Naturally Occurring Genetic Variants in the Oxytocin Receptor Alter Receptor Signaling Profiles

HEK293T cells were maintained in Dulbecco’s modified Eagle’s medium/Ham’s F12 medium without phenol red and supplemented with 10% fetal bovine serum and 25 μg/mL gentamicin. Cells were kept in a humidified cell culture incubator at 37 °C with 5% CO₂.

The wild-type (WT) OXTR and P108A OXTR constructs in pcDNA3.1(+) vector were a kind gift from Dr. Jeffrey Murray (University of Iowa). Other missense single nucleotide variants were introduced by site-directed mutagenesis (Genewiz, South Plainfield, NJ). The WT OXTR sequence was identical to the coding region of the National Center for Biotechnology Information reference sequence NM_000916.3↗.

The β-arrestin-1-Rluc8 fusion construct in the vector pcDNA3.1(+) encoded β-arrestin-1 with a C-terminal linker SGGSTSA followed by Rluc8. The β-arrestin-2-Rluc8 fusion construct in the vector pcDNA3.1(+) encoded β-arrestin-2 with a C-terminal linker GGGSEF followed by Rluc8. The template cDNA clones for β-arrestin-1 (ARRB100002) and β-arrestin-2 (ARRB200001) were obtained from the cDNA Resource Center (Bloomsberg, PA, www.cdna.org↗). A plasmid containing the Rluc8 cDNA was a kind gift from Dr. Brian Finck (Washington University in St. Louis).

The OXTR-GFP10 fusion construct in the vector pcDNA 3.1(+) encoded OXTR with a C-terminal linker SGGKL followed by GFP10. A plasmid containing the GFP10 cDNA was a kind gift Dr. Céline Gales (INSERM, France).

The plasmid encoding OXTR-GFP was a gift from Christian Gruber (Addgene plasmid #67848; http://n2t.net/addgene:67848↗; RRID: Addgene_67848).²⁰ Note that this plasmid includes the missense single nucleotide variant A218T, which was corrected before introducing the variants of interest. An N-terminal HA tag was added (linker GPT) to generate the HA-OXTR-GFP construct.

All plasmids were confirmed by bidirectional Sanger sequencing.

Oxytocin (Tocris Bioscience, Minneapolis, MN) stock solutions diluted to 500 μM in water were stored at −80 °C until just before use.

HEK293T cells (2 × 10⁴) were plated in each well of 96-well black-walled, clear-bottom polystyrene microplates coated with poly-d-lysine. The following day, cells were transfected with a construct encoding WT or variant OXTR. Each variant was tested alongside WT controls on the same plate. For transfections, 50 ng of DNA and 0.5 μL of TransIT-293 reagent (Mirus Bio, Madison, WI) diluted in Opti-MEM reduced-serum media (Thermo Fisher Scientific, Waltham, MA) were added to each well. After 24 h, media was removed and replaced with 100 μL of Brilliant Calcium indicator solution (Ion Biosciences, San Marcos, TX), which was prepared by diluting Brilliant Calcium indicator, DrySolv, and TRS reagent in assay buffer. After incubation for 1 h, a Synergy2 plate reader (BioTek, Winooski, VT) was used to add 100 μL of oxytocin of the appropriate concentration and record the fluorescence intensity (excitation filter = 485/20 nm, emission filter = 528/20 nm) every 0.14 s for 20 s/well. Fluorescence increase (increase in intracellular Ca²⁺) was calculated as the average of fluorescence intensity readings from 10 to 20 s after oxytocin addition minus the minimum fluorescence intensity averaged over five points from 0 to 10 s.

For desensitization assays, transfected cells were pretreated with the indicated oxytocin concentrations for 30 min. Then, without washing out the pretreatment oxytocin, a Synergy 2 plate reader was used to add a challenge dose of 1 μM oxytocin and record response as above.

HEK293T cells (4 × 10⁴) were plated in each well of 96-well white-walled, clear-bottom polystyrene microplates coated with poly-d-lysine. The following day, cells were transfected with WT or variant OXTR-GFP10 and β-arrestin-1-Rluc8 or β-arrestin-2-Rluc8 at a ratio of 15:1 (w/w). For transfections, 50 ng of DNA and 0.5 μL of Lipofectamine 2000 reagent (Thermo Fisher Scientific), both diluted in Opti-MEM reduced-serum media, were added to each well. After 24 h, media was removed and replaced with 100 μL of Hank’s buffered salt solution (HBSS) supplemented with 20 mM HEPES. A Synergy2 plate reader was used to add 100 μL of assay buffer containing 10 μM coelenterazine 400a (Biotium, Fremont, CA) and the indicated concentrations of oxytocin to 10 wells at a time. Luminescence at 520 and 400 nm was read every 26 s for a total of 182 s. The BRET ratio was calculated as the average ratios of emission at 520 nm/400 nm at the five time points from 78 to 182 s. WT controls were tested on each plate in parallel with variants.

HEK293T cells (1 × 10⁶) were plated in T25 flasks and transfected the next day with HA-OXTR-GFP, OXTR-GFP, or HA-OXTR. Cells were transfected with 300 ng of plasmid DNA and 4 uL of TransIT-LT1 reagent (Mirus Bio). Cells were detached 24 h later with CellStripper (Corning) and collected by centrifugation. To measure receptor internalization, cells were incubated with the indicated concentration of oxytocin for 30 min before and during detachment. Cells were incubated with an empirically determined saturating concentration (8–16 μg/mL) of phycoerythrin (PE)-conjugated anti-HA antibody (901518, Biolegend, San Diego, CA) in staining buffer (0.5% BSA and 0.1% sodium azide in Ca²⁺/Mg²⁺-free PBS) on ice for 40 min and then washed twice in staining buffer before flow cytometry to quantify cell surface OXTR. For quantification of total OXTR, the PE-labeled living cells were fixed with 2% paraformaldehyde and permeabilized with 0.5% Tween20 in PBS. Cells were washed with 0.1% Tween 20 in PBS, incubated with 16 μg/mL PE anti-HA antibody for 40 min at room temperature, and washed twice before flow cytometry.

Flow cytometry was performed on a CytoFLEX flow cytometer (Beckman Coulter, Indianapolis, IN). Three technical replicates were performed for each experimental condition, and data from 5000 transfected cells were collected from each replicate. Three independent trials were performed. SYTOX Blue (Thermo Fisher Scientific) was used to exclude dead cells where appropriate. PE Quantibrite beads (BD Biosciences) were used for calibration. Flow cytometry gating was performed as follows: (1) forward and side scatter were used to exclude debris; (2) forward scatter width vs height was used to exclude doublets; (3) SYTOX blue staining was used to identify dead cells; (4) GFP fluorescence was used to gate transfected cells (GFP+ population). The GFP+ threshold was determined relative to the GFP signal in the GFP-negative control (cells transfected with HA-OXTR).

The number of receptors on transfected cells was calculated from the geometric mean of PE fluorescence intensity calibrated to PE standards as previously described.²¹ Values from nonspecific binding of PE-HA antibody to HA-negative cells (cells transfected with OXTR-GFP) were subtracted from all samples.

For Ca²⁺ and BRET assays, responses were normalized by subtracting the average basal response from all samples and then dividing by the average WT response at the highest oxytocin concentration for each trial. For desensitization and internalization experiments, responses were normalized by dividing values from all samples by the average response from the corresponding nonpretreated sample(s). Normalization was performed separately for each replicate experiment.

Nonlinear regression with least-squares fitting was used to generate dose–response curves and calculate E_max, EC₅₀, and IC₅₀ values (GraphPad Prism 8). The three-parameter regression method, which was used to fit the BRET data and internalization data, used the model: Y = Bottom + (Top – Bottom)/(1 + 10^{(log(EC₅₀) or IC₅₀–X)}). The four-parameter regression method, which was used to fit the Ca²⁺ activation and desensitization data, used the equation Y = Bottom + (Top – Bottom)/(1 + 10^{((log(EC₅₀) or IC₅₀–X)×HillSlope))}. In these models, Y = response, X = log(oxytocin concentration), and no constraints were placed on any values. Buffer controls were assigned a nominal concentration value of 10^–9 M for BRET assays or 10^–12 M for all other assays.

All experiments were performed in triplicate, with WT controls tested alongside each variant on the same plate to control for day-to-day variation in assay response. Average values from three biological replicates were used to construct dose–response curves for each variant and the matched WT controls, which were compared by performing nested extra sum-of-squares F tests. F statistics were calculated and P-values were determined as previously described.^22,23P-values shown reflect comparisons of log EC₅₀ values or Top values (see equations above), as indicated.

The initial homology model of WT OXTR was provided by the I-TASSER GPCR homology model database.²⁴ This model was then prepared for simulation by the CHARMM-GUI membrane protein input generator.^{25 −28} Mutations (e.g., V281M) and palmitate lipid tails on C346 and C347 were introduced by the CHARMM-GUI PDB manipulator.²⁹ All proteins were simulated in 0.15 M KCl (111 K⁺ ions and 92 Cl^– ions) in a rectangular box of size 99.5 × 99.5 × 171.2 Å with a membrane consisting of 121 (upper leaflet) or 120 (lower leaflet) POPC molecules and 12 cholesterol molecules (upper and lower leaflet). All systems contained ∼100 000 TIP3P³⁰ water molecules. Systems were minimized in the default manner supplied by CHARMM-GUI. Briefly, using the CHARMM36m force field,³¹ each system’s energy was minimized by using gradient descent, then simulated NVT with progressively weaker and fewer restraints on positions of atoms and membrane components.

Production runs were performed in GROMACS.³² Hydrogen bonds were constrained with the LINCS algorithm.³³ Cutoffs of 1.2 nm were used for the neighbor list, Coulomb interactions, and van der Waals interactions. The force-switch modifier was used to smoothly switch forces from van der Waals interactions to zero between 1.0 and 1.2 nm. The Verlet cutoff scheme was used for the neighbor list. The Nosé–Hoover thermostat was used to hold the temperature at 300 K.³⁴ The semi-isotropic Parrinello–Rahman barostat was used to maintain constant pressure of 1 bar as is standard in protein–membrane simulations.³⁵ Conformations were stored every 20 ps.

The FAST algorithm^36,37 was used to enhance conformational sampling for each OXTR sequence (WT, P108A, V281M, and V45L). Five FAST simulation rounds were conducted with 10 simulations per round. Each simulation was 50 ns in length (2.5 μs aggregate simulation). To explore away from the starting structure, the FAST ranking function favored restarting simulations from states that had the fewest number of preserved native contacts. Additionally, a similarity penalty was added to the ranking to promote conformational diversity in starting structures, as described previously.³⁸

DiffNets can perform dimensionality reduction in a way that highlights biochemically relevant differences between data sets.³⁹ Two DiffNets were independently trained to learn about impairment of β-arrestin and Gq signaling. All DiffNet training and analysis was conducted under the assumption that the regions of Gq and β-arrestin binding were most likely to contain differences that explained impaired Gq or β-arrestin signaling. Therefore, the DiffNet analysis only considered atoms in the binding region (as shown in Figure S1↗). All simulation data (2.5 μs per variant) was converted to DiffNet input as described previously.³⁹ Briefly, XYZ atom coordinates from simulations were mean-shifted to zero and then multiplied by the inverse of the square root of a covariance matrix, which was calculated from simulations. To learn about β-arrestin impairment, a DiffNet was trained to classify all structures from V45L and P108A as β-arrestin impaired (i.e., initial labels of one) and WT and V281M simulations as normal (i.e., initial labels of zero). To learn about Gq impairment, a DiffNet was trained to classify structures from V281M simulations as potentially Gq impaired and WT, V45L, and P108A simulations as normal. In both cases, the labels were iteratively updated in a self-supervised manner described previously³⁹ in which expectation maximization bounds of [0.1–0.4] were chosen for normal variants and [0.6–0.9] for impaired variants. Both training sessions used 10 latent variables, 10 training epochs in which the data were subsampled by a factor of 10 in each epoch, a batch size of 32, and a learning rate of 0.0001.

A Markov State Model (MSM) is a statistical framework for analyzing molecular dynamics simulations and provides a network representation of a free energy landscape.^{40 −42} To quantify differences between variants, several measurements were made that relied on MSMs, each built with 2.5 μs of simulation data for each variant. All MSMs were constructed with Enspara,⁴³ a python library for clustering and building MSMs from molecular simulation data. In this work, Enspara was used to cluster OXTR structures, count transitions between clusters, and derive equilibrium probabilities of structural states explored during simulation. A separate MSM was built for each variant, using the same methodology for each variant. Namely, simulation frames were converted from XYZ atom coordinates to a vector containing a value indicating the amount of solvent-accessible surface area (SASA) of each residue side chain (i.e., the data was SASA featurized). SASA calculations were computed by using the Shrake–Rupley algorithm⁴⁴ (with a solvent probe radius of 0.28 nm) as implemented in the python package MDTraj.⁴⁵ SASA featurization was used for subsequent clustering because, unlike other clustering schemes (e.g., RMSD-based), SASA emphasizes the conformational changes of surface residues over internal residues, which should be most useful for understanding signaling of a transmembrane receptor that has a surface for binding ligands. Next, the SASA-featurized data were clustered with a hybrid clustering algorithm. First, a k-centers algorithm⁴⁶ was used to cluster the data into 1000 clusters. Next, three sweeps of k-medoids update steps were applied to refine the cluster centers to be in the densest regions of conformational space. Then, transition probability matrices were produced by counting transitions between states (i.e., clusters) using a 2 ns lag time, adding a prior count of and row-normalizing, as described previously.⁴⁷ Equilibrium populations were calculated as the eigenvector of the transition probability matrix with an eigenvalue of one. For the distance histograms in Figures and 7, the distance for each cluster center (i.e., representative structure of the cluster) was calculated and the distance was weighted by the corresponding equilibrium population calculated with the MSM. Similar calculations performed with an MSM built on an RMSD-based clustering scheme produced similar results (Figure S2↗).

We searched the worldwide gnomAD v2.1 data set,¹⁸ which includes 141,456 exomes, to identify the most prevalent single nucleotide missense variants in OXTR. We identified 11 OXTR variants (Table 1) with allele counts greater than 50, indicating that they were detected in more than 50 heterozygous individuals.¹⁸ These variants affected residues in multiple domains, including six residues in transmembrane domains (TMs), one in the first extracellular loop (ECL1), two in the third intracellular loop (ICL3), and two in the C-terminal tail (Table 1, FigureA). The gnomAD cohort includes homozygotes for the four most common variants: A218T, A238T, V172A, and L206V. The most prevalent variant, A218T, was found in 27% of gnomAD participants; the 11th most prevalent variant, P108A, was found in 0.05% of participants.

We reasoned that the missense variants most likely to affect clinical oxytocin response would alter oxytocin-induced Ca²⁺ signaling, which is required for myometrial smooth muscle contraction, or recruitment of β-arrestin, which is thought to mediate OXTR desensitization.¹³ Therefore, to prioritize variants for further study, we transiently transfected plasmids encoding wild-type (WT) OXTR or the 11 variants into HEK293T cells and then performed high-throughput assays to measure effects on these pathways. First, to measure increases in intracellular Ca²⁺ in response to oxytocin, we used a fluorescent Ca²⁺ indicator dye. Second, to measure β-arrestin recruitment in response to oxytocin, we performed bioluminescence resonance energy transfer assays in HEK293T cells transfected with green fluorescent protein (GFP)-tagged OXTR and luciferase-tagged β-arrestin-1 or β-arrestin-2. V45L, P108A, L206V, V281M, and E339K had the largest statistically significant effects on EC₅₀ or E_max in two or more assays and were therefore selected for further study (Figure, Tables S1–S3↗). V45L decreased the E_max for β-arrestin-1 recruitment and increased the EC₅₀ for β-arrestin-2 recruitment (Figure S3↗). P108A increased the EC₅₀ for β-arrestin-1 recruitment and increased both the EC₅₀ and the E_max for β-arrestin-2 recruitment. L206V increased the E_max for β-arrestin-1 and β-arrestin-2 recruitment. V281M increased the EC₅₀ for Ca²⁺ signaling and decreased the E_max for Ca²⁺ signaling and β-arrestin-2 recruitment. Finally, E339K increased the EC₅₀ for Ca²⁺ signaling and decreased the E_max for Ca²⁺ signaling, β-arrestin-1 recruitment, and β-arrestin-2 recruitment (Figure).

To quantify the effect of these five genetic variants on OXTR quantity and localization to the plasma membrane, we performed quantitative flow cytometry. A specific OXTR antibody is not commercially available, so we created a plasmid encoding the OXTR fusion protein HA-OXTR-GFP. We used GFP fluorescence to differentiate transfected from untransfected cells, and a phycoerythrin (PE) -conjugated anti-HA antibody to quantify the HA epitope on the extracellular N-terminus of OXTR. To quantify surface OXTRs, living cells were labeled by PE; to quantify total OXTRs throughout the cell, an additional PE-labeling step was performed after fixing and permeabilizing the PE-labeled living cells.

No variants had a statistically significant effect on the total number of OXTRs per cell after adjusting for multiple comparisons (P > 0.01 in one-sample t tests, FigureA). However, two variants (P108A and L206V) increased the number of cell surface OXTRs by 23 ± 3% and 41 ± 4%, respectively (P = 0.0003 and P = 0.0002, one sample t tests). Conversely, two variants (V281M and E339K) decreased the number of cell surface OXTRs by 49 ± 0.7% and 36 ± 2%, respectively (P < 0.0001, one-sample t tests, FigureB).

When we graphed cell surface OXTRs as a percentage of total OXTRs (FigureC), we found that 21 ± 2% of total WT OXTRs were localized to the plasma membrane. P108A and L206V increased OXTR surface localization to 25 ± 1% and 27 ± 1%, respectively (adjusted P = 0.03 for both). Conversely, V281M and E339K decreased OXTR surface localization to 12 ± 1% and 17 ± 1%, respectively (adjusted P = 0.01 for both).

OXTR internalization and desensitization, mediated in part by β-arrestin recruitment, are thought to be responsible for some adverse effects associated with oxytocin exposure, including uterine atony and postpartum hemorrhage.¹³ Thus, to assess the potential clinical implication of variants, we aimed to define their effects on OXTR desensitization and internalization. As expected, for all five variants, relative differences in the number of cell surface receptors (Figure) corresponded to the differences seen in maximal β-arrestin recruitment assays (FigureE, G). For example, P108A and L206V had elevated E_max values for β-arrestin-2 recruitment and elevated membrane localization, whereas V281M and E339K had decreased E_max values for β-arrestin recruitment decreased membrane localization. In contrast, differences in the EC₅₀ of β-arrestin recruitment did not correspond to changes in cell surface receptor number. For example, V45L increased the EC₅₀ of β-arrestin-2 recruitment but had no effect on membrane localization, and P108A increased the EC₅₀ of both β-arrestin-1 and β-arrestin-2 recruitment and increased membrane localization. We hypothesized that increased EC₅₀ values would reflect functional deficits in OXTR desensitization and internalization.

To measure desensitization, we pretreated cells expressing WT OXTR or the five variants with varying concentrations of oxytocin for 30 min and then used Ca²⁺ indicator assays to measure the cellular response to a saturating concentration (1 μM) of oxytocin (Figure). To measure internalization, we incubated cells with varying concentrations of oxytocin for 30 min and then performed quantitative flow cytometry to measure surface OXTRs (Figure). We found that V281M and L206V had no effect on either receptor desensitization or internalization (P > 0.05, extra sum-of-squares F test). In contrast, V45L, P108A, and E339K caused a rightward shift in the dose–response curve and increased the IC₅₀ for desensitization (P = 0.0001, P < 0.0001, and P < 0.0001, extra sum-of-squares F test, FigureB, Table S4↗). V45L and P108A caused a similar rightward shift in internalization assays (P = 0.0098 and P = 0.0003, extra sum-of-squares F test, FigureC, Table S4↗). Although E339K did not cause a statistically significant increase in EC₅₀ for internalization (P > 0.05), it prevented maximal internalization, with 44% of E339K OXTRs versus 24% of WT OXTRs remaining on the cell surface (P = 0.0001, FigureC).

Three of the five variants investigated had differential effects on OXTR activation (oxytocin-induced Ca²⁺ signaling in FigureA), desensitization (FigureB), and internalization (FigureC). These variants altered the balance between OXTR desensitization and activation at any given dose of oxytocin (FiguresD and S4↗). Of the three variants that impaired OXTR internalization and desensitization, only one, E339K, also altered potency and efficacy for OXTR activation, potentially due to decreased cell surface localization (FigureB). V281M had similar effects as E339K on OXTR cell surface localization and OXTR activation but had no effect on OXTR internalization or desensitization (Figure). In contrast, V45L and P108A impaired OXTR internalization and desensitization without altering OXTR activation (Figure).

In our in vitro assays, two variants (V45L and P108A) reduced β-arrestin recruitment, OXTR internalization, and OXTR desensitization compared to WT OXTR. Thus, three lines of evidence suggest that V45L and P108A decrease OXTR’s ability to activate β-arrestin. To define the structural basis of β-arrestin impairment, we used molecular dynamics simulations to computationally model the motions of all atoms in WT and variant OXTRs in solution over time (FigureA, B). We paired these simulations with the FAST algorithm (see Methods(36,37)) to enhance sampling of the conformational ensemble (i.e., the set of structural poses the receptor adopts) of each variant.

To identify the conformational changes most associated with β-arrestin impairment, we used DiffNets, deep-learning algorithms that are trained to identify biochemically relevant differences between multiple conformational ensembles (see Methods).³⁹ We first trained a DiffNet to identify differences between conformational ensembles of the two β-arrestin-impaired OXTRs (V45L and P108A) and two OXTRs (WT and V281M) with normal desensitization and internalization. From this training, the DiffNet learned a label for each simulation frame (structural configuration) from zero to one that indicated the probability that it was associated with this classification. To interpret these labels, we calculated the correlation between interatom distances in the OXTR cytosolic region (71 289 possible distances, Figure S1↗) and changes in the DiffNet label. We then plotted the 100 distances that were most correlated with the DiffNet label (FigureC). This analysis showed clear enrichment in distances that cluster at the interface between transmembrane domain 1 (TM1) and the first intracellular loop (ICL1), indicating that changes in this region were associated with β-arrestin impairment.

DiffNets identified locations associated with reduced β-arrestin function without any prior information about functional sites in OXTR. To determine whether the DifffNet predictions corresponded to functional locations, we used the simulation data to build Markov State Models. Markov State Models provide a discrete map of structural configurations and an equilibrium population value that corresponds to the proportion of time a protein spends in a given configuration.^{40 −42} The DiffNet prediction implicated the TM1-ICL1 region in β-arrestin impairment, so we used Markov State Models to more closely examine this region. In this analysis, V45L and P108A introduced an additional helical turn at the C-terminus of TM1 that was not present in WT and V281M OXTR. Specifically, we found that the hydrogen bond between Val⁶⁰ and Leu⁶⁴ was shorter in V45L and P108A OXTR than in WT and V281M OXTR (0.2 vs 0.6 nm) (FigureA). Thus, β-arrestin-impaired OXTRs were predicted to have a shorter ICL1 than OXTRs with normal β-arrestin function.

This conformational change has important implications for β-arrestin binding. First, shortening ICL1 may prevent the interactions between ICL1 and the bottom loop of β-arrestin (FigureB) previously described by Yin et al.⁴⁸ Second, shortening ICL1 reduces the distance between ICL1 and helix 8 (H8), causing a collapsed state (FigureC). When we superimposed bound structures of β-arrestin and G protein (from other GPCRs^48,49) onto the OXTR homology model, the model predicted that this shortened distance created a steric clash between ICL1 and the β-arrestin finger loop, but not between ICL1 and the G protein (FigureD). Taken together, our data suggest that the mechanism underlying reduced β-arrestin function was similar in V45L and P108A OXTR.

Our results in FigureD indicated that the balance between OXTR activation and desensitization in V281M OXTR deviated significantly from WT, with greater relative desensitization for any given unit of activation. We observed the opposite deviation in V45L and P108A OXTR, both of which had less relative desensitization for any given unit of activation. To investigate the structural basis of this difference, we used a similar approach as above and trained a second DiffNet to identify differences between conformational ensembles of V281M OXTR and V45L, P108A, and WT OXTR. We plotted the 100 distances that were most correlated with the DiffNet label in FigureD. This analysis showed enrichment for distances between transmembrane domains 3 and 5 (TM3 and TM5), indicating that structural rearrangements in this region were associated with V281M.

We then used Markov State Models to plot the probability that OXTR adopts a conformation with a given distance between TM3 and TM5. V281M OXTR was more likely to adopt conformations with a shorter distance between TM3 and TM5 than were WT, V45L, and P108A OXTR (0.8 nm versus 1.2–1.4 nm, FigureA). When we superimposed the bound β-arrestin and G protein structures, we saw that this collapsed state caused a steric clash with the G protein but not with β-arrestin (FigureB). This finding suggests that V281M disrupted the binding of Gq to OXTR without affecting β-arrestin recruitment.