Structurally defining neurokinin selectivity to improve NK2R agonists Authors: Artem Pavlovskyi, Yinhang Ren, María Gestal-Mato, Ole H Olsen, Thue W Schwartz, Zachary Gerhart-Hines, Xiangyu Liu Published: 2026 ABSTRACT AbstractIncretin-based obesity pharmacotherapies have revolutionized patient care but act predominantly by reducing food intake. Approaches that increase energy expenditure could improve efficacy but remain challenging to harness. Recently, neurokinin 2 receptor (NK2R) activation was shown to both lower food intake and stimulate energy expenditure in preclinical models. However, the endogenous NK2R ligand, NKA, crossreacts with other receptor family members that are linked to unwanted adverse effects. Therefore, understanding NK2R selectivity is the key to unlocking its therapeutic potential. Here we generated cryo-electron microscopy complexes of NK2R bound to NKA and several synthetic agonists to discover candidate interactions driving selectivity. Targeted receptor and ligand mutagenesis was then used to functionally validate the specific residues in the NK2R binding pocket and the C terminus of synthetic peptide agonists that were responsible for selectivity. These findings provide a structural framework for defining neurokinin selectivity and enable the development of improved NK2R agonists for clinical investigation.Subject terms: Cryoelectron microscopy, Diseases, Cell signalling, Cryoelectron microscopy, Peptides Main Incretin-based pharmacotherapies, driven by glucagon-like peptide-1 (GLP-1) and gastric inhibitory polypeptide (GIP) receptor agonism have dramatically changed the treatment of obesity and type 2 diabetes (T2D)1,2. However, their efficacy comes with notable limitations, including side effects, such as nausea, and significantly reduced weight lowering efficacy in people living with both obesity and T2D3–8. The underlying mechanism driving weight loss in such therapies relies solely on changing energy balance by decreasing food intake. The ability to additionally increase energy expenditure is an appealing strategy to further shift energy balance and achieve higher quality, sustained weight loss9. However, despite promising early examples of efficacy10, previous attempts to elevate energy expenditure or metabolic rate have been hampered by marked safety concerns1. Thus, leveraging energy expenditure remains an unmet need. Recently we reported that long-acting neurokinin 2 receptor (NK2R) agonism not only increased energy expenditure, without changes in heartrate, but also reduced food intake in a nonaversive manner11 (Fig. 1). Furthermore, no evidence of delayed gastric emptying or loss of lean mass, both undesirable effects of GLP-1R agonists, was observed11. NK2R agonism resulted in weight loss and significant decrease of blood lipids in both mice and macaques. We also observed peripheral insulin sensitization without the need for insulin secretion11, which would represent a differentiating feature to incretin-based strategies. This work opens up the potential for leveraging neurokinin biology for the treatment of cardiometabolic diseases12. Cross-reactivity of NKA and selective action of EB1002. The tachykinin receptor family is a group of neuropeptide receptors whose endogenous ligands exhibit high cross-reactivity. Thus, although NKA induces beneficial cardiometabolic effects through NK2R, it also activates NK1R, which is associated with side effects such as gastrointestinal motility issues, bronchoconstriction and psychiatric disorders. The selective NK2R agonist EB1002 does not activate NK1R; however, the mechanisms underlying its selectivity remain unknown. Figure created in BioRender; Gerhart-Hines, Z. https://biorender.com/onngbif (2026). NK2R is a Class A G protein-coupled receptor (GPCR) and is a member of the tachykinin receptor family, which also includes NK1R and NK3R. NK2R is expressed in many tissues, including smooth muscle of intestines and stomach, immune cells and the central and peripheral nervous systems13. When activated by its endogenous agonist neurokinin A (NKA), NK2R will recruit primarily downstream Gq/11 proteins to activate phospholipase C (PLC), thereby inducing intracellular inositol trisphosphate (IP3) production and Ca2+ release. Its activity is linked classically to intestinal motor functions, smooth muscle contraction, inflammation and asthma13. However, determining the specific physiological contributions of NK2R and leveraging it as a drug target have remained challenging given that NKA exhibits almost equal potency to all three tachykinin receptor subtypes11,14. The cross-reactivity of NKA with NK1R is particularly problematic for druggability, given that NK1R is implicated in stress signaling, gastrointestinal motility, pain, emesis, bronchoconstriction, depression and inflammation13,15–19 (Fig. 1). The lack of selectivity NKA has towards other tachykinin receptor family members stems from both the high structural similarity between neurokinin receptors and the resemblance between NKA and the endogenous ligand for NK1R, substance P (SP), with which it shares a nearly identical C-terminal sequence FXGLM-NH2 (where ‘X’ stands for F in SP and for V in NKA)20. Despite several previous high-resolution active-state cryo-electron microscopy (cryo-EM) structures of NK1R21–23 and NK2R24 with endogenous ligands, the molecular mechanism underlying NK2R selectivity remains unresolved. In our recent work characterizing the potential of NK2R for counteracting obesity and T2D, we developed an NK2R agonist, named EB1002 (ref. 11), with both high stability and selectivity. The NK2R selectivity of EB1002 was evidenced by activation curves of human NK1R and NK2R overexpressed in Cos7 cells, with measurements performed on IP3 as the relevant secondary messenger. Building on earlier attempts at selectivity25, EB1002 is a seven-residue N-terminally truncated NKA analog with modified amino acids and an attached lipid protractor. EB1002 contains substitutions in positions 2 (S2K) and 3 (F3Y), which correspond to positions 5 and 6 in NKA. Furthermore, there are chemical modifications at positions 6 (L6 backbone methylation) and 7 (conversion of methionine to methoxinine (Mox) through sulfur-to-oxygen replacement), which correspond to positions 9 and 10 in NKA (Fig. 2a). How these modifications confer such selectivity to this heptapeptide remains unknown. Such an understanding would greatly accelerate drug design and, ultimately, clinical evaluation of NK2R activation for the treatment of cardiometabolic disease. Cryo-EM structures of NK2R complexes with endogenous or synthetic peptide agonists. a, Sequences of the endogenous NK2R agonist NKA and the synthetic peptide agonists. b, Cryo-EM structures of NK2R-miniGsq-Gβγ-Nb35 complex bound to different agonists. In this study, we generated five cryo-EM structures of Gq-coupled active-state NK2R bound with either the endogenous agonist, NKA or four new synthetic agonists, including EB1002, that each possessed different chemical modifications or properties (Fig. 2a,b). We achieved a global resolution of around 2.8–3.4 Å, which enabled us to compare several ligand-receptor interactions and structural features distinguishing NK2R orthosteric binding site from NK1R. We then performed mutagenesis studies on the receptor and engineered additional ligands to ascertain the contribution of each specific interaction or peptide modification to selectivity. Once the residues that were most crucial for selectivity were identified, we mutated the corresponding amino acids in NK1R and could fully impart NK2R-like signaling in NK1R. By delineating the underlying structural features and mechanisms driving EB1002 selectivity, we gained a deeper understanding of neurokinin activation that will guide the future development of improved agonists. Structural determination of NK2R bound by endogenous versus synthetic agonists Wild-type (WT) NK2R is highly unstable and forms large aggregates on size-exclusion chromatography (SEC) (Extended Data Fig. 1). To improve the stability and proportion of active-state NK2R proteins suitable for cryo-EM interrogation, we took advantage of a previous validated approach22 that fused the engineered Gq protein α subunit, miniGs/q7026, to the C terminus of the receptor. MiniG proteins are truncated versions of the intact G proteins with complete deletion of the alpha-helical domains and several thermostable mutations, which could largely recapitulate the properties of the entire proteins27,28 and thus have been used vastly in GPCR structure determination. The resulting NK2R-miniGs/q70 fusion protein was purified in complex with Gβγ and nanobody 35 (Nb35). We then acquired cryo-EM structures of these NK2R fusion proteins bound to the highly cross-reactive endogenous ligand, NKA and four synthetic agonists: EB1001, EB1002, P336 and P383 (Extended Data Table 1), with a global resolution of 2.8–3.4 Å (Fig. 2a,b). The four agonists were chosen to provide a more complete understanding of the effect of separate modifications on selectivity and receptor activation. Purification and data processing of NKA-NK2R-miniGs/q70-Gβ1γ2-Nb35 protein complex. a, Wild-type (WT) NK2R could only form large aggregates on fluorescent SEC (F-SEC), indicating low stability and poor behavior. b, Fusion of miniGs/q70 protein to the C-terminus of WT NK2R leads to the formation of monomeric protein peak on F-SEC. c, SDS-PAGE gel of NKA-NK2R-miniGs/q70-Gβ1γ2-Nb35 protein complex after SEC. The experiment was independently repeated five times with similar results. d, Workflow of the cryo-EM data processing for the NKA-NK2R-miniGs/q70-Gβ1γ2-Nb35 protein complex. e, The gold-standard Fourier shell correlation (GSFSC) curves for the density map of the NKA-NK2R-miniGs/q70-Gβ1γ2-Nb35 protein complex. Extended Data Table 1EC50 and efficacy (Emax) values of NKA and selective NK2R agonists taken from United States of America patent US 2023/0406883 A1Measures performed using IP3-assay, efficacy normalized to IP production induced by NKA. EC50 and efficacy (Emax) values of NKA and selective NK2R agonists taken from United States of America patent US 2023/0406883 A1 Measures performed using IP3-assay, efficacy normalized to IP production induced by NKA. A similar conformational arrangement was observed among all the five active-state NK2R structures, with the maximal all-atom root mean square deviation less than 0.8 Å between receptors (Extended Data Fig. 2a). Coverage of the transmembrane (TM) region was nearly complete and a conserved disulfide bond was identified between C18145.50 and C1063.25 (superscriptions according to Ballesteros–Weinstein numbering)—a signature for many Class A GPCRs. Residues at the N terminus before T24 and at the C terminus after A321 of NK2R were not visualized in our structures, probably due to their high flexibility and incapability to maintain a solid conformation. Moreover, the density around the unstructured extracellular and intracellular loops (ECLs and ICLs) connecting TM helices was predictably less well captured. ECL2 stands out as an exception due to its close association with the agonists (Extended Data Fig. 2b–d), similar to the phenomenon that has been reported previously in the structures of NK1R bound with its endogenous agonist, SP21,22. Structure−based activation mechanism of NK2R. a, High global resemblance is observed among NK2R bound with five different agonists – NKA, EB1001, EB1002, P336 and P383 – with all-atom RMSD less than 0.8 Å. b, Local resolution distribution of the NKA-NK2R-miniGs/q70-Gβ1γ2-Nb35 protein complex. The resolution around the receptor part is optimal for structural interrogation. c, Angular distribution of the NKA-NK2R-miniGs/q70-Gβ1γ2-Nb35 protein complex. Slight orientation preference is observed. d, Cryo-EM density for every transmembrane helix and the ECL2 of NK2R. ECL2 is well resolved due to its close interaction with the agonist NKA. e, The intracellular end of NK2R TM6 goes through a conserved and drastic outward movement when compared with the inactive-state NK1R (PDB ID: 6E59), which is a hallmark of class A GPCR activation. f, NK2R activation is accomplished through a series of structural motif rearrangements, including the movement of Toggle Switch and PIF motif. Critically, the NK2R-interacting residues in the agonists and surrounding orthosteric binding pockets were well resolved in our structures, enabling the precise analysis of selectivity mechanisms. The benzene or phenol rings on the agonists served as references for us to accurately map each ligand into the cryo-EM density with high confidence. The only residue lacking in each of the synthetic ligands was the N terminus-conjugated C18diacid-2*OEG-gammaGlu moieties. The first two N-terminal residues of NKA were missing, probably due to their low degree of involvement with NK2R activation, in agreement with previous pharmacological data29,30. Cryo-EM data collection, refinement and validation statistics is provided in Extended Data Table 2. Extended Data Table 2Cryo-EM data collection, refinement and validation statistics Cryo-EM data collection, refinement and validation statistics Once the active-state NK2R structures were obtained, we first tried to analyze the conformational changes during activation and provide a detailed interpretation of the NK2R activation mechanism. Considering the absence of available inactive-state NK2R structure and the high homology between NK2R and NK1R, a previously published inactive-state NK1R structure31 was chosen for comparison and analysis. Given the high degree of similarity observed in the receptor region across different agonist-bound NK2R structures, we selected the NKA-NK2R complex for further elucidation. As expected, the intracellular end of TM6 undergoes an outward swing, moving approximately 8.9 Å away from the TM core (measured from the Cα atom of R2406.25 in NK2R, and the corresponding E2386.25 in NK1R) (Extended Data Fig. 2e). This movement is a conserved hallmark of Class A GPCR activation, facilitating the opening of the ‘intracellular pocket’ to accommodate the C-terminal α5 helix of the miniGs/q70 protein. However, the inward movement of the intracellular end of TM7, another characteristic feature of Class A GPCR activation, was less pronounced in the active-state NK2R. This observation, noted previously in both NK1R and NK2R structures22,24, may indicate a unique activation profile specific to the tachykinin receptor subfamily. The overall conformational changes in NK2R are achieved through a series of structural motif rearrangements following agonist engagement, largely adhering to the established principles of Class A GPCR activation. Ligand recognition by NK2R involves a hydrogen bond formed between the agonist and Y2666.51 of the receptor, which drives the inward and downward movement of this residue. The bulky sidechain of Y2666.51 further facilitates the displacement of W2636.48, the canonical ‘toggle switch,’ thereby aiding in the repositioning of the ‘P5.50I3.40F6.44’ connector motif, as evidenced by the outward rotation of F2596.44 (Extended Data Fig. 2f). The contraction of the orthosteric pocket and the subsequent cascade of motif rearrangements induced by agonist binding result in the outward rotation of TM6 around its intracellular end. Structures of NK2R complexes with endogenous and synthetic agonists reveal selectivity drivers Ligand selectivity is crucial for GPCR drug development, as a lack of selectivity can lead to unintended interactions with other receptors, potentially increasing the risk of ‘off-target’ side effects. This is particularly important for NK2R, where off-target activation of the closely related NK1R may induce inflammation and pulmonary side effects. In this study, the high-resolution structures of NK2R bound with NKA and four different agonists provided us with an opportunity to more completely understand the effect of separate modifications on selectivity and receptor activation. When comparing the binding modes of different agonists, it is important to consider the corresponding EC50 values of each peptide (Extended Data Table 1). All agonists had similar EC50 values for hNK2R, from approximately −8.8 (P383) to −8.4 (P336). For hNK1R binding, all the selective agonists exhibited lower efficacy than NKA (by ~30%) and EC50 values around −6.5 (compared to −8.15 for NKA). P336 had the weakest binding to hNK1R but was also comparably lower in activating hNK2R. The binding pose of NKA in our structure was very similar to that of a previously reported structure24, validating our structural models. We then compared the binding poses of NKA (Fig. 3a) with that of EB1002 (Fig. 3b) and EB1001(Fig. 3c), the earlier generation selective NK2R agonist that exhibits anti-obesity effects in nonhuman primates11, and identified several notable distinctions. One such distinction involves the F6 residue in NKA that was replaced with a tyrosine in EB1002 (hereby referred to as Y3, due to EB1002 being an N-terminally truncated NKA analog). This tyrosine introduction allowed the formation of a hydrogen bond (2.8 Å) between the hydroxyl group and the carbonyl group in the backbone of NK2R A942.65 (Fig. 3b, middle panel). Given that the H-bond is formed between Y3 and the backbone of A942.65 and position 2.65 is occupied by alanine in both NK1R and NK2R, it is unlikely to contribute directly to receptor selectivity. However, this residue may still play asymmetric roles in receptor activation. To investigate this possibility, we checked the function of NK1R A932.65V and NK2R A942.65V mutants. IP3 production assays showed a notable loss of activity in both mutants relative to their respective wild types (Extended Data Fig. 3), with the substitution having a more pronounced effect on NK1R. Differential binding features of endogenous and selective NK2R agonists. a–c, Depiction of NK2R interactions with NKA (a), EB1002 (b) and EB1001 (c) present in obtained cryo-EM structures. Each row presents different parts of the ligand: N terminus, central region and C terminus. The values assigned to each bond show the distance between the donor and the acceptor, rounded and presented in Ångstroms. IP3 production curves of Cos7 cells transfected with NK2R, NK1R and their mutants at residue 2.65 in response to NKA. IP production was measured by an inositol trisphosphate accumulation assay based on the application of 3H-myo-inositol. Emax of NK1R in response to NKA is defined as 100% IP production, with other values normalized accordingly. Mutants show markedly lower response, suggesting secondary structure destabilization or residue importance in receptor activation for both NK1R and NK2R. Experiment was performed in one biological replicate, with two technical replicates run in parallel. Due to this no statistics were derived. Source data EB1002 Y3 is additionally predicted to set up T-shaped pi stacking with NK2R F26N-ter based on distance and ring alignment. Although the distance between NKA F6 and NK2R F26N-ter could also allow for pi stacking, the corresponding Y3-mediated hydrogen bond in EB1002 would probably lead to additional stabilization of the aromatic ring and thus a stronger interaction. The backbone of NK2R Y280ECL3 was predicted to establish a hydrogen bond with the backbone of NKA, but such a bond was not detected in complex with EB1002 (Fig. 3a,b, middle panels). However, this is probably due to conformational changes in the EB1002 backbone, which instead introduce cation-pi stacking between K2 of EB1002 and Y280ECL3, as well as a hydrogen bond between Y3 of EB1002 and NK2R A942.65. The interactions between the N-terminal peptide residues and NK2R are largely similar for either agonist, with some notable distinctions (Fig. 3a,b, upper panels). The backbones of NK2R T24N-ter, A25N-ter and Q176ECL2 establish hydrogen bonds with the carboxyl group of the terminal aspartic acid of the peptides. NK2R K180ECL2 similarly forms a salt bridge with the same aspartic acid residue carboxyl group when either agonist is bound. However, one difference is that NK2R K180ECL2 is capable of forming an additional hydrogen bond with the backbone of NKA specifically but not EB1002. This is probably caused by the lysine introduced at position 2 in EB1002 (corresponding to serine at position 5 in NKA), which has a strong affinity towards Y280ECL3 and thereby pulls the EB1002 backbone further from K180ECL2 compared to NKA. An additional association with both ligand backbones involves hydrogen bonds with NK2R N972.68 (Fig. 3a,b, middle panels). Here NKA is able to form two such bonds, whereas only one hydrogen bond is found between N972.68 and bound EB1002. In contrast to the N terminus, we detect much more striking differences in how the C-terminal residues of each peptide interact with the deepest region of the receptor binding pocket (Fig. 3a,b, lower panels). NK2R N902.61 forms a hydrogen bond only in complex with NKA, whereas the introduction of an oxygen atom with the terminal methoxinine in EB1002 enables the formation of a hydrogen bond with Q1664.60. The amidated C termini of EB1002 and NKA, which are crucial for activity, engage N862.57 and Y2666.51 in both structures, albeit with varying distances, suggesting differences in interaction strength (Fig. 3a,b, lower panels). We next interrogated the structure of NK2R bound to agonist, EB1001, which has a norleucine (Nle) residue at position 7 instead of the methoxinine in EB1002. EB1001 was not able to engage as many interactions compared to EB1002 and there were many differences in the binding mode between the two agonists. In contrast to EB1002, EB1001 residues D1 and K2 were directed outwards from the binding pocket and the carbonyl group of G5 and the methylated nitrogen from its peptide bond with leucine are oriented in the opposite direction from how EB1002 binds (Fig. 3b,c, upper and lower panels). The N terminus of EB1001 establishes hydrogen bonds to the backbone of NK2R D175ECL2 and K180ECL2 (Fig. 3c, upper panel). Y3 in the EB1001 is not positioned as deeply in the binding pocket as EB1002, which disrupts the formation of a hydrogen bond with A942.65 (Fig. 3c, middle panel). Nevertheless, EB1001 is still able to potentially pi-stack with F26N-ter and establish interactions with N972.68, Y2897.35 and N862.57 similarly to EB1002 (Fig. 3c, lower panel). However, the bond to N902.61 is not established by the amidated C terminus as in case of NKA, but rather by carbonyl group from the leucine backbone. Figure 4a depicts the striking differences in binding between these two agonists despite the only difference between EB1001 and EB1002 being the substitution of an oxygen atom to carbon in methoxinine, turning it into norleucine. Structural comparison of these two ligand-bound complexes shows how such a seemingly minor modification alone can lead to profound effects in ligand positioning in the binding pocket. This subtle atomic change in the deep region of the pocket has broader ramifications, leading to differences in the backbone conformation as well as sidechain shifts at the top of the binding site. Thus, these findings highlight the allosteric nature of peptide binding, whereas a small change at one end of the ligand can propagate through the molecule and substantially influence receptor interaction at the opposite end. Comparison of EB1001, EB1002, P336 and 383 binding poses in NK2R. a, EB1001 and EB1002 differ by only one residue at the C terminus, yet this subtle difference leads to distinct binding poses in their N termini and central region of the peptide. b, EB1001 and 336 differ by only one residue at the N terminus. This minor variation results in distinct binding poses in the central region of the peptide, whereas the C terminus maintains a relatively similar conformation. c, EB1002 and 383 differ by two residues in the central region of the peptide, yet this variation is associated with distinct binding poses not only in the central region but also at both the N- and C termini. d, Although the overall receptor conformation is highly conserved, subtle structural variations are observed in regions associated with peptide agonist binding due to differences in peptide residues, particularly the N-terminal domain and ECL3. In contrast, the receptor–G protein interface remains largely unchanged, suggesting a common activation mechanism across the different peptide agonists. To further investigate this allosteric effect, we compared the binding poses of EB1001 and 336—two peptides that differ by only a single amino acid at the N terminus. EB1001 contains an aspartate (D1), whereas 336 contains a glutamate (E1). The structural differences extend beyond the N-terminal residue. Residues K2, Y3, V4, G5 and even norleucine at position 6 exhibit distinct conformations, including changes in their main chains (Fig. 4b). This comparison further underscores the presence of allosteric effects, whereby modifications at the N terminus influence the overall binding conformation of the ligand. Finally, we extended this comparative analysis to EB1002 and 383 (Fig. 4c), which share identical N- and C-terminal residues but differ by two central residues (Y3/F3 and V4/T4). Strikingly, both the N-terminal and C-terminal regions adopt different conformations in these two peptides, reinforcing the notion that allosteric effects are not limited to terminal modifications but also occur when changes are introduced in the middle of the peptide sequence. To extend the analysis beyond the ligand-binding site, we examined and compared the receptor–G protein interfaces across all the NK2R structures obtained (Fig. 4d), which has shown a high degree of alignment. This consistency supports the idea that the different peptide-bound states share a common G-protein activation mechanism. Despite these conformational variations, all four synthetic peptides exhibit selectivity for NK2R. This led us to investigate how the individual features of the peptides contribute functionally to NK2R selectivity. C-terminal ligand modifications are critical for NK2R selectivity We showed previously that the energy expenditure-inducing, appetite-suppressing features of synthetic agonist peptide, EB1002, are completely NK2R-selective in vivo11. EB1002 is derived from NKA but has been modified heavily at several residues. To determine the contributions of each EB1002 modification, we designed and tested NKA analogs with each individual residue change as well as select combinations. The C18diacid-2*OEG-gammaGlu protractor was not added to the peptides for in vitro screening; however, all analogs were acetylated at their N terminus to remove the charge similar to lipidation protraction. We then compared how the series of acylated peptides activated human NK2R versus NK1R through production of IP3 (Fig. 5) and cAMP (Extended Data Fig. 4). IP production curves of Cos7 cells transfected with NK2R and NK1R in response to NKA and its shortened N-terminally acylated analogs. a–f, IP production was measured by an IP3 accumulation assay based on the application of 3H-myo-inositol. Emax of NK2R (a, c, e) or NK1R (b, d, f) in response to NKA is defined as 100% IP production, with other values normalized accordingly. Dose–response curves are shown as the mean of three biological replicates. Each biological replicate consists of two technical replicates in a, b, e and f, and three technical replicates in c and d. Dose–response curves for NKA-Memox represent six biological replicates, of which three include three technical replicates each and three include two technical replicates each. Data are presented as mean values ± s.e.m. Source data cAMP production curves of Cos7 cells transfected with NK2R (a, c, e) and NK1R (b, d, f) in response to NKA and its shortened N-terminally acylated analogues. cAMP production was measured by a BRET-based assay using CAMYEL biosensor. Emax of NK2R or NK1R in response to NKA is defined as 100% cAMP response, with other values normalized accordingly. Dose–response curves are shown as the mean of three biological replicates. Each biological replicate consists of two technical replicates in Figures a, b, e, and f, and three technical replicates in Figures c and d. Dose–response curves for NKA-Memox represent six biological replicates, of which three include three technical replicates each and three include two technical replicates each. Data are presented as mean values ± standard error of the mean (SEM). Source data We started by investigating the F3Y substitution, which we had shown previously was essential for selectivity in the protracted agonist, EB1002 (ref. 11). Moreover, our cryo-EM structures suggested that this tyrosine in EB1002 enabled the formation of a unique hydrogen bond and pi stacking in the NK2R binding pocket, distinct from NKA. Notably, these results were obtained in the context of the heavily modified, protracted peptide. We found that point mutagenesis of the phenylalanine residue in position 3 of NKA to tyrosine diminished NK1R potency more significantly compared to its effect on NK2R, thus supporting a contribution to NK2R selectivity (Fig. 5a,b). The S2K substitution alone did not lead to marked changes in activation; however, it improved the potency for both receptors when introduced together with F3Y (logEC50 change from −7.2 to −7.6 for NK2R and change from −6.8 to −7.6 for NK1R) (Fig. 5a,b). We next evaluated the respective impacts of methylated leucine and methoxinine substitutions. Both are typically incorporated to increase protease stability, modulate pharmacokinetic properties and prevent oxidative degradation32–34. However, we found that those modifications also increased NK2R selectivity individually (Fig. 5c,d), with the methoxinine introduction leading to a bigger change than leucine methylation (logEC50 −7.1 and −7.8, respectively, compared to −8.4 for NKA). The combination of the two modifications provided a cumulative improvement in NK2R selectivity. Introducing the S2K and F3Y on top of the methylated leucine and methoxinine substitutions led to an even greater shift in response between the two receptors (Fig. 5e,f), indicating that each of the four modifications contributed to maximizing NK2R selectivity over NK1R. We obtained similar conclusions for all modifications on receptor activation using BRET-based cAMP measurements (Extended Data Fig. 4). Notably, in the context of these agonist peptides that do not contain a C18diacid*OEG-gammaGlu lipid protractor, both the individual and paired S2K and F3Y substitutions do not seem to contribute to NK2R potency. In fact, F3Y even slightly worsened potency as a single substitution. Yet, in the lipidated agonist, EB1002, these residues, particularly, F3Y are critical substitutions for conferring NK2R selectivity11. This difference could be due to the way the lipidation alters the position of the agonist in the binding pocket. Ultimately, this underscores the influence that acyl protractions can impart on agonist binding and receptor activation beyond their purpose to extend half-life in blood. Two deep binding-pocket residues are fully sufficient to confer NK2R selectivity Based on the analysis of the different EB1002 peptide modifications that revealed the importance of the C terminus for selectivity, we next investigated interactions within the deeper orthosteric pocket of tachykinin receptors. We defined the deep NK2R and NK1R binding pockets as the regions reached by the last three C-terminal residues of NKA analogs. The ligand-exposed surfaces of these regions of tachykinin receptors are comprised of many residues, but only four are different between NK1R and NK2R: F2676.54 and Y2696.54, F2646.51 and Y2666.51, M2917.39 and F2937.39, and V1163.36 and M1173.36, respectively (Fig. 6a,b). To test the contribution of these residues to selectivity, we generated NK1R mutants in which NK1R residues were substituted for the corresponding residues in NK2R. We compared the activation of these mutants with WT NK1R and WT NK2R on cAMP and IP3 production in response to (1) NKA (Fig. 6c); (2) the unprotracted, selective NKA analog with methylated leucine and methoxinine C-terminal substitutions (hereby referred to as Memox-NKA) (Fig. 6d) and (3) the protracted, selective agonist EB1002 (Fig. 6e). We started with Y2666.51 as this residue was already linked to selectivity based on our cryo-EM analysis of synthetic agonists. A single substitution of F2646.51 to tyrosine or of M2917.39 to phenylalanine had a restrained effect on the activation of NK1R by Memox-NKA and EB1002 (Extended Data Fig. 4); however, the simultaneous substitution of both residues was sufficient to make NK1R sensitive to selective NK2R agonists (Fig. 6c–e). The addition of the F2676.54 substitution to tyrosine on top of F2646.51 and M2917.39 double substitutions slightly increased the potency and efficacy of the receptor. However, given that F2676.54 improved activation for both selective (Memox-NKA, EB1002) and nonselective (NKA) ligands (Extended Data Fig. 5), we conclude that this residue did not exert a specific impact on selectivity. This is further supported by the finding that the F2676.54 substitution, on top of the combined Y2666.51 and F2937.39 double substitution, decreased the EC50 of Memox-NKA only modestly and had no appreciable effect on the EC50 of EB1002 (Table 1). Finally, the addition of substitution of V1163.36 to methionine on top of F2646.51 and M2917.39 led to decreased receptor response to all ligands. A summary of corresponding logEC50 and maximum effect (efficacy) (Emax) values for all aforementioned activation curves are shown in Table 1 (for IP3 production) and in Extended Data Table 3 (for cAMP production). Taken together, although F2676.54 had a minor and positive impact on receptor activation, our data show that the two deep pocket residues, Y2666.51 and F2937.39 are the key drivers of NK2R selectivity and represent a mechanistic avenue for selectively leveraging NK2R signaling to treat cardiometabolic disease (Fig. 7). Mutations enable NK1R activation by selective agonists. a, Depiction of NK2R deep binding pocket surface. Residues that differ in NK2R and NK1R sequence and comprise the pocket surface are highlighted in green. b, Superimposition of our EB1002-bound NK2R structure with the previously published NK1R structure (PDB ID 7RMG). Four residues differing between NK2R and NK1R in the ligand-binding pocket are indicated. Key structural features and critical interactions underlying selectivity of EB1002 for NK2R are highlighted. c–e, cAMP (top) and IP (bottom) production curves of Cos7 cells transfected with NK2R, NK1R and NK1R F6.51Y, M7.39F in response to NKA (c), Memox-NKA (d) and EB1002 (e). cAMP production was measured by a BRET-based assay using CAMYEL biosensor. IP production was measured by an IP3 accumulation assay based on the application of 3H-myo-inositol. Emax of NK2R in response to the corresponding ligands is defined as 100% of the secondary messenger response, with other values normalized accordingly. Dose–response curves are shown as the mean of three biological replicates, each consisting of three technical replicates. Data are presented as mean values ± s.e.m. Source data cAMP (a-c) and IP3 (d-f) production curves of Cos7 cells transfected with NK2R, NK1R and different NK1R mutants in response to NKA, Memox-NKA and EB1002. cAMP production was measured by a BRET-based assay using CAMYEL biosensor. IP production was measured by an inositol trisphosphate accumulation assay based on the application of 3H-myo-inositol. Emax of NK2R in response to the corresponding ligands is defined as 100% of the secondary messenger response, with other values normalized accordingly. Dose–response curves are shown as the mean of three biological replicates, each consisting of three technical replicates. Data are presented as mean values ± standard error of the mean (SEM). Dose-response curves for NK1R F6.51Y, M7.39F, V3.36M mutant are shown as mean of only two biological replicates. Due to this, standard errors of the mean (SEM) calculations were not applied. Source data Table 1logEC50 and Emax values of NK2R, NK1R and its mutants, activated by NKA, Memox-NKA and EB1002NKAMemox-NKAEB1002log(EC50)Emax, (percentage of NK2R response)log(EC50)Emax, (percentage of NK2R response)log(EC50)Emax, (percentage of NK2R response)NK1R−8.42130−6.8799≥ −6.551aNK2R−8.16100−7.96100−8.15100NK1R F6.51Y−8.35161−7.09135≥ −6.790aNK1R M7.39F−8.39135−7.56130≥ −7.0118aNK1R F6.51Y, M7.39F−8.71134−7.98126−7.47125NK1R F6.51Y, M7.39F, V3.36M−8.12159−7.44147≥−7.0127aNK1R F6.51Y, F6.54Y, M7.39F−8.73143−8.38136−7.63138Measures performed using IP3-assay, efficacy normalized to IP production of NK2R induced by corresponding ligand. Values were obtained by fitting the dose–response data presented in Extended Data Fig. 5 using a four-parameter logistic model. Source data are provided as for Extended Data Fig. 5.aEmax was not reached, values presented at agonist concentration of −6 of logarithmic scale. logEC50 and Emax values of NK2R, NK1R and its mutants, activated by NKA, Memox-NKA and EB1002 Measures performed using IP3-assay, efficacy normalized to IP production of NK2R induced by corresponding ligand. Values were obtained by fitting the dose–response data presented in Extended Data Fig. 5 using a four-parameter logistic model. Source data are provided as for Extended Data Fig. 5. aEmax was not reached, values presented at agonist concentration of −6 of logarithmic scale. Extended Data Table 3logEC50 and efficacy (Emax) values of NK2R, NK1R and its mutants, activated by NKA, Memox-NKA and EB1002Measures performed using CAMYEL-based cAMP production assay, efficacy normalized to cAMP production of NK2R induced by corresponding ligand. Values were obtained by fitting the dose–response data presented in Extended Data Fig. 5 using a four-parameter logistic model. Source data are provided as for Extended Data Fig. 5.*Emax was not reached, values presented at agonist concentration of −6 of logarithmic scale. logEC50 and efficacy (Emax) values of NK2R, NK1R and its mutants, activated by NKA, Memox-NKA and EB1002 Measures performed using CAMYEL-based cAMP production assay, efficacy normalized to cAMP production of NK2R induced by corresponding ligand. Values were obtained by fitting the dose–response data presented in Extended Data Fig. 5 using a four-parameter logistic model. Source data are provided as for Extended Data Fig. 5. *Emax was not reached, values presented at agonist concentration of −6 of logarithmic scale. Graphical depiction of key mechanistic findings on NK2R selectivity. NK2R promotes energy expenditure, insulin sensitization, lipid-lowering and appetite control leading to weight loss. However, targeting NK2R has remained challenging because of the cross-reactivity of the endogenous NK2R ligand, NKA, with family member NK1R, which is linked to several severe adverse effects. The combination of NK2R deep pocket binding residues Y6.51 and F7.39 and C-terminal modifications to the modified NKA analog serve to shift activation away from NK1R and favor NK2R. Thus, engineering agonists that further exploit this structural distinction could hold the key to unlocking the potential cardiometabolic benefits of NK2R agonism without the adverse effects of NK1R signaling. Figure created in BioRender; Gerhart-Hines, Z. https://biorender.com/5wgdh80 (2026). Discussion Here we present five new cryo-EM structures of NK2R in complex with its nonselective endogenous agonist, NKA and four new synthetic agonists. We found that the D4 of NKA or D1 at the N terminus of the synthetic peptides EB1001, EB1002 and 383, exhibited similarly strong binding interactions with NK2R residues T24N-ter, A25N-ter, D175ECL2, Q176ECL2 and K180ECL2. In compound 336, these binding interactions were mediated by the N-terminal E1. We did not observe interactions between NK2R and the first three residues of NKA, consistent with previous reports on the negligibility of these amino acids for binding30. Based on the structural data obtained, the N termini of ligands can form intrapeptide hydrogen bonds with side chains of D1/E1 or K2 in nonendogenous ligands. The same lysine can form cation-pi stacking to Y280ECL3 when peptide ligands are oriented inwards in the binding pocket. However, in structures where the K2 of ligands are directed outwards, we instead observed the formation of hydrogen bonds between the peptide chain and the backbone of Y280ECL3. We observed only minor differences in the binding mode of the residues in positions 5–6 of NKA and 3–4 of the synthetic ligands. Based on the structural distance, all ligands possess the ability to form pi stacking with F26N-ter and hydrogen bonds with N972.68. Pi stacking with F26N-ter was further stabilized by a hydrogen bond between the backbone of A942.65 and tyrosine in position 3 of EB1002, EB1001 and compound 336. The positioning of compound 336 additionally facilitated the formation of a hydrogen bond involving residue Y932.64 and the peptide backbone. In NKA and all synthetic peptides, except compound 383, the position 4 residue is a valine that is located in close proximity to V182ECL2 in the receptor, conferring beneficial hydrophobic interactions. Based on cryo-EM structures, all peptides were able to form hydrogen bonds involving their C-terminal residues and NK2R residues, N862.57, N902.61 (excluding EB1002), Y2666.51 and Y2897.35. The strength of these interactions was highly impacted by changing the C-terminal peptide residue. Substitution of methionine with norleucine improved hydrophobic interactions with I1143.33 and F2706.55 in the receptor, whereas substitution of methionine with methoxinine enabled the formation of new hydrogen bonds with Y2666.51 and Q1664.60. C-terminal modifications shape selectivity not only by strengthening interactions with NK2R but also by impeding binding to the off-target receptor, NK1R. Indeed, the hydrophobic sidechain of F2646.51 in NK1R is unable to form hydrogen bonds, and thus would probably impose steric hindrance with the more hydrophilic methoxinine of synthetic NK2R agonist peptides. We also observed that modification of even a single residue can influence the binding poses of distant regions of the peptide. This is evident from the binding pose comparisons between EB1001 and EB1002 (which differ by only one residue at the C terminus), between EB1001 and P336 (which differ by only one residue at the N terminus), and between EB1002 and P383 (which differ by two residues in the middle portion). In all these cases, small sequence changes in one part of the peptide lead to conformational changes in other regions. We further resolved the contributions of each modification of the EB1002 peptide by designing and testing NKA analogs with individual residue changes as well as select combinations. NKA analogs were acetylated on the N terminus to simulate the lack of charge caused by the C18diacid-2*OEG-gammaGlu protraction in EB1002. This was critical in the interrogation of selectivity drivers as we found that the presence of a charged N-terminal residue diminished receptor activation when C-terminal modifications are introduced (Extended Data Figs. 6 and 7). Notably, N-terminal acetylation is not able to fully mimic the lipid protractor, which must be taken into consideration during data analysis. Moreover, the methods we used for assessing receptor activation, using IP3 and cAMP production, do not allow us to distinguish changes in ligand binding from changes in the ability of the ligand to shift the receptor towards an active conformation. To make such distinction, dedicated receptor–ligand binding assays would need to be conducted. IP3 production curves of Cos7 cells transfected with NK2R (a, c) and NK1R (b, d) in response to NKA and its shortened non-acylated analogues. IP production was measured by an inositol trisphosphate accumulation assay based on the application of 3H-myo-inositol. Emax of NK2R or NK1R in response to NKA is defined as 100% IP production, with other values normalized accordingly. Dose–response curves are shown as the mean of three biological replicates, each consisting of two technical replicates. Data are presented as mean values ± standard error of the mean (SEM). Source data cAMP production curves of Cos7 cells transfected with NK2R (a, c) and NK1R (b, d) in response to NKA and its shortened non-acylated analogues. cAMP production was measured by a BRET-based assay using CAMYEL biosensor. Emax of NK2R or NK1R in response to NKA is defined as 100% cAMP response, with other values normalized accordingly. Dose–response curves are shown as the mean of three biological replicates, each consisting of two technical replicates. Data are presented as mean values ± standard error of the mean (SEM). Source data Our results suggest that the S2K substitution in EB1002 does not affect ligand selectivity but rather improves activation potency for both NK2R and NK1R, especially when introduced in combination with the F3Y substitution. Our cryo-EM structures indicate that this is caused by the ability of the lysine to set up a cation-pi stacking with Y280ECL3/Y278ECL3 residues that are present in both NK2R and NK1R, respectively. We reported previously that F3Y was a main driver of selectivity in protracted NK2R agonists11; however, we did not observe this in the nonprotracted, acetylated NKA analogs. In fact, F3Y substitution in nonprotracted peptides impinged negatively on receptor potency. This discrepancy is probably due to the impact that lipid protractors have on the positioning of a peptide in the binding pocket. Although it is challenging to resolve lipid chain conformations in cryo-EM structures, we could infer structural insights into the differences in peptide positioning. Both the NK2R:EB1002 and NK2R:compound 336 complex structures indicated that the hydroxyl group of tyrosine and the backbone of A942.65 residue could form a hydrogen bond, which was not possible in the structure with the endogenous, unprotracted NKA. Thus, the lipid protraction seems to shift agonist peptides in the binding pocket to not only accommodate the hydrophobicity of the orthosteric site but to facilitate additional hydrogen bonding. Future investigations to generate the cryo-EM structure of the acetylated peptide DKYVGmeLMox (described in this paper) bound to NK2R and comparing it to the binding mode of acylated EB1002 would be valuable to understand how the lipid protraction changes the agonists’ binding conformation. We hypothesize that the conformations will be different depending on N-terminal modification based on the activation data where nonacetylated peptides exhibit impaired activation curves compared to acetylated ones. Differences are also observed when comparing the activation curves of acetylated and lipidated peptides. At the C terminus of agonist peptides, methylation of the L6 backbone and Met7Mox substitution led to the most striking improvements in selectivity both separately and in combination. The structural rationale for methylation-dependent improvements is probably because the methylation of the leucine backbone nitrogen interferes with a possible hydrogen bond between the peptide chain and Y2877.35 in NK1R. Disruption of similar hydrogen bond does not seem to occur for the comparable residue in NK2R, Y2897.35. Furthermore, methylation can affect the backbone conformation, making the peptide chain less mobile. For the methoxinine modification, our data are consistent with a model whereby introduction of an oxygen atom allows for additional hydrogen bond formation with NK2R residues Y2666.51 and Q1664.60. This model is further supported by our cryo-EM structures of both EB1002 and compound 383, which indicate the presence of such hydrogen bonds. These bonds are not possible in NK1R, given that position 6.51 is occupied by hydrophobic phenylalanine. Thus, our findings provide mechanistic insights into how C-terminal substitutions shift ligand preference towards NK2R. To assess more directly the sufficiency of binding pocket residues for selectivity, we mutated residues in NK1R to their corresponding amino acids in NK2R as single, double or triple mutants and evaluated activation by EB1002 (with lipid protraction) and Memox-NKA (with acetylation). We focused on the deep region of the binding pocket, which we defined as the regions of NK2R and NK1R, where the last three C-terminal residues of NKA analogs reach. In this area, there are only four residues that are different between the two receptors and thus represent the most likely candidates for driving selectivity. No single mutation fully recapitulated NK2R activity; however, combined substitutions of NK1R phenylalanine to tyrosine at position 6.51 and methionine to phenylalanine at position 7.39 were sufficient to improve the response to selective agonists. Notably these NK2R-like mutations in NK1R also caused an increased response to the endogenous NK2R ligand, NKA. Conversely, replacing Y2666.51F in NK2R with the corresponding NK1R residue led to ~50% and ~75% Emax loss35 for cAMP and IP3 production, respectively, underscoring the particular importance of this residue for receptor activity. This is probably because Y6.51 enables additional hydrogen bonding, whereas the introduction of F7.39 in place of M7.39 provides a ‘bulkier’ and more hydrophobic residue, thus profoundly reshaping the binding pocket. Collectively, our data demonstrate that C-terminal peptide modifications, including backbone methylation of leucine and methionine-to-methoxinine substitution, together with NK2R residues in position 6.51 and 7.39, are key drivers of EB1002 selectivity (Fig. 7). GPCRs are the most prominent therapeutic targets in biology36,37. This is especially evident in the cardiometabolic disease indications where GLP-1R and GIPR agonists have revolutionized the treatment of obesity and T2D and amylin and glucagon-containing polyagonists are nearing the market. Although the first generations of these drugs have been injectable proteins that are engineered from the endogenous peptide ligands, newer innovations include small molecule formats1. Such advances have the potential to introduce cheaper, more easily produced and orally available compounds that could disrupt the peptide-based incretin market. This would theoretically also hold true for NK2R peptides agonists as well. However, given the high cross-reactivity of the tachykinin family members and structural similarity between binding pockets, achieving a safe, potent small molecule NK2R agonist probably represents a steeper hurdle. Methods Plasmids All receptor constructs of WT and modified human TACR1 and TACR2 were inserted into the pcDNA3.1(+)-C-DYK vector, whereas CAMYEL26 was expressed using the pcDNA3.1(+) vector. All plasmids were manufactured by GenScript Biotech. To clone plasmids, XL1-Blue competent Escherichia coli bacteria were transformed using heat shock protocol. Ampicillin resistance was used as a selectable marker. Plasmid isolation was performed using NucleoBond Xtra Midi Kit (Macherey–Nagel), according to the manufacturerʼs manual. Cell culture and transfection Cos7 African green monkey kidney fibroblast-like cell line was purchased from ATCC and cultured at 37 °C with 10% CO2 in Dulbecco’s Modified Eagle’s Medium 1× (DMEM) 1885 containing GlutaMAX, supplemented with 10% fetal bovine serum (FBS) (Sigma–Aldrich) and 1% penicillin-streptomycin (final concentration 100 U ml−1 and 0.1 mg ml−1, respectively). Cell lines were purchased directly from the vendor and thus were not authenticated at the University of Copenhagen. Cells have been tested continuously for mycoplasma contamination with negative outcomes. Cos7 cells were plated in poly-D-lysine (Sigma–Aldrich)-coated 96-well clear or white assay plates (20,000 cells per well). The subsequent day cells were transiently transfected with 200 ng per well DNA using calcium phosphate coprecipitation. In brief, DNA (GPCR, 0.3 μg per well; CAMYEL, 1 μg per well) was mixed with 2 M CaCl2 and TE buffer (10 mM Tris, 0.1 mM EDTA, pH 7.5). The DNA mixture was dropwise added to 2× herpes buffered saline (HBS) under slow vortex and incubated at room temperature for 45 min. After the incubation, DNA was added to the cells along with a final concentration of 100 nM Chloroquine (Sigma–Aldrich) and incubated for 5 h at 37 °C. Transfection was stopped by removing the transfection medium and replacing it with fresh growth medium. Bioluminescence resonance energy transfer-based cAMP assay The intracellular level of cAMP was determined in real-time using a bioluminescence resonance energy transfer (BRET)-based cAMP assay following GPCR stimulation. This approach relies on the cAMP biosensor CAMYEL, which was cotransfected into Cos7 cells along with the corresponding constructs for the different GPCRs 24 h before the assay. CAMYEL consists of a cAMP binding protein (exchange protein activated by cAMP (Epac)) flanked by a BRET pair (Renilla luciferase (Rluc) and yellow fluorescent protein (YFP)). In brief, Epac changes its conformation upon cAMP binding, resulting in a loss of BRET signal. This enables cAMP production to be measured as BRET intensity decreases. On the day after the previous transfection, Cos7 cells were washed twice with 100 µl per well of Hank’s balanced salt solution (HBSS) (GIBCO, Life Technologies) and were incubated with a total of 100 µl of 5 mM luciferase substrate coelenterazine H (Thermo Fisher Scientific) in HBSS for 5 min at 37 °C. Dose–response curve dilutions of ligands were prepared in DMSO, prediluted in HBSS at a 1:5 ratio and added to wells at a volume of 5 μl. As a next step, bioluminescence measurements were performed every 2 min for 12 min on a CLARIOstar Plus plate reader (BMG LabTech). A BRET filter set (475-30/535-30) was used to measure emission signals from Rluc and YFP simultaneously. BRET signal was determined as the ratio of the emission intensity at 535 nm to the emission intensity at 475 nm. Experiments were performed in triplicate and cAMP production was plotted as a percentage of response normalized to the cAMP production level of NKA activated WT NK2R or NK1R, which Emax assigned to 100%. Because not all tested mutants could be analyzed on the same plate, interplate differences in signal range were controlled by including the WT NK2R on every plate and measuring it in parallel with the corresponding mutants. Responses were subsequently normalized to the NK2R stimulated with NKA on the same plate before comparing mutant dose–response curves across plates. Biological replicas were made using cells with different passage numbers that were plated separately. IP3 accumulation assay The inositol phosphate (IP) accumulation assay allows the indirect determination of increasing intracellular levels of IP3 following GPCR stimulation. This approach relies on the assumption that the total mass of soluble IP in the cell derives from the degradation of IP3. For this, lithium ions (Li+) are added to the assay buffer to block inositol polyphosphate-a-phosphatase (IPPase) and inositol monophosphatase (IMPase), thus inhibiting further degradation of IP and IP2. To assess IP accumulation, [3H]-myo-inositol is added to cell medium before the assay, which leads to the incorporation of a radiolabel into phosphatidynositol 4, 5-biphosphate (PIP2) in the membrane, functions as a tracer. The measurements of increasing levels of IP and IP2 are used as a readout of IP3 accumulation. Cos7 cells were incubated with 0.5 mCi ml−1 [3H]-myo-inositol (Perkin Elmer) in 100 ml of growth medium overnight, following the transfection with the different GPCRs constructs. The subsequent day cells were washed twice with 200 ml per well of HBSS (Gibco, Life Technologies) and pre-incubated for 5 min at 37 °C with 100 ml per well of assay buffer (HBSS, 10 mM LiCl). Dose–response curve dilutions of ligands in DMSO were added in volume of 1 μl per well, followed by 90 min incubation at 37 °C. To stop receptor activation, assay buffer was removed, and plates were placed on ice. Immediately after buffer removal cells were lysed with 40 ml of 10 mM formic acid per plate. After a 30 min incubation on ice, 35 µl of the lysis solution from each well was transferred into a white 96-well plate with addition of 60 µl per well of YSI SPA scintillation beads (Perkin Elmer) prepared with 1 g of beads diluted in 80 ml of distilled water. Plates were sealed, shaken vigorously for 15 min, centrifuged for 5 min at 430g and left at room temperature for 4 h. After this time, light emission was recorded on a Microbeta counter (Perkin Elmer). Experiments were performed in triplicate and IP production was plotted as a percentage of response normalized to the IP production level of NKA activated WT NK2R or NK1R, with its Emax assigned to 100%. Because not all tested mutants could be analyzed on the same plate, interplate differences in signal range were controlled by including the WT NK2R on every plate and measuring it in parallel with the corresponding mutants. Responses were subsequently normalized to the NK2R stimulated with NKA on the same plate before comparing mutant dose–response curves across plates. Biological replicas were made using cells with different passage numbers that were plated separately. Construct cloning and protein expression Full-length WT human NK2R (amino acids 1–398) was cloned into the pcDNA3.1 expression vector, incorporating a hemagglutinin (HA) signal peptide and a FLAG tag (DYKDDDDA) at the N terminus; a 6× His, a GGSGG linker and engineered miniGs/q70 (ref. 26) were appended at the C terminus. This construct was subsequently transfected into Expi293F cells (Thermo Fisher Scientific) at a density of 2 million cells ml−1 using polyethylenimine (Polysciences) as a transfection reagent. Following transfection, the cells were incubated for 48 h to facilitate receptor expression, after which the cell pellets were harvested and stored at −80 °C for subsequent protein purification. Purification of human NK2R from Expi293 cells The cell pellets were subjected to lysis in a buffer solution composed of 20 mM Tris (pH 7.5), 1 mM EDTA, 2.5 μg ml−1 leupeptin, 0.16 mg mL−1 benzamidine and 10 μM TCEP. Following a 25-min incubation at room temperature under stirring conditions, the lysates were centrifuged at 37,000g for 15 min to facilitate the separation of cellular components. Subsequently, the membrane fractions were isolated and washed using a high-salt buffer consisting of 50 mM HEPES (pH 7.5), 750 mM NaCl, 2 mM MgCl2, 5 mM ATP, 10 μM TCEP, along with leupeptin and benzamidine as protease inhibitors. The membranes were homogenized by gentle dounce homogenization approximately five times until thoroughly mixed. After an additional centrifugation step, the membrane pellets were resuspended in a low-salt wash buffer, which contained 100 mM NaCl while maintaining the other constituents identical to those of the high-salt buffer. The suspension was then subjected to gentle dounce homogenization approximately five times to ensure thorough mixing, followed by incubation at 4 °C overnight to facilitate complex formation. Subsequently, the mixture was centrifuged to isolate the membrane fractions. These fractions were further processed through 30 cycles of dounce homogenization on ice, using a solubilization buffer that included 50 mM HEPES, 150 mM NaCl, 1% lauryl maltose neopentyl glycol (LMNG, Anatrace), 0.2% cholesteryl hemisuccinate (CHS, Sigma), 2 mM MgCl2, 2 mM CaCl2, 5 mM ATP and 10 μM TCEP, along with protease inhibitors and benzonase. To stabilize the NK2R, Gβ1γ2 protein and Nanobody 35 (Nb35) were added to the solution in a molar ratio of 1:1.2:1.2, respectively, ensuring the assembly of these components into a stable complex. The membranes were solubilized in the detergent solution for 1.5 h at 4 °C under stirring conditions, followed by centrifugation at 37,000g for 20 min to separate the soluble components from the insoluble material. The resulting supernatant was then applied to an M1 anti-Flag affinity column at a flow rate of 1.0 ml min−1. To remove impurities, the column was washed with a buffer containing 20 mM HEPES (pH 7.5), 150 mM NaCl, 0.1% LMNG, 0.02% CHS, 2 mM MgCl2, 2 mM CaCl2 and 10 μM TCEP. The target protein was subsequently eluted using a FLAG elution buffer consisting of 20 mM HEPES (pH 7.5), 150 mM NaCl, 0.01% LMNG, 0.01% CHS, 10 μM TCEP, 5 mM EDTA and 0.2 mg ml−1 Flag peptide. Finally, the eluate was concentrated using a spin concentrator (Millipore) with a 50 kDa molecular weight cutoff to achieve the desired concentration. To make the NK2R-miniGs/q-Gβ1γ2-Nb35 complex more stable, it is necessary to supplement Gβ1γ2 and Nb35; the mixture was incubated at 4 °C for 1 h. The assembled complex was then subjected to SEC using a Superdex S200 Increase 10/300 GL column (GE Healthcare). The SEC column was equilibrated with a buffer containing 20 mM HEPES (pH 7.5), 150 mM NaCl, 0.002% LMNG, 0.002% CHS, 2 mM MgCl2 and 100 μM TCEP. The peak fractions of complexes were collected. The purified complex was subsequently concentrated to a final concentration range of 4–8 mg ml−1. The concentrated sample was flash-frozen in liquid nitrogen and stored at −80 °C. This purified and concentrated complex was then used for subsequent cryo-EM sample preparation. Cryo-EM sample preparation and data collection For cryo-EM sample preparation, two kinds of grids—holey carbon gold grids (QUANTIFOIL R1.2/1.3, 300 mesh) or amorphous nickel-titanium foil grids (R1.2/1.3, Au 300 mesh)—were initially subjected to glow discharge before use. Then, 4 μl of purified NKA/EB1001/EB1002/P336/P383-bound NK2R-miniGs/q-Gβ1γ2-Nb35 protein complexes with concentration range between 2.3 mg ml−1 and 9.0 mg ml−1 were then applied on glow-discharged grids, blotted for 4.5 s with filter paper, and subsequently plunge-frozen in liquid ethane by a FEI Vitrobot Mark IV (Thermo Fisher Scientific) under constant conditions of 8 °C and 100% humidity. Cryo-EM data were acquired on a 300 kV Titan Krios G3i (Thermo Fisher Scientific) TEM supplemented with a K3 direct electron detector (Gatan) and a GIF quantum energy filter with an energy slit width of 20 eV. Videos were obtained at a nominal magnification of ×64,000, corresponding to a pixel size of 0.54 Å, with a defocus range from −1.1 μm to −1.6 μm, a total of 32 frames and an accumulated dose of 50 e−/Å−2 per micrograph. Cryo-EM data processing All micrographs used for data processing were generated and motion-corrected from raw movie stacks by MotionCor2 to a final pixel size of 1.08 Å, with a sum of 1,328, 1,372, 1,133, 1,472 and 1,572 micrographs for NKA/EB1001/EB1002/P336/P383-bound NK2R-miniGs/q-Gβ1γ2-Nb35 protein samples, respectively. The following data processing procedures were all conducted in cryoSPARC (version 3.1), unless stated otherwise. Contrast transfer function (CTF) parameters were determined using Patch CTF Estimation, and patched micrographs were further picked manually based on CTF fit resolution and relative ice thickness to exclude flawed ones. For NKA-bound NK2R whose structure was resolved first, particles were identified preliminarily by automatic blob picker with a particle diameter range from 90 Å to 140 Å, extracted from micrographs with a box size of 256 pixels, and subjected to two-dimensional classification to produce templates with clear structural features for template picker. Another two rounds of two-dimensional classification were performed upon particles from template picking, and the particles from the top classes were selected for ab initio reconstruction of an unbiased initial three-dimensional (3D) volume. Several rounds of heterogenous refinement were then carried out based on the reconstructed good volume and the other two dreadful volumes to dig for superior particles, leading ultimately to 181,339 particles remaining. Nonuniform refinement and local refinement to further enhance the density around the receptor part results in a density map with a global resolution of 2.97 Å after sharpening, as determined by gold-standard Fourier shell correlation, enabling a detailed structural interrogation. Based on the resolved high-resolution NKA-bound NK2R structure, prime particles in EB1001/EB1002/P336/P383-bound NK2R samples can be distinguished readily within several rounds of heterologous refinement after template picking, using NKA-bound NK2R structure as an initial reference; 3D ab initio reconstruction from the pools of high-quality particles and subsequent nonuniform and local refinement were also performed to obtain well-resolved maps with no bias. As for EB1001-bound NK2R structure, an additional one round of 3D classification with focus on the receptor part was applied to improve resolution that had possibly been impaired by structure dynamics. The whole data processing progress finally gives rise to the density maps at global resolution of 3.40 Å/3.15 Å/2.93 Å/2.82 Å for EB1001/EB1002/P336/P383-bound NK2R-miniGs/q-Gβ1γ2-Nb35 protein complexes, respectively. Model building and refinement For NKA-bound NK2R-miniGs/q-Gβ1γ2-Nb35 complex, a preliminary model was obtained by docking each complex component manually into the cryo-EM density map and with automatic adjustment by the ‘Fit in Map’ routine implemented in UCSF Chimera X. The NK2R model was predicted by AlphaFold2, and the other components, miniGs/q, Gβ1γ2 and Nb35, were derived from previous NK1R-Gq complex structure (Protein Data Bank (PDB) ID: 7RMG). NKA was built manually in Coot (version 0.9.8.7). As with EB1001/EB1002/P336/P383-bound NK2R-miniGs/q-Gβ1γ2-Nb35 complexes, the overall models were adjusted from NKA-bound NK2R structure, and all the agonists were depicted initially in ChemDraw and generated subsequently through the ‘eLBOW’ software package from Phenix. All the initial structure models were tuned manually in Coot, followed by several rounds of real-space refinement in Phenix and subtle manual readjustment in Coot to exclude outliers and clashes. The quality of eventual models was validated by Phenix and MolProbity before PDB deposition. Software Cryo-EM data were collected using Thermo Scientific Smart EPU Software combined with MotionCor2. Gene sequences were analyzed in SnapGene software. Cryo-EM data were processed in cryoSPARC (version 3.1), and subsequent model building and refinement was accomplished with ChimeraX_Daily, Coot (version 0.9.8.7) and Phenix (version 1.20.1). Dose–response curves were prepared using GraphPad Prism (ver. 11.0.0). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41594-026-01820-9. Supplementary information Reporting Summary Source data Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Acknowledgements We wish to thank the members of the Gerhart-Hines and Liu laboratories for valuable discussions. We thank the Tsinghua University Branch of China National Center for Protein Sciences (Beijing) as well as Shuimu Biosciences for support with cryo-EM data collection. Extended data Author contributions Z.G.-H. and X.L. conceived the project. Y.R. and X.L. designed, executed and analyzed the cryo-EM studies. A.P. led the in vitro pharmacology studies with the help of M.G.-M. O.H.O. and T.W.S. designed the unprotracted peptide agonists. A.P., M.G.-M., Y.R., Z.G.-H. and X.L. wrote the first draft. All authors contributed to data analysis, experimental design and revising the manuscript. Peer review Peer review information Nature Structural & Molecular Biology thanks Zhenhua Shao, Alejandra Tomas and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Katarzyna Ciazynska and Sara Osman, in collaboration with the Nature Structural & Molecular Biology team. Funding This project was supported by the Novo Nordisk Foundation (125973 to Z.G.-H.; NNF18OC0033444 to the Center for Adipocyte Signaling). This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 Research and Innovation Program (Starting Grant aCROBAT agreement no. 639382 to Z.G.-H. and Consolidator Grant HEAT-UP agreement no. 101088636 to Z.G.-H.). The Novo Nordisk Foundation Center for Basic Metabolic Research is an independent research center at the University of Copenhagen, partially funded by an unrestricted donation from the Novo Nordisk Foundation (NNF18CC0034900 and NNF23SA0084103). This work was supported by the Tsinghua-Peking Center for Life Sciences (X.L.), by Beijing Frontier Research Center for Biological Structure, Tsinghua University (X.L.), by National Natural Science Foundation of China (grant 32122041 to X.L.), and by Tsinghua University Initiative Scientific Research Program (X.L.). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Data availability The cryo-EM density maps for NKA-, EB1001-, EB1002-, 336- and 383-bound NK2R-Gq complexes have been deposited in the Electron Microscopy Data Bank (EMDB) under accession codes EMDB-65549, EMDB-65570, EMDB-65571, EMDB-65572, and EMDB-65573, respectively. Atomic coordinates for the atomic models of NKA-, EB1001-, EB1002-, 336-, and 383-bound NK2R-Gq complexes have been deposited in the PDB under the accession numbers 9W1J, 9W2G, 9W2H, 9W2I and 9W2J, respectively. Source data are provided with this paper. Competing interests T.W.S. and Z.G.-H. are cofounders of Embark Laboratories ApS and Incipiam Pharma ApS, which are both companies developing therapeutics for the treatment of diabetes and obesity. The other authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Artem Pavlovskyi, Yinhang Ren. Contributor Information Zachary Gerhart-Hines, Email: zpg@sund.ku.dk. Xiangyu Liu, Email: liu_xy@mail.tsinghua.edu.cn. Extended data is available for this paper at 10.1038/s41594-026-01820-9. Supplementary information The online version contains supplementary material available at 10.1038/s41594-026-01820-9. References Associated Data Supplementary Materials Reporting Summary Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Statistical Source Data—plate-reader measurements. Data Availability Statement The cryo-EM density maps for NKA-, EB1001-, EB1002-, 336- and 383-bound NK2R-Gq complexes have been deposited in the Electron Microscopy Data Bank (EMDB) under accession codes EMDB-65549, EMDB-65570, EMDB-65571, EMDB-65572, and EMDB-65573, respectively. Atomic coordinates for the atomic models of NKA-, EB1001-, EB1002-, 336-, and 383-bound NK2R-Gq complexes have been deposited in the PDB under the accession numbers 9W1J, 9W2G, 9W2H, 9W2I and 9W2J, respectively. Source data are provided with this paper.