This study presents seven cryo-electron microscopy (cryo-EM) structures showing how different classes of psychedelic and non-psychedelic compounds interact with the serotonin (5-HT) 2A receptor-the primary target for classical psychedelics' therapeutic effects-revealing both shared and distinct binding patterns that could guide the development of new therapeutic compounds with improved side effect profiles.
There is currently a resurgence in exploring the utility of classical psychedelics to treat depression, addiction, anxiety disorders, cluster headaches, and many other neuropsychiatric disorders. A biological target of these compounds, and a hypothesized target for their therapeutic actions, is the 5-HT2A serotonin receptor. Here, we present 7 cryo-EM structures covering all major compound classes of psychedelic and non-psychedelic agonists, including a β-arrestin-biased compound RS130-180. Identifying the molecular interactions between various psychedelics and the 5-HT2A receptor reveals both common and distinct motifs among the examined psychedelic chemotypes. These findings lead to a broader mechanistic understanding of 5-HT2A activation, which can catalyze the development of novel chemotypes with potential therapeutic utility and fewer side effects.
Gumpper and colleagues situate this work within a renewed interest in classical psychedelics as potential treatments for depression, addiction, anxiety, cluster headaches and other neuropsychiatric conditions. They note that although many psychedelics share agonism at the 5-HT2A serotonin receptor, the molecular basis for differences in hallucinogenicity, signalling bias (G-protein versus β-arrestin pathways), receptor selectivity and partial agonism remains poorly understood. Two competing molecular hypotheses motivating the field are summarised: hallucinations may arise from β-arrestin2 biased signalling at 5-HT2A, or alternatively a threshold of G-protein activation may determine hallucinogenic effects, with partial agonists producing non-hallucinogenic activity. The authors also mention circuit-level hypotheses linking therapeutic effects to rapid induction of neuroplasticity after single doses, as observed in some clinical trials of psilocybin. This study sets out to resolve ligand-specific molecular interactions by determining seven cryo-EM structures of the active, Gq-coupled 5-HT2A receptor bound to representative ligands from major chemotypes: 5-HT, psilocin, DMT (tryptamines), mescaline and phenethylamine-derived RS130-180 (phenethylamines), LSD and its non-hallucinogenic analogue 2-bromo-LSD (BOL, ergolines). By combining structural, mutational and functional data, the investigators aim to identify common and distinct receptor contact motifs that can explain biased signalling, selectivity, and partial agonism, and thereby inform rational design of novel chemotypes with therapeutic potential and fewer side effects.
Papers cited by this study that are also in Blossom
Carhart-Harris, R. L., Giribaldi, B., Watts, R. et al. · New England Journal of Medicine (2021)
Carhart-Harris, R. L., Bolstridge, M., Rucker, J. et al. · Lancet Psychiatry (2016)
Carhart-Harris, R. L., Bolstridge, &. M., Day, C. M. J. et al. · Psychopharmacology (2017)
Agin-Liebes, G. I., Lancelotta, R., Uthaug, M. V. et al. · ACS Pharmacology and Translational Science (2021)
The principal experimental approach was cryo-electron microscopy of active-state 5-HT2A receptor complexes. The investigators used a previously described receptor construct co-expressed with a stabilised mini-GαqiN–Gβ1–Gγ2 heterotrimer (mini-Gαq) and a single-chain antibody (scFv16) in Spodoptera frugiperda (Sf9) cells. Complexes with seven ligands (5-HT, psilocin, DMT, mescaline, BOL, LSD, RS130-180) were co-expressed, purified and subjected to cryo-EM. Global reconstructions of the receptor–heterotrimer–scFv16 assemblies were obtained, and because ligand density in the receptor was sometimes weak, focused local refinement on the receptor using cryoSPARC produced receptor-only maps suitable for ligand modelling. Ligand placements were further validated using the GemSpot docking pipeline that uses the cryo-EM map as a restraint. Maps and models were deposited (details in supplementary material). Structural model building began from a previously published 5-HT2A coordinate set and used fit-to-map procedures in ChimeraX, Phenix real-space refinement, iterative manual rebuilding in COOT, and final cleanup with ISOLDE. Ligand restraints were generated with Phenix Elbow where necessary; maps and models were validated with Phenix Mtriage and MolProbity. Global resolution was assessed by gold-standard Fourier shell correlation (0.143 cutoff) and alternative sharpening (deepEMhancer) was applied for visualisation where noted. To capture conformational heterogeneity, the authors applied cryoSPARC's 3DFlex pipeline to each particle set, generating ensembles of reconstructions (123 per structure from three latent dimensions) which were converted to model trajectories via rigid-body refinement in Phenix and analysed with MDAnalysis. Dimensionality reduction (UMAP) on Cα coordinates was used to compare ligand-stabilised conformational spaces. Biochemical and cell-based assays complemented the structural work: TRUPATH (BRET2) assays measured Gαq heterotrimer dissociation kinetics, BRET1 assays measured β-arrestin2 recruitment (with GRK2 co-transfection where indicated), and an Nb6 dissociation assay using a 5-HT2A–κOR chimera served as a conformational sensor for inactive→active transitions and ligand dissociation kinetics. Site-directed mutagenesis followed by functional assays tested roles of individual residues (examples reported include N343A, L229A, F234A and reciprocal V235M/M218V swaps between 5-HT2A and 5-HT2B). Chemical synthesis and characterisation procedures are reported for RS130-180 and intermediates, with standard purification and NMR/HRMS validation.
Seven active-state 5-HT2A–Gq cryo-EM structures were obtained with representative ligands across tryptamine, ergoline and phenethylamine chemotypes. Despite chemical diversity, the receptor backbone showed a high degree of homology across complexes; Cα RMSD relative to the 5-HT complex was reported as 0.6 Å for BOL, DMT and LSD, 0.5 Å for mescaline and psilocin, 0.8 Å for RS130-180, and 1.0 Å for a previously published 25CN-NBOH structure. Ligand placements were unambiguous after focused receptor refinement and validated computationally. Tryptamines (5-HT, psilocin, DMT) occupy a canonical orthosteric pose. All make conserved contacts within a pocket defined by residues D155, F339 and F340, in agreement with decades of biochemical data. 5-HT uniquely forms an H-bonding interaction with N343, and the N343A mutation reduced 5-HT potency, whereas psilocin and DMT show slight indole-ring shifts: psilocin’s 4'-OH points toward the amino tail rather than N343, and DMT lacks an accessory moiety to anchor the indole, consistent with the mutational and functional data presented. Ergoline structures (LSD and BOL) revealed a mechanism relevant to 5-HT2A versus 5-HT2B selectivity and partial agonism. BOL, long reported non-hallucinogenic in humans, was characterised biochemically here as a potent G-protein–biased partial agonist, which can antagonise LSD’s psychedelic actions. Structurally, the bromine at BOL’s 2-position makes van der Waals contacts with I163 and F340 deep in the orthosteric pocket; I163 corresponds to the I in the canonical PIF motif (a key activation switch). Interaction of BOL with I163 may impede the conformational rearrangement required for full activation, explaining weak partial agonism. Reciprocal mutations at the 5.39 position (V235M in 5-HT2A and M218V in 5-HT2B) altered BOL pharmacology as predicted, supporting the structural hypothesis. Phenethylamines showed distinct features. Mescaline and RS130-180 adopt related poses and contact D155; mescaline additionally contacts helix 5 and makes a hydrophobic contact via its 3'-methoxy group with L229 on extracellular loop 2 (ECL2). L229 forms a lid over the orthosteric pocket implicated previously in LSD’s long residence time. Functionally, an L229A mutation converted mescaline from an agonist to an inverse agonist, and F234A impaired ligand-dependent activation for both 5-HT and mescaline, indicating those residues’ roles in signal transmission through TM5. Using a Nb6–5-HT2A–κOR BRET sensor to monitor conformational dynamics, 5-HT and mescaline produced Nb6 dissociation reversibly reversed by antagonist, whereas LSD produced prolonged stabilisation consistent with long residence time; mescaline’s engagement with L229 did not produce LSD-like slow dissociation. The N‑benzylated phenethylamine RS130-180 (and previously reported 25CN-NBOH) directly engages the 'toggle-switch' tryptophan (W336, 6.48) and induces a distinctive receptor conformation. RS130-180 sterically displaces W336 into a downward-facing position not seen in other 5-HT2A structures, which produces an inward rotation of F332 (PIF motif) and an outward shift of Y380 (NPxxY motif) at the base of TM7. The result is a non-canonical (NC) conformation in which TMs5 and 6 resemble an active state while TM7 assumes an inactive-like configuration. Functionally, RS130-180 is β-arrestin2–biased in early time points in TRUPATH and β-arrestin assays, and cryo-EM plus kinetic data indicate this bias arises from conformational stabilisation rather than long receptor occupancy: RS130-180 and 25CN-NBOH dissociated rapidly in the Nb6 assay after antagonist challenge, while ergolines (LSD, BOL) showed prolonged occupancy. Time-dependent signalling assays showed that G-protein signalling for the N-benzylated phenethylamines catches up at later time points, producing more even signalling and indicating a kinetic component to observed bias in some cases. Analyses of conformational ensembles using 3DFlex and UMAP revealed ligand-dependent 'rocks' and 'twists' of the receptor–heterotrimer complex. Each ligand stabilised distinct regions of conformational space despite alignment prior to dimensionality reduction, consistent with the biochemical readouts (Nb6 and β-arrestin responses) that different ligands promote different active-state conformational ensembles.
Gumpper and colleagues interpret their cryo-EM and complementary data as providing mechanistic explanations for several debated phenomena in psychedelic pharmacology. They argue that tryptamines occupy the validated orthosteric site and make conserved contacts that explain family-specific signalling features, while ergolines can exploit deep orthosteric-pocket interactions to influence both selectivity and efficacy. In particular, the bromine substitution in BOL contacting I163 (PIF motif) is proposed to underlie partial agonism and may account for BOL’s lack of hallucinogenic activity despite retaining receptor engagement. For phenethylamines, engagement of the ECL2 lid (L229) can shape ligand interactions and, in the case of mescaline, is functionally important for agonism versus inverse agonism. The investigators further propose that direct targeting of the toggle-switch W336 by N-benzylated phenethylamines produces an NC conformation that favours β-arrestin2 recruitment, offering a structural basis for arrestin bias. The authors position these findings relative to prior structural and biochemical work by noting concordance with long-standing models (for example D155 and F340 contacts) while clarifying points of contention, such as the correct orthosteric pose for tryptamines versus a recently proposed extended pocket. They also emphasise that bias can be achieved through observable conformational stabilisation rather than solely by ligand residence time, supported by the rapid dissociation of arrestin-biased N-benzylated phenethylamines in the Nb6 assay. Key limitations acknowledged in the text include that cryo-EM structures represent averages over multiple microstates and that the complexes were formed with a mini-Gαq rather than native full-length Gαq, which may influence conformational sampling. The authors note these caveats when interpreting the ensemble analyses. They also acknowledge that some conclusions (for example on kinetics versus conformational stabilisation as mechanisms of bias) are informed by their assays but remain to be fully generalised. Finally, the investigators state implications for drug discovery: the structure–function relationships revealed here point to concrete molecular strategies to design chemotypes with desired selectivity, efficacy and bias profiles, such as exploring deep orthosteric-pocket interactions for ergoline selectivity or targeting the toggle-switch to engineer arrestin bias. They present the dataset as a resource to accelerate rational design of novel psychiatric therapeutics with improved safety and pharmacology.
To reveal the molecular determinants for psychedelic drug actions, we determined seven active state cryo-EM structures of the agonist-bound 5-HT 2A R in complex with heterotrimeric Gq. This set of structures contains ligands encompassing: (1) known psychedelic chemotypes (tryptamines, phenethylamines, and ergolines); (2) non-psychedelic analogs; and (3) a phenethylamine-derived β-arr2 biased agonist (Fig.). We utilized our previously employed construct for the 5-HT 2A R and previously designed mini-GαqiN-Gβ1-Gγ2 (mini-Gαq) heterotrimer with stabilizing single-chain antibody (scFv16) expressed in Spodoptera frugiperda (Sf9). For higher throughput structure determination, the complexes of receptor/heterotrimer were co-expressed, and purified. Global reconstructions containing receptor, heterotrimer, and scFv16 were obtained for all the ligands tested (Supplementary Fig.). However, due to the inherent flexibility of the receptor across the heterotrimer, the ligand density in the receptor portion of the reconstruction was occasionally inadequate for modeling (Supplementary Fig.). Ligand densities were refined by performing a focused local refinement in cryoSPARC utilizing a loose mask (Supplementary Fig.) around the receptor alone. In that way we obtained two maps for each structure: the first representing a global (receptor:heterotrimer complex) and the second representing local (receptor alone) reconstructions. We deposited maps and models with identical receptor coordinates based on the focused receptor refinement (Supplementary Fig.and Supplementary Tables). For the tryptamines, we). All subsequent receptor-focused reconstructions allowed us to unambiguously model ligand densities as described here (Fig., Supplementary Fig., C and Supplementary Table-4 for data collection and model statistics, and Supplementary Fig.) and resolution estimated by GSFSC (Gold Standard Fourier Shell Correlation) (Supplementary Fig.and Supplementary Fig.). Importantly, ligand placements were verified by the Gemspot pipeline, a docking protocol that utilizes the cryo-EM map as restraints (Supplementary Fig.). Despite the substantial chemical space that is recognized by the 5-HT 2A R, there is a remarkable global structural homology across the tryptamines, ergolines, and phenethylamines at the receptor level. Using the 5-HT:5-HT 2A R structure as a reference, a structural alignment yields a C-α RMSD of 0.6 Å for BOL, 0.6 Å for DMT, 0.6 Å for LSD, 0.5 Å for mescaline, 0.5 Å for psilocin, and 0.8 Å for RS130-180 (Fig.). We also compared our previously published 25CN-NBOH structure to the 5-HT structure and found it exhibits an RMSD of 1.0 Å (Fig.).
Examining the orthosteric site, all ligands make predicted receptor contacts and sit within a pocket defined by D155, F339, and F340, thereby validating more than 30 years of prior biochemical data and computational models for ligand interactions(Figs.). Significantly, we find that psilocin interacts with the G protein-coupled state distinct from that recently proposed by another group for psilocin. We also compared similar previously experimentally validated 5-HT receptor structures with our current structures (Supplementary Fig.). The 5-HT bound 5-HT 1A /5-HT 1D and the 5-CT bound 5-HT 5A structures were recently solvedby cryo-EM and aligning these with our structures, it is evident that the tryptamine substructure inhabits a similar position within the orthosteric pocket (Supplementary Fig.). In addition, we compared the closely related 5-HT 2C receptor bound with psilocinand our structure of psilocin bound to 5-HT 2A R. These structures exhibit significant overlap with the interacting residues in the orthosteric pocket, with psilocin exhibiting a small shift in the tryptamine core compared to the 5-HT 2A structure (Supplementary Fig.). Thus, the overwhelming structural evidence across multiple serotonin receptor families suggests that the functionally active form of all examined tryptamine is in the validated orthosteric site and not in the recently hypothesized extended binding pocket (Supplementary Fig.). To examine the key interacting residues within the orthosteric pocket for tryptamines, we quantified the differences in the contacts and validated the structural models by site-directed mutagenesis and subsequent functional assays (Supplementary Fig.and Supplementary Table). First, we observed that 5-HT, psilocin, and DMT all make consistent ionic interactions with D155and the primate-specific residue S242as previously hypothesized, due to the similar positioning of the indole ring seen in other 5-HT 2A agonist structures. However, when comparing the three tryptamines, we noted that there are slight shifts within the indole ring (Fig.). Most notable is the change in the 4'-OH position of psilocin and the 5'-OH of 5-HT. The hydroxyl moiety on 5-HT sits within the H-bonding range with N343, whereas the psilocin hydroxyl group points towards the amino tail of the molecule (Fig.). This finding is consistent with our mutagenesis studies where we found that the N343A 6.55 mutation influences 5-HT potency, suggesting electrostatic and/or potential water-mediated interaction between the endogenous ligand and N343A(Supplementary Fig.and Supplementary Table). However, that was not found with psilocin. DMT also shows a slight shift in the indole ring compared to 5-HT, as there is no accessory moiety on the indole ring to position the ligand (Fig.).
The LSD analog BOL has been reported to be non-hallucinogenic in humans. Previously, it was speculated that BOL acted as a 5-HT 2A R antagonist as reports of it being administered to humans in doses as high as 10 mg three times daily without any psychoactivity, whereas it subsequently blocked the hallucinogenic actions of LSD. However, recapitulating recently published work, we found that BOL, instead, is a potent 5-HT 2A R G-protein biased partial agonist (Supplementary Fig.), which explains at least in part its ability to block the psychedelic actions of LSD in humans. Within the orthosteric pocket, both ergolines make the expected hydrogen bonding contacts with D 3.32 and S(Fig.). BOL is especially intriguing as it differs by a single atom compared to LSD, bromine at the 2 positions (Figs.) and is nonhallucinogenic. Comparing the binding poses of BOL with LSD in these structures, we observe that both ligands occupy similar binding positions (Fig.). The most notable difference being the Br of BOL, which makes van der Waals contact with I163deep within the orthosteric pocket and F340(Fig.). I163is the I in the canonical PIF motif that undergoes a significant conformational change upon receptor activation, and interaction with this residue could potentially impede receptor activation resulting in weak-partial agonism. In addition, F340is conserved throughout all 5-HT 2 family members and plays a universal role in ligand recognition. When comparing our cryo-EM structures of 5-HT 2A -LSD and D 2 -LSD (Fig.), we observed significant overlap in the overall placement of the ligands. Also, we compared these structures with our recently solved 5-HT 2B -LSD and the recently solved 5-HT 1A -LSD cryo-EM structures. When aligning just the receptor portions of the complexes, we observed a considerable shift upward of the indole part of the ergoline ring (Fig.and Supplementary Fig.) for 5-HT 2B . This is due to two factors: (1) 5-HT 2B does not contain a serine at position 5.46 (it is an A) and loses the interaction with the indole N; and (2) the ring is pulled upwards through hydrophobic interactions with M 5.39 (Fig.). At the 5.39 position in 5-HT 2A R, D 2 R, and 5-HT 1A R, this is a valine, and in this regard, it was recently reported that BOL is a potent D 2 R/5-HT 1A R agonist and a 5-HT 2B antagonist. We hypothesize that BOL may also sit higher within the orthosteric pocket and act as an antagonist at 5-HT 2B , either potentially blocking the structural rearrangement of I163or engaging these residues in a way that does not allow activation. This hypothesis was evaluated biochemically by testing reciprocal V235M 5.39 /5-HT 2A and M218V 5.39 /5-HT 2B mutations. Supporting this hypothesis, we found that switching these residues transforms BOL into a partial agonist at 5-HT 2B and greatly attenuates the potency and efficacy for BOL at 5-HT 2A (Supplementary Fig.and Supplementary Table). Importantly, similar effects were not seen with LSD as it is not stabilized in this deep portion of the orthosteric pocket. This binding mode could, conceivably, be exploited as an approach for creating ergoline-based 5-HT 2A selective agonists through further exploration of the space deep within the orthosteric pocket.
The phenethylamines mescaline and RS130-180 exhibit similar binding poses (Fig., B and Supplementary Fig.). Both mescaline and RS130-180 contact D(Fig.), whereas mescaline also contacts Son helix 5 (Fig.). Moreover, the mescaline-bound complex revealed a hydrophobic interaction with L229 on extracellular loop 2 (ECL2) through its 3'-position methoxy group (Fig.). This interaction was previously found to be responsible for creating a lid on top of the orthosteric pocket with 5-HT 2A and 5-HT 2B LSD crystal structures to 'trap' LSD in the binding site. This ECL2 lid was found to be responsible for LSD's long residence time on the receptor and, potentially, its extended duration of action. In addition, L229 has a tight hydrophobic interaction with F234, which we hypothesized may be important for transmitting the signal through TM5 during receptor activation (Fig.). Unexpectedly, the L229A mutant transformed mescaline from an agonist to an inverse-agonist (Fig.). Furthermore, site-directed mutagenesis of F234Ashowed that this residue is important for ligand-dependent activation of the receptor for both 5-HT and mescaline, indicating that stabilization of this residue is important for transmitting activation signals through TM5. Because mescaline is stabilized by the same residue essential for LSD's prolonged receptor residence time, we wondered if the same process might apply to mescaline. As radioactive mescaline is not available, instead, we utilized a 5-HT 2A -κOR chimeric receptor and nanobody 6 (Nb6) pair to probe the conformational states of the receptor in a ligand-dependent manner as previously described. We have previously characterized this Nb6 5-HT 2A -κOR chimera BRET pair and found it to be functionally relevant for 5-HT 2A R signaling. This BRET-based assay affords us the ability to probe the active→inactive state transition and to compare the kinetics of ligand-stabilized conformations. Utilizing this technology, we probed 5-HT, mescaline, and LSD (Supplementary Fig.). After a baseline for 15 min, the agonist was added for a period of 50 mins. Immediately following agonist administration Nb6 dissociates from the inactive state consistent with an inactive→active state transition. After 50 min of agonist exposure, the potent 5-HT 2A antagonist risperidone (10 μM) was added, and the 5-HT and mescaline BRET response quickly returned to near basal levels. However, that is not seen with LSD similar to direct radioligand binding studies indicating a prolonged ligand residence time (and presumed long-term stabilization of the 'active' state). Our findings indicate that the closing of the ECL2 lid is essential for mescaline's interaction with L229 but that this interaction does not significantly modify apparent conformational transition kinetics. Furthermore, we examined all the ECL2 lids across all the structures and observed that they are closed (Supplementary Fig.) adding further evidence for the importance of this conformation in the transducer-coupled state. However, simple engagement with L229 may not drive dissociation kinetics for every class of compounds like it does with ergolines, but this has yet to be examined from the point of view of ECL2 loop dynamics. A β-arrestin-biased 5-HT 2A agonist stabilizes a Noncanonical State We next examined the binding mode of the βarrestin-2 biased N-benzylated phenethylamine, RS130-180 (see Supplementary Fig.for NMR validation and Supplementary Fig.for the synthetic scheme), which has shown utility as an in vitro tool compound, although it has suboptimal in vivo pharmacokinetic properties. RS130-180 was optimized for potency and bias from ZINC000341335936 ('5936), which was identified from a large-scale docking campaign. One particularly interesting feature of the N-benzylated phenethylamines is their influence on the positioning of the "toggle-switch" tryptophan (W). Using the structural information from the 25CN-NBOH 5-HT 2A -Gq complex, we hypothesized that 'pushing' on the toggle switch was essential for its potency and efficacy. However, this pushing on Whas not been replicated in any of the previously solved 5-HT 2A structures, including the ones in this work. To understand the potential structural features contributing to RS130-180's bias profile, and potential receptor changes during G-protein activation, we obtained a cryo-EM structure with RS130-180. Like 25CN-NBOH, we found that RS130-180 directly interacts with W. However, due to the steric hinderance, and the additional bulk of RS130-180, the toggle switch adopts an entirely downward facing position unseen in any reported 5-HT 2A R structures. The displacement of Wcauses an inward rotation of F332(or the F in the PIF motif) towards the receptor core. This inward rotation of F332 6.44 pushes Y380(Y of the NPxxY motif) outwards compared to the active state 5-HT structure and all other active state structures (Fig.). We observed that this outward shift at the bottom of TM7 is reminiscent of the inactive state crystal(Fig.). That creates a non-optimal conformation for G-protein activation as TMs 5 and 6 are in activate state conformations whereas TM7 is in an inactive state configuration. This type of intermediate state, the non-canonical (NC state), has been proposed through MD simulations and observed in recent structures. To our knowledge, this represents the first occurrence in which the NC state is stabilized by an arrestin-biased ligand, as RS130-180 imperfectly stabilizes the Gq conformation, thereby favoring arrestin signaling and, perhaps, explaining the observed bias of the ligand.
Examining the structural data in aggregate, we noticed distinct chemotype-and ligand-stabilized interactions based on the modeled coordinates. All ligands examined contact D, which is a long validated contact within the orthosteric pocket(Fig.)(Supplementary Fig.for interaction maps prepared by Maestro (Schrodinger)). Interestingly, 5-HT is the singular ligand in the structures that we solved that contacts N, and this contact could play an important role in 5-HT 2 family specific signaling as it is non-conserved throughout all 5-HT receptors. Another residue, S, which is unique to the primate 5-HT 2A receptor and not found in either the mouse or rat, interacts with all ligands examined, except for the N-benzylated phenethylamines 25CN-NBOH and RS130-180. In addition, psilocin was the only ligand to contact T160. L229made hydrophobic contacts with DMT, mescaline, LSD, and BOL. Finally, a potential targeting mechanism for ergolines lies in W151as both LSD and BOL wereand Supplementary Table. The bottom portion is close-up panels showcasing the structural mechanism that RS130-180 utilizes to achieve its arrestin bias. found to interact with this residue, whereas the other chemical classes do not. Finally, we validated long-proposed and important pi-stacking interactions in the orthosteric pocketwith F340with all of the ligands except mescaline. However, mescaline does make a pication interaction with the adjacent F339. These results reveal potential avenues for selective drug design by targeting various residues within the orthosteric pocket. The finding that bias can arise in concert with observable conformational changes in 5-HT 2A structures, as exemplified by the NCstate produced by RS130-180 in this work, led us to investigate the potential temporal effects of ligand bias. Figureshows bias plotsas a function of time for all the ligands presented in this work and 25 CN-NBOH. To create these plots, dose-response curves utilizing our TRUPATH system (Gαq-heterotrimer dissociation) and β-arr2 recruitment were measured (see Supplementary Fig.for dose-response curves and Supplementary Tablefor fit parameters) in a timedependent fashion. Interestingly, utilizing these assays we noticed that all the tested ligands exhibited Gαq bias except for RS130-180 and 25CN-NBOH. Both ligands are N-benzylated phenethylamines and directly interact with W336 6.48 (the "toggle switch"), causing a deviation in the tryptophan conformation compared to the other agonists, potentially revealing a mechanism for producing β-arr2 biased compounds. Subsequently, we noticed that the G-protein signaling for both compounds catches up at later time points as the plots move towards "even" signaling, suggesting a kinetic component to this type of bias. It was previously shown in D2R that ligand binding kinetics can modulate signaling bias through receptor occupancy timei.e., the longer the ligand activates the receptor, the longer all the various signal transduction pathways can be activated in the system, leading to changes in bias in a time-dependent manner. However, due to both RS130-180 and 25CN-NBOH molecularly targeting the important toggle switch residue (W336), which is important in modulating inac-tive→active state transition, we hypothesized that the observed bias is being achieved through a conformational stabilization of the receptor and not the ligand binding kinetics. To probe both ligand dissociation and the potential inactive→active state conformational transitions we utilized the 5-HT 2A -κOR and Nb6 BRET pair against the remaining ligands as well as 25CN-NBOH. We found that both 25CN-NBOH and RS130-180 dissociated rapidly from the receptor after the addition of risperidone (Supplementary Fig.), unlike the ergolines LSD and BOL, suggesting shorter receptor occupancy times. This would indicate that in the case of 25CN-NBOH and RS130-180, the change in the observed signaling bias is not due to ligand kinetics, but through the stabilization of a specific conformational state of the receptor, which in turn modulates the kinetics of the activation/recruitment of the transducer. In addition, we observed that several agonists induce different maximal responses when compared to other BRET probes (i.e., β-arr2 recruitment or heterotrimer dissociation) (Fig.). To best compare % activation of these three datasets, we matched them for time (in this case ~55 min for Nb6 or 60 min for Gq/β-arr2) and ligand concentration. In examining this plot, we noticed that the NB6 response for all compounds closely mimicked the β-arr2 response, except for LSD (Fig., Supplementary Fig., and Supplementary Fig.). In addition, both the N-benzylated phenethylamines (25CN-NBOH and RS130-180) showed a larger response than the β-arr2 response. Because Nb6 is sensing the inactive→active state transition, the variances in maximal responses suggest a fundamental difference in how each ligand stabilizes this conformational transition (Fig.). This corroborating biochemical evidence from the structural data further suggests these compounds adopt distinct conformational states upon ligand activation, leading to their signaling behaviors. To explore the features of the different conformational transitions observed by the ligand-dependent NB6 dissociation, we examined the observable structural distributions available to us in our cryo-EM ensembles (Fig.and Supplementary Fig.). Although we acknowledge that a cryo-EM structure is an average of many microstates, analyzing the distribution of conformations is informative. In addition, we also realize that these structures do not utilize the native Gαq as a transducer, which may bias conformations of the receptor, but it should be noted that the use of mini-Gαq does not change the efficacy/potency of 5-HT 2A R signaling. Unlike 3DVAcomponent, 3DFlexmodels the deformation flow field of the consensus map in a nonlinear fashion, capturing more information about the overall movement of the conformational ensemble of the particle stack. Once the ensemble of maps was generated, we used phenix.refineand did two rounds of rigid body refinement into the maps. We observed that all structures contained conformational "rocks" and "twists," as has been reported for other GPCRs. However, each structure exhibited distinct patterns of these "rocks" and "twists" containing ligand-dependent variations in both the direction and magnitude (Supplementary Fig.). Importantly, both the direction and magnitude of the "rock" and "twist" patterns were not affected by the size of the dataset as the 5-HT-complexes exhibited large movements (~157 K particles) whereas BOL (~603 K particles) was relatively stable. To compare the high-dimensional structural information, we employed dimensionality reduction (UMAP 46 ) (Fig.). Remarkably, despite aligning all structures before dimensionality reduction, each ligand stabilizes distinct conformational spaces (Fig.).
The cryo-EM structures presented represent the known agonist classes related to psychedelic drug action at the 5-HT 2A R. Utilizing this structural information will lead to insights into structure-aided drug design for novel psychedelic therapies. Although the exact mechanisms of action of psychedelic drugs are still unknown, this work represents a leap forward in the molecular understanding leading to this phenomenon. This work provides several mechanistic breakthroughs in understanding 5-HT 2A R actions by various psychedelics: mescaline interacts with the ECL2; a method for designing ergolinebased compounds for 5-HT 2A selectivity over 5-HT 2B revealed by the non-hallucinogenic BOL; a potential mechanism of partial agonism through BOL targeting I 3.40 a key molecular switch (PIF-motif) for class A GPCR activation, and validation that targeting W(toggle switch) by N-benzylated phenethylamines produces arrestin biased compounds through modulation of receptor structure. Moreover, we have observed that different agonists stabilize varying conformational ensembles of the receptor coupled to the active-state heterotrimer. Due to a recent resurgence in interest for utilizing psychedelic compounds as intermittently dosed therapeutics to treat depression, anxiety, addiction, cluster headaches, and many other neuropsychiatric disordersit essential that we understand the details of action for these historically ostracized compounds. This study reveals the molecular underpinnings of ligand-receptor interactions while offering insights that will accelerate the pursuit of safer psychiatric therapeutics.
The expression and purification of scFv16 was carried out as described previously. In short, Sf9 cells were infected with an MOI of 3. 96 h post-infection (hpi) the cells were centrifuged at 3000 × g for 15 min. The supernatant media was collected, and the pH was adjusted to 7.8 using a Tris base. Chelating agents in the media were quenched with the addition of 1 mM NiCl 2 and 5 mM CaCl 2 and allowed to stir for 1 hr at RT. The precipitants were removed by additional centrifugation. After centrifugation, 1-2 mL of His60 Ni Superflow Resin was added to the media and allowed to stir overnight at 4 °C. The resin is collected the next day and washed with 20 CV of 20 mM HEPES pH 7.5, 500 mM NaCl, and 10 mM imidazole. The protein is then eluted with 20 mM HEPES pH 7.5, 100 mM NaCl, and 250 mM imidazole. The protein is then concentrated for further purification using a Superdex 200 16/60 column (GE). The peak fraction is collected/concentrated and stored at 4 °C or flash frozen to be kept in the -80 °C for future use.
All constructs used in a complex generation were published in our previous structural work on the 5-HT 2A R. An MOI of 3 (5-HT 2A R receptor) and an MOI of 1.5 (miniGq heterotrimer) was used to infect Sf9 cells. 48 hpi cells were harvested through centrifugation at 3000 × g for 15 min and washed with HN buffer (20 mM HEPES pH 7.4, 100 mM NaCl), and stored at -80 °C until purification. Cells were thawed on ice and incubated with a buffer containing 20 mM HEPES pH 7.5, 50 mM NaCl, 1 mM MgCl2, 2.5 units of Apyrase, proteinase inhibitors (500 μM AEBSF, 1 μM E-64, 1 μM Leupeptin, 0.15 μM Aprotonin), and between 5-50 μM of ligand of choice at RT on a rotator. After 2 hr, the cell suspension was dounce homogenized, and membranes were collected by centrifugation at 30,000 rpm (71,000 × g) for 30 min (Ti45 rotor). The membrane pellet was collected and solubilized in 40 mM HEPES pH 7.5, 100 mM NaCl, 5% (w/v) glycerol, 0.06% cholesteryl hemisuccinate (CHS), 0.6% lauryl maltose neopentyl glycol (LMNG), 500 μg of scFv16 purified protein, and between 5-50 μM ligand of choice for 5 hr at 4 °C. The solubilized lysate was clarified through centrifugation for 1 hr at 60,000 rpm (264,000 × g) at 4 °C (Ti70 rotor). 20 mM imidazole and 2.5 μL of PNGaseF was then added to the supernatant and incubated overnight at 4 °C with Talon IMAC resin. The next day the resin was collected and washed with 25 CV of 20 mM HEPES pH7.5, 100 mM NaCl, 30 mM imidazole, 0.01% (w/v) LMNG, 0.001% (w/v) CHS, with 5-50 μM of ligand of choice, and 5% glycerol. The protein was then eluted with the same buffer but with 250 mM imidazole. After elution, the protein was immediately concentrated and subjected to SEC (Superose 6 equilibrated 20 mM HEPES pH 7.5, 100 mM NaCl, 100 μM TCEP, 0.00075% (w/v) LMNG, 0.00025% (w/v) glycol-diogenin (GDN), 0.00075% CHS, and 5-50 μM ligand of choice). The appropriate peak fractions were collected and subjected to cryo-EM analysis. The protein was then concentrated to 3-5 mg/mL, and an additional 5-50 μM of ligand is added and allowed to incubate for 1-2 hr before making grids.
Quantifoil R 1.2/1.3 Au 300 holey carbon film grids were glow discharged and individually frozen in a 60/40 ethane propane mixture using a Vitrobot mark IV (FEI). The blot time of each grid ranged from 2.5-5 s with the humidity set to 95% at 4 °C. All images were collected on a 200 keV G3 Talos Arctica with a Gatan K3 direct electron detector with a pixel size of 0.88 Å for ~2.7 seconds for 60 subframes with a total exposure of ~50 electrons/ Å 2 . A multi-shot array utilized and recorded automatically using SerialEM. The micrographs were manually curated, inspected, and processed using cryoSPARC v3.1 or v4.1. All data collection statistics and example processing tree canand Supplementary Table. B Cartoon of Nb6 dissociation and the time course of 5-HT. Data represent N = 3 biological replicates and mean ± SEM. Created in BioRender. Gumpper, R. (2025)(C) The calculated activation of each signaling assay matched for time (55 mins for Nb6 and 60 min for Gq/β-arr2) and concentration. For the NB6 assays, this data represents N = 3 biological replicates. For Gq/β-arr2, the calculated activation is based on the fit parameters at 60 minutes and matched for concentration. These fit parameters can be found in Supplementary Fig.and Supplementary Table, where the number of biological replicates is also recorded. D Workflow of the output from cryoSPARC 3DFlex to generate models for downstream analysis and dimensionality reduction. E Aggregate data of the flexibility analysis using the Flex-refine protocol in cryoSPARC reveals conformational selectivity of the ligands. This is a UMAP of the models derived from the maps output by 3DFlex. Each color represents a different ligand, and abbreviations are as follows: mescaline (MSC), N,N-dimethyltryptamine (DMT), serotonin (5HT), 2-bromo-LSD (BOL). be found in Supplementary Fig.and Supplementary Tables. The processing tree is like our previously published structures. A gold-standard Fourier shell correlation cutoff of 0.143 Å was used to determine global resolution. The maps and models were validated through using the half-maps and B-factor sharpened maps using Mtriage in the Phenix Software package. To assist visualization of the maps in Fig., alternative sharpening was performed on the halfmaps using deepEMhancer. Other than the densities shown in Fig.for visualization purposes, only the automatically sharpened maps that are output from cryoSPARC non-uniform refinement or local refinement jobs were deposited and used for model building and subsequent structural analysis.
The initial model for each structure was derived from our previously published structure coordinates from PDB 6WHA. Initial placement of the complex was done using the fit-to-map function in ChimeraX. The model was then subjected to Phenix Real-space refinement with a rigid body turned on for a single round. Further modeling and structure validation was done using COOT, with a final round of refinement model cleanup done using the ChimeraX plugin ISOLDE and COOT. For ligands that did not already have restraints deposited in the PDB, they were generated using Phenix Elbow. The final model was further validated using Phenix Mtriage to examine the map-to-model quality and model statistics were generated by MolProbity. All the structure figures were generated using ChimeraX. Model statistics can be found in Supplementary Tables.
To capture potentially non-linear and complex motions of the receptor-heterotrimer, we utilized cyroSPARC's 3DFlex pipeline to generate the conformational landscape from each set of particles. Uniformly, each dataset was down-sampled 4.3 x (i.e., from a box size of 312 pixels to 72 pixels) during the 3D-Flex preparation step. For the flex-mesh preparation, the mask was set to a level to exclude the micelle, while keeping the protein parts. During Flex training 3 latent dimensions were used, and the centering strength was adjusted to make sure that all particles stayed with -1.5 to 1.5 across the 3 latent dimensions. Finally, Flex Generate was used to output 41 frames per latent dimension for a total of 123 reconstructions per structure. To fit models to begin a more in-depth structural analysis, 6WHA was used to be manually fit into the first frame by ChimeraX. Utilizing phenix.refine the same model was then fit into each frame via 2 cycles of rigid body refinement. All of these PDBs were then input as trajectories into MDAnalysis 62 , and every structure was aligned to the original 5-HT structure. From here, UMAP was carried out using only the Cαs. All plots shown were done in matplotlib and Seaborn.
The TRUPATH (BRET2) assays were carried out. Briefly, 2 hrs prior to transfection in 10 cm dishes media was replaced with DMEM containing 1% dialyzed FBS. A ratio of 1:1:1:1 of receptor:Gα:Gβ:Gγ with 1 μg of each plasmid was transfected using Transit 2020 (Mirus Biosciences) following the manufacturers protocol. The next day the cells were plated into poly-L-lysine coated 96-well plates at a concentration of 40-50 k cells per well in DMEM containing 1% dialyzed FBS. 48 h posttransfection, the media was removed from the 96 well plates and covered with 60 μL of assay buffer (1x HBSS, 20 mM HEPES pH 7.4) and allowed to incubate at 37 °C for 10 minutes. After the incubation of 30 μL of 3X drug binding buffer (1X HBSS, 20 mM HEPES pH 7.4, 0.3% (w/v) BSA, 0.03% (w/v) Ascorbic Acid, ligand to be tested) was added and allowed to incubate for 10 min at RT. For specific kinetic assays the ligand was allowed to incubate with the cells for the listed amount of time before reading. Coelenterazine 400a was added to each well, followed by an additional 10-minute incubation. The BRET ratio was read (395 nm/510 nm) on a PHERAstar FSX for a total of 5 scans. The final read was taken for each plate for further analysis.
The β-arrestin recruitment assays were carried out using the same protocol as mentioned for the TRUPATH assays above, except a 1:2.5:5 ratio of receptor:GRK2:β-arr2 was used during transfection, and coelenterazine H was used as the RLuc substrate. For the kinetic time point assay, each ligand was incubated with the cells for that specific amount of time before reading. Nb6 Dissociation from 5-HT 2A -κOR chimera Utilizing the previously made and optimized constructs, the κOR ICL3 was inserted into the ICL3 of 5-HT 2A 33 . This allows for Nb6 binding and acts as a conformational sensor for inactive/active states, as previously shown. HEK293T cells were co-transfected and plated similarly to the BRET2 assays but with the chimeric 5HT 2A -κOR-Rluc and Nb6-mVenus ratio 1:5, respectively. Kinetic traces were recorded on the Pherastar FSX for a total of 120 minutes every 30 s: 15 mins for baseline, 50 mins with agonist addition, and 55 mins with 10 µM antagonist (risperidone) addition using BRET1 plus optic module.
All commercial chemical reagents and solvents were used for the reactions without further purification. Flash column chromatography was performed on a Teledyne ISCO CombiFlash Rf+ instrument equipped with a 220/254/280 nm wavelength UV detector and a fraction collector. Normal phase column chromatography was conducted on silica gel columns with either hexane/ethyl acetate or dichloromethane/methanol as eluent. All final compounds were purified with preparative high-performance liquid chromatography (HPLC) on an Agilent Prep 1200 series with the UV detector set to 220/254 nm at a flow rate of 40 mL/min. Samples were injected onto a Phenomenex Luna 750 × 30 mm, 5 μm C18 column, and the gradient was set to 10% of acetonitrile in H 2 O containing 0.1% TFA progressing to 100% of acetonitrile. For chiral separation, a Lux R 5 μM i-Amylose-3 column was used and samples were separated with preparative highperformance liquid chromatography (HPLC) on an Agilent Prep 1200 series with the UV detector set to 220/254 nm at a flow rate of 40 mL/min (method: solvent: H 2 O (0.1% TFA): CH 3 CN, 0 -30 min (90% :10% -0%: 100%), 30 min -35 min (0%: 100%). All final compounds prepared had purity > 95% as determined by an Agilent 1200 series system with a DAD detector and a 2.1 mm × 150 mm Zorbax 300SB-C18 5 μm column for chromatography and high-resolution mass spectra (HRMS) that were acquired in positive ion mode using an Agilent G1969A API-TOF with an electrospray ionization (ESI) source. Samples (2 μL) were injected onto a C18 column at room temperature, and the flow rate was set to 0.4 mL/min with water containing 0.1% formic acid as solvent A and acetonitrile containing 0.1% formic acid as solvent B. Nuclear magnetic resonance (NMR) spectra were acquired on Bruker DRX 400 MHz for proton ( 1 H NMR) and 101 MHz for carbon ( 13 C NMR). Chemical shifts for all compounds are reported in parts per million (ppm, δ). The format of the chemical shift was reported as follows: chemical shift, multiplicity (s = singlet, d = doublet, t = triplet, q = quartet, m = multiplet), coupling constant (J values in Hz), and integration. All final compounds had > 95% purity using the HPLC methods described above. To a solution of commercially available 2,5-dimethoxy-N,N-dimethylaniline (1) (400 mg, 2.2 mmol) in acetone (5 mL) was added N-bromosuccinimide (NBS) (431 mg, 2.42 mmol, 1.1 equiv). The resulting suspension was stirred at 0 o C for 2 h. The reaction mixture was concentrated and purified by flash chromatography (hexane / ethyl acetate = 10:1 to 3:1) to afford 4-bromo-2,5-dimethoxy-N,N-dimethylaniline (2) as a yellow solid (464.1 mg, yield 82%). Intermediate 2 (464.1 mg, 1.78 mmol) was then dissolved in toluene (6 mL) and water (3 mL), then (2-((tert-butoxycarbonyl)amino)ethyl)trifluoroborate potassium (670 mg, 2.67 mmol, 1.5 equiv), Pd(OAc) 2 (40 mg, 0.178 mmol, 0.1 equiv), Ruphos (166 mg, 0.356 mmol, 0.2 equiv), and cesium carbonate (1.74 g, 5.34 mmol, 3 equiv) were added. The reaction mixture was stirred at 110 o C overnight. The reaction mixture was then cooled to room temperature, and extracted with ethyl acetate, washed with brine, dried over Na 2 SO 4, filtered, and then evaporated. The residue was purified by flash chromatography (hexane/ethyl acetate = 3:1 to 1:1) to afford tert-butyl (4-(dimethylamino)-2,5-dimethoxyphenethyl)carbamate (3) as a yellow solid (510 mg, yield 88%). To a solution of compound 3 (150 mg, 0.46 mmol) in DCM (1 mL) was added TFA (1 mL). The resulting suspension was stirred at rt for 1 h. The resulting mixture was concentrated and purified by reverse phase column (10%-100% acetonitrile/ 0.1% TFA in H 2 O) to afford 4-(2aminoethyl)-2,5-dimethoxy-N,N-dimethylaniline (4) as a yellow oil (100 mg, yield 97%). To a solution of compound 4 (100 mg, 0.45 mmol) in methanol (1 mL) were added commercially available 2-(prop-2-yn-1yloxy)benzaldehyde (5) (71.4 mg, 0.45 mmol, 1 equiv), Et 3 N (5 drops) and AcOH (10 drops), the mixture was stirred for 1 h at rt, then sodium cyanoborohydride (84.1 mg, 1.3 mmol, 3 equiv) was added. The resulting mixture was stirred at rt for 1 h. The mixture was filtered through celite. The resulting mixture was purified by prep-HPLC (10% -100% acetonitrile/ 0.1% TFA in H 2 O) to afford the final compound in TFA salt form (121 mg), then added 1 mL saturated NaHCO 3 solution, extracted with ethyl acetate, washed with brine, dried over Na 2 SO 4, filtered, and evaporated. Then added 1 equiv. HCl.dioxane (4 N solution, 50 μL), and stirred at rt for an additional 1 h. The reaction mixture was then concentrated and the residue lyophilized to afford 2,5dimethoxy-N,N-dimethyl-4-(2-((2-(prop-2-yn-1-yloxy)benzyl)amino) ethyl)aniline (RS130-180) as a light yellow solid (81 mg, 2HCl salt, yield 41%).H NMR (400 MHz, methanol-d 4 ) δ 7.46 (q, J = 7.6, 6.8 Hz, 2H), 7.36 (d, J = 4.3 Hz, 1H), 7.24 (t, J = 4.1 Hz, 2H), 7.09 (d, J = 6.7 Hz, 1H), 4.91 (t, J = 3.3 Hz, 2H), 4.30 (s, 2H), 4.01 (s, 3H), 3.91 (s, 3H), 3.29-3.26 (m, 8H), 3.11-3.06 (m, 3H).C NMR (101 MHz, MeOD) δ 156.02, 152. 13, 144.66, 131.60, 131.15, 129.45, 127.91, 121.51, 119.48, 115.51, 112.39, 103.92, 77.90, 76.42, 56.23, 55.76, 55.70, 46.22, 46.16, 44.99, 26.88
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Create a free account to open full-text PDFs.
Ley, L., Holze, F., Arikci, D. et al. · Neuropsychopharmacology (2023)
Garcia-Romeu, A., Davis, A. K., Fire Erowid et al. · Journal of Psychopharmacology (2019)
Gonza ´lez-Maeso, J., Weisstaub, N. V., Zhou, M. et al. · Neuron (2007)
Preller, K. H., Burt, J. B., Adkinson, B. et al. · eLife (2018)
Kim, K., Che, T., Panova, O. et al. · Cell (2020)
Cao, D., Yu, J., Wang, H. et al. · Science (2022)
Nichols, D. E. · Current Topics in Behavioral Neurosciences (2017)
Lyu, J., Kapolka, N., Gumpper, R. et al. · Science (2024)
Wacker, D., Wang, S., Mccorvy, J. D. et al. · Cell (2017)
Kaplan, A. L., Confair, D. N., Kim, K. et al. · Nature (2022)
Lewis, V., Bonniwell, E. M., Lanham, J. K. et al. · Cell Reports (2023)
Papers in Blossom that reference this study
Yacoub, J., Bray, E., Bayyat, J. et al. · Biorxiv (2026)
Jain, M. K., Gumpper, R. H., Slocum, S. T. et al. · Neuron (2025)
McCulloch, D. E-W., Larsen, K., Johansen, A. et al. · MedRvix (2025)