001% GDN, 0.002% CHS, 10 μM TVP or MQ1^K39A. Each complex displayed a monodisperse peak whose analysis by SDS polyacrylamide gel and Coomassie blue staining confirmed the presence of all proteins (Supplementary Fig. 2). Peak fractions were pooled, supplemented with 0.001% amphipol A8-35 and concentrated onto a 50-kDa concentrator to ~12–20 mg/ml with an excess of inverse agonist (200 μM for TVP or MQ1^K39A) for cryo-EM studies. Negative stain-EM was used to assess the sample quality before cryo-EM characterization. Briefly, 3 µl of diluted sample to 50 nM was applied to glow-discharge grid and incubated 2 min before treatment with uranyl acetate 0.75% for an additional 2 min. Grids were then visualized in a 120 kV TEM, and the collected images were then treated using Relion for 2D classification to determine the complex quality.

Cryo-EM sample preparation and image acquisition

Samples (3 µl) of the purified TVP-V2R-BRIL-Fab-Nb and MQ1^K39A-V2R-BRIL-Fab-Nb complexes obtained from SEC and concentrated to 22 mg/mL and 13 mg/mL, respectively, were applied to glow-discharged (25 mA, 10 s) QuantiFoil Gold R 0.6/1 300-mesh holey carbon grids (QuantiFoil, Micro Tools GmbH), blotted for 4.5 s, and then flash-frozen in liquid ethane using a Leica EM GP2. Images were acquired using a Titan Krios G3i microscope operating at 300 kV at the 3D-EM facility at Karolinska Institutet, Sweden. Micrographs were recorded on a Gatan K3 detector in super-resolution mode with EPU software (v. 2.14.0). In total, 11,583 and 12,309 movies were respectively captured at a magnification of 165,000x, yielding a calibrated pixel size of 0.5076 Å and an exposure dose of 80 e/Ų, with defocus values ranging from −0.6 µm to −2.0 µm (Supplementary Fig. 3).

Cryo-EM data processing

Cryo-EM data for the TVP-V2R-BRIL-Fab-Nb and MQ1^K39A-V2R-BRIL-Fab-Nb complexes were processed using cryoSPARC (v4.4–v4.5)⁵⁸. Movie frames were aligned with Patch Motion Correction, and Contrast Transfer Function (CTF) parameters were estimated through Patch CTF correction. Automatic Gaussian blob detection (estimated diameter = 80 to 180, elliptical and circular blob) was used for particle picking, producing particles that underwent reference-free 2D classification (100 classes, mask diameter = 100). Particles were extracted with a box size of 250 Å and downscaled to 2.5 Å/pixel. The best 2D classes were used as references to train a model with Topaz, a convolutional neural network for particle picking⁵⁹, which identified 1,925,457 and 4,758,606 particles, respectively, for further 2D classification. Selected particles from the best classes from both 2D classification (blob picking and Topaz) were pooled, duplicates were removed, and the pools were subjected to multiple rounds of ab-initio model reconstruction in 2, 3, or 4 classes. Particles from the best class were re-extracted at a pixel size of 1.0152 Å for refinement. After NU-refinement, particles underwent global CTF refinement, reference-based motion correction, and a final NU-Refinement, resulting in a map with an overall resolution of 2.7 Å (FSC 0.143) (Supplementary Fig. 3). Regarding the TVP-V2R-BRIL-Fab-Nb complex, to circumvent protein dynamic, series of signal subtraction and local refinement toward two sub-volumes (1. TVP-V2R; 2. Fab-Nb) yielded maps with improved quality and less distortion. Regarding the MQ1^K39A-V2R-BRIL-Fab-Nb complex, the same strategy was applied; local refinement toward three sub-volumes was carried out (1. MQ-V2R; 2. Fab-Nb; 3. MQ). For both samples the different maps were then combined with UCSF Chimera (1.13.1) vop maximum function (Supplementary Fig. 3).

Model building and refinement

AlphaFold3 predictions for V2R-BRIL-Fab-Nb and MQ1^K39A-V2R-BRIL-Fab-Nb served as initial models. The Tolvaptan isomeric smile (https://pubchem.ncbi.nlm.nih.gov/compound/Tolvaptan) was used to generate a ligand restraints dictionary file with Grade Web Server (https://grade.globalphasing.org). The models were corrected through manual inspection in Coot (v0.9), and further refined by global refinement and relaxation in Rosetta (2022.45+release.20a5bfe), and global minimization in Phenix (v1.20.1-4487) real-space refinement.

Receptor mutagenesis

Mutations of multiple V2R residues into an alanine were performed as previously described³⁵. Q92A, Q96A, W99A, K100A, D103A, R104A, K116A, Q119A, M120A, Q174A, F178A, E184A, W193A, R202A, I209A, W284A, F287A, Q291A, M311A and S315A mutations were introduced in a plasmid coding for the human V2R sequence fused at its N-terminus to the enzyme-based self-labeling SNAP-tag (pRK5-SNAP vector, PerkinElmer Revvity). For each mutation, forward and reverse primers partially overlapped each other (Supplementary Table 8). The enzyme and reaction buffers were commercially available using the KOD Hot Start DNA Polymerase kit (MERK-Millipore). PCR reactions were carried out in a final volume of 50 µl using the buffer conditions and enzyme amounts of the manufacturer protocol. The reaction mix included 0.2 mM dNTP, 0.3 µM of each primer, 1.5 mM MgSO₄, 0,02 U/µl of KOD Hot start DNA polymerase, and 200 ng of template DNA. The cycling conditions were 2 min at 95 °C, 30 s at oligo’s Tm, then 29 cycles at 72 °C for 3 min, and a final cycle at 72 °C for 1 minute. Following the PCR reaction, 1 µl of Dpn1 (New England Biolabs) was then added to the PCR product for 1 h at 37 °C. Then 2.5 µl of the PCR solution was used to transform 50 µl of commercially DH5$\propto$ competent cells (Life Technologies), before to be plated on agar ampicillin plate. Following an overnight growth, bacterial plates were stored at 4 °C. Unique clones for each condition were then picked and grown overnight at 37 °C in 3 ml of LB medium. Finally, we performed plasmid DNA mini-preparations (QIAGEN kit) to transfect HEK cells for binding assays. In addition, all mutated plasmids were sequenced to confirm each mutation with a SP6 primer (ATTTAGGTGACACTATAG) using molecular biology services (Eurofins Genomics).

TR-FRET binding assays

V2R binding studies were based on time-resolved fluorescence resonance energy transfer (TR-FRET) measurements using Tag-Lite assays (PerkinElmer Revvity) as previously described^21,42,60. Briefly, HEK cells were plated (15,000 per well) in white-walled, flat-bottom, 96-well plates (Greiner CELLSTAR plate, Sigma-Aldrich) precoated with poly-L-ornithine (14 μg/ml, Sigma-Aldrich) and incubated in Dulbecco’s minimum essential medium (DMEM) containing 10% fetal bovine serum (FBS, Eurobio), 1% nonessential amino acids (GIBCO), and penicillin/streptomycin (GIBCO). Cells were transfected 24 h later with the pRK5 plasmid coding for the human V2R (wild-type or mutant) fused at its N-terminus to the SNAP-tag (pRK5-SNAP vector, PerkinElmer Revvity). Transfections were performed with X-tremeGENE 360 (Merck), according to the manufacturer’s recommendations: 10 μl of a premix containing DMEM, X-tremeGENE 360 (0.3 μl per well), SNAP-V2 coding plasmid (from 10 ng to 100 ng per well to reach similar expression level for each construct), and noncoding plasmid (up to a total of 100 ng DNA) were added to the culture medium. After a 48-h culture period, cells were rinsed once with Tag-lite medium (PerkinElmer Revvity) and incubated in the presence of Tag-lite medium containing 100 nM benzylguanine-Lumi4-Tb for one hour at 37 °C. Cells were then washed four times. At this step, the cell surface expression level of the different V2R (wild-type or mutants) was measured in Tag-lite medium using the fluorescence value of the Lumi4-Tb donor at 620 nm. For saturation studies, cells were incubated for four hours at 4 °C in the presence of benzazepine-red nonpeptide vasopressin antagonist (BZ-DY647, PerkinElmer Revvity) at various concentrations ranging from 10⁻¹⁰ to 10⁻⁷M. Non-specific binding was determined in the presence of 10 μM AVP or TVP (when affinity of AVP was significantly decreased). For competition studies, cells were incubated for four hours at 4 °C with a fixed benzazepine-red ligand concentration (using a concentration equivalent to two Kd determined for each construct as described above) and increasing concentrations of AVP, TVP or MQ1^K39A ranging from 10⁻¹² to 10⁻⁵M. Fluorescent signals were measured at 620 nm (fluorescence of the donor) and at 665 nm (FRET signal) on a PHERAstar (BMG LABTECH). Results were expressed as the 665/620 ratio [10,000  × (665/620)]. A specific variation of the FRET ratio was plotted as a function of benzazepine-red concentration (saturation experiments) or competitor concentration (competition experiments). All binding data were analyzed with GraphPad 9.1.1 (GraphPad Prism Software Inc.) using the one site-specific binding equation. All results are expressed as the means ± SEM of at least three independent experiments performed in triplicate. K_i values were calculated from median inhibitory concentration values with the Cheng-Prusoff equation.

Statistical analysis

As indicated in TR-FRET binding assays, Kd, Ki values, and SEM were calculated form at least 3 independent experiments. The p values were obtained by One-way ANOVA analysis. Statistical significance was defined as ns, p > 0.05;*p < 0.05;**p < 0.01;***p < 0.001.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

B.M. and S.G. initiated and designed the project. A.F., P.C., H.O., and B.M. carried out purification of V2R. A.F. carried out purification of Fab and Nb. N.G. synthesized the toxin MQ1^K39A. A.F., P.C., H.O., and B.M. prepared the complexes and cryo-EM samples. J.B. conducted cryo-EM image acquisition and analysis, built and refined the 3D models. A.F. and C.Ma. performed V2R site-directed mutagenesis. P.C., H.O., L.L., T.P., and C.Me. carried out pharmacological binding assays. A.F., J.B., and B.M. designed the figures, A.F. and J.B. prepared the figures. BM wrote the initial version of the manuscript. S.G., G.S., A.F., and J.B. contributed to and reviewed the manuscript writing. B.M., S.G., and G.S. supervised and coordinated the project.

Peer review

Peer review information

Nature Communications thanks Shoji Maeda, Jagannath Maharana and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The cryo-EM density maps for TVP-V2R-BRIL-Fab-Nb and MQ1^K39A-V2R-BRIL-Fab-Nb complexes have been deposited in the Electron Microscopy Data Bank (EMBD) under accession codes EMD-51988 and EMD-52012. The coordinates for the corresponding models have been deposited in the Protein Data Bank (PDB) under accession numbers 9HAP and 9HB3. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-59114-5.