Introduction

Light-activated drugs offer powerful tools to modulate receptor activity reversibly and selectively1,2,3,4,5, enabling both the dissection of complex signaling networks and the development of therapeutics with minimal off-target effects3,5,6,7,8,9,10. By incorporating a photoswitchable moiety such as azobenzene into a bioactive molecule, drug activity can be reversibly controlled with specific wavelengths of light11,12,13,14,15. Upon irradiation, the azobenzene moiety undergoes cis-trans isomerization16,17, inducing structural and electronic changes that can alter ligand-receptor interactions and functional outcomes at diverse targets, including G protein-coupled receptors (GPCRs)18,19,20,21,22,23,24,25, ion channels26, membrane transporters27, and enzymes28,29.

Despite such successful applications, the rational design of photoswitchable ligands exhibiting efficacy switch behavior remains challenging. Ideally, one would like to design a ligand that switches from antagonist to agonist conformation simply through light exposure. In practice, this is done empirically19,30,31,32,33,34, without knowing the molecular principles that govern whether a given photochemical isomer will act as an agonist or antagonist.

A striking illustration of this knowledge gap is provided by two recently reported photoswitchable N,N-dimethyltryptamine derivatives targeting the human serotonin 5-HT2A receptor (Fig. 1a)25. The 5-HT2A receptor is a class A GPCR with critical roles in neuropsychiatric function, serving as the primary target for both atypical antipsychotics and psychedelic compounds35,36. These two closely related compounds, para-methoxy-5-azo-DMT (1, Fig. 1b) and meta-methoxy-5-azo-DMT (2, Fig. 1b), differ only in the position of a single methoxy substituent on the azobenzene moiety. Yet, this seemingly minor structural variation produces dramatically different pharmacological outcomes upon photoisomerization. In cell-based β-arrestin recruitment assays25, 1 behaves almost like an on/off switch: its cis isomer exhibits moderate agonist activity (~45% of the response elicited by the partial agonist LSD), whereas its trans isomer yields only weak antagonism (~20% of LSD’s maximal response). In contrast, 2 retains substantial agonist activity in both isomeric states (cis ~65% and trans ~55% of LSD’s response), failing to achieve the desired “off” state. This inability to fully suppress receptor activation severely limits its pharmacological applications.

Fig. 1: The 5-HT2A receptor system and the photoswitchable ligands.
Fig. 1: The 5-HT2A receptor system and the photoswitchable ligands.
Full size image

a Representation of the X-ray inactive (PDB id: 6A9444) and cryo-EM Gq-coupled active (PDB id: 6WHA45) 5-HT2A structures. b Chemical structures of the photoswitchable ligands para-methoxy-5-azoDMT (1) and meta-methoxy-5-azoDMT (2) in their trans configuration, with atom numbering used throughout. The labels of the ligand atoms used in the manuscript are also reported. c Key microswitches analyzed in this work, including the Trp3366.48 toggle switch, Arg1733.50-Glu3186.30 ionic lock, Cys3226.34-Arg1733.50 TM6 displacement, and Na+ pocket.

Recent evidence suggests that the exact binding mode of a ligand within the orthosteric binding pocket of a GPCR can dramatically influence its signaling outcome37. Structural studies have shown that traditional, non-photoactive ligands may engage either deeply buried subpockets or more superficial vestibular regions of the receptor, and that this binding pose diversity can modulate ligand affinity, efficacy, and even establish a signaling bias in GPCRs other than 5-HT2A38,39,40. Recently, a photoswitchable drug was designed to achieve signaling bias in the 5-HT2A receptor itself41. Based on these insights, we hypothesized that the divergent efficacy profiles of compounds 1 and 2 towards 5–HT2A arise also from isomer-specific differences in the binding depth of the receptor pocket.

To test this hypothesis, we performed all-atom molecular dynamics (MD) simulations of compounds 1 and 2 in both cis and trans configurations (trans–1, cis–1, trans–2, and cis–2) bound to 5-HT2A inactive and active states. Our simulations reveal that the para versus meta positioning of the methoxy substituent fundamentally alters ligand anchoring within the binding pocket. In the inactive receptor, the para-methoxy group of trans-1 enables formation of a hydrogen bond network involving Asp2315.35 in extracellular loop 2 and Thr1603.37 at the base of the orthosteric pocket (Fig. 2). This dual anchoring mechanism pulls trans-1 significantly deeper into the binding site compared to trans-2, which lacks these stabilizing interactions due to its meta substitution pattern. After photoisomerization, both cis isomers adopt shallower, agonist‑like poses that destabilize the inactive state; a similar behavior can also be found for the trans isomers in the active receptor. Instead, cis‑2 forms a persistent hydrogen bond with Thr1603.37 in the active receptor that allows for a deeper penetration in between TM4 and TM5, whereas cis‑1 cannot form this interaction. Together, our findings reveal the binding pose depth as a key determinant of efficacy for photoswitchable 5-HT2A ligands, providing a mechanistic foundation for the rational design of photopharmacological switches for this receptor and potentially other GPCRs with similar structural motifs to those of 5-HT2A31,42,43.

Fig. 2: Hydrogen bond network analysis in the inactive receptor.
Fig. 2: Hydrogen bond network analysis in the inactive receptor.
Full size image

Kernel density estimation (KDE) distribution of the H-bonding interactions with Thr1603.37, Ser2425.46, Asp2315.35 (O1–H-N-Asp), and Asn6.55 residues for (a) trans-1, (b) trans-2, (c) cis-1, and (d) cis-2 in the inactive receptor. Insets show representative binding poses. Analysis was performed on trajectory frames extracted every 1 ns.

Results

Simulations of ligand-receptor complexes

To understand why compounds 1 and 2 exhibit such different pharmacological profiles despite their structural similarity, we performed eight independent MD simulations of the 5-HT2A receptor: each photoswitchable ligand (1 or 2) in each isomeric form (trans or cis) was simulated in both the inactive and active receptor states. This approach yielded eight distinct systems with a cumulative simulation time of ~80 microseconds (Table S1). The inactive receptor model was based on the antagonist risperidone-bound X-ray structure (PDB ID: 6A9444), while the active state model was derived from the Gq-coupled 5-HT2A cryo-EM structure (PDB ID: 6WHA45). Importantly, we retained the Gq subunit in all active-state simulations, as preliminary MDs showed that its absence caused the receptor to relax towards an inactive-like conformation46. Each receptor complex was embedded in a neuronal-mimetic lipid bilayer, and initial ligand-receptor poses were generated by constrained molecular docking (see Methods and Supplementary Note 1 for detailed protocols). In all cases, the docked pose preserved the hallmark salt bridge between the ligand’s protonated amine and Asp1553.32, a conserved anchor for aminergic GPCR ligands.

Activation microswitches

GPCRs transition between inactive and active states through coordinated conformational changes in conserved “microswitches”, structural motifs that relay ligand binding to G protein coupling. Activation involves several key events: rotation of the Trp3366.48 toggle switch, breaking of the Arg1733.50–Glu3186.30 ionic lock, displacement of an allosteric Na+ ion, and the outward movement of transmembrane helix 6 (TM6) that enables G protein or β-arrestin binding47,48,49,50,51,52,53,54,55.

We therefore monitored these activation markers: (i) the χ1 and χ2 dihedral angles of Trp3366.48, (ii) the local environment of the sodium pocket, (iii) the Arg3.50–Glu6.30 ionic lock (dIL), and (iv) the TM3–TM6 cytoplasmic gap (dTM). These metrics distinguish inactive states (dTM ≈ 9 Å, dIL ≈ 4 Å, χ1 ≈ −80°, Na+ coordinated) from active states (dTM ≈ 12 Å, dIL ≈ 16 Å, χ1 ≈ −160°, Na+ expelled).

To determine whether our simulated binding poses were compatible with receptor activation, we compared them to experimentally determined agonist binding modes. We used the tetracyclic core of LSD, a well–characterized partial agonist, as our reference, aligning against structures in (i) the fully active cryo-EM structure (PDB 9AS356) and (ii) the X-ray crystal structure in the absence of G-protein that resembles features of the inactive state (PDB 6WGT45). By calculating the root-mean-square deviation of each ligand’s indole core relative to LSD (RMSDi), we could quantify how closely our compounds mimicked established agonist binding. Poses maintaining RMSDi < 2 Å were considered to adopt similar, potentially activation-compatible orientations. Here, we discuss only the long-lived binding modes; transient poses are detailed in Supplementary Note 2.

Trans isomers exhibit opposite effects in the inactive receptor through distinct binding depths

In the inactive receptor, trans-1 establishes a unique hydrogen bond network involving its para-methoxy group and the backbone carbonyl of Asp2315.35 (Fig. 2, S1, 43% of occurrence overall the MD trajectories), combined with the indole N1-H bond to Thr1603.37 (70% occupancy) and the conserved salt bridge to Asp1553.32 (99% occupancy). This triad of interactions creates a rigid binding pose with minimal fluctuation (Fig. S2). Extensive hydrophobic contacts (Val2355.39, Val1563.33, Ser1593.36, and Leu230 are present in ≥ 75% of frames) further stabilize this pose. This interaction network (Asp2315.35 and Thr1603.37) pulls trans-1 deeper into the orthosteric pocket compared to typical tryptamine agonists such as LSD, DMT, and psilocin56. This over-insertion prevents trans-1 from engaging the polar triad, i.e., Ser2395.44, Ser2425.46, and Asn3436.55, known to be critical for agonist activity at the 5–HT2A receptor57. The latter, in particular, has been shown to be essential for 5–HT’s agonist potency45. The incompatibility of this orientation with receptor activation is also supported by multiple metrics: RMSDi > 2.0 Å for 55% of the simulation time, narrow TM3-TM6 distance, intact ionic lock, and vertical toggle switch orientation (Fig. S3–S5, Table S2). Together, these observations indicate that trans-1 stabilizes the receptor’s inactive conformation.

In contrast, trans-2 cannot form the same anchoring interactions due to its meta-methoxy position and consequently adopts a different binding mode (Fig. 2b). Trans-2 maintains the Asp1553.32 salt bridge, but exhibits greater conformational flexibility within the pocket and adopts a pose closely resembling that of the partial agonist LSD in the active receptor (RMSDi<2 Å for 78% of simulation time). This allows intermittent engagement of the canonical polar network (Fig. S1): trans-2 transiently forms H-bonds with Thr1603.37 (~10% occupancy) and Ser2425.46 (~20%), while still maintaining hydrophobic contacts with Leu229, Leu230, Val2355.39, Val1563.33, and Val3667.40 (each ~50–65% occupancy). Overall, however, trans-2 on its own did not trigger substantial receptor activation on the simulation timescale, keeping a vertical toggle switch, narrow TM3–TM6, and closed ionic lock (see Fig. S3–S5, Table S2).

Cis isomers partially destabilize the inactive receptor

The cis isomers, upon photoisomerization from trans, adopt shallower binding poses, similar to those adopted by LSD. Cis-1 maintains the critical Asp1553.32 salt bridge (as expected for tryptamine analogs), but it is no longer able to reach Asp2315.35, and its binding pose shifted upward relative to trans-1. Consequently, cis-1 partially restored some agonist-like contacts: it forms stable hydrogen bonds with Ser2395.44 and Ser2425.46 (each ~50–70% occupancy) and even engaged Asn3436.55 (Fig. 2c, S1). Indeed, the indole ring of cis-1 overlays with the tetracyclic core of LSD for ~75% of the trajectory frames (RMSDi <2 Å), highlighting that cis-1 binds in an LSD-like orientation in the inactive state. Cis-1 also engages a network of hydrophobic contacts with residues such as Phe234, Gly2385.43, Trp3366.48, Val1563.33, and Leu229 (occurrences ~45–87%). As a result, this gives cis-1 a well-anchored yet not over-deep pose. However, cis-1 does not fully activate the receptor on its own (see Fig. S4–S5, Table S2). The Trp3366.48 toggle also stays in the inactive (vertical) rotamer in almost all trajectories. Only in one case, the temporary cleavage of the cis-1-Ser2395.44 hydrogen bond causes the ligand to reposition within the binding pocket. This repositioning led to a temporary loss of π–π stacking between Phe2435.47 and Phe3406.52, two aromatic residues that line the binding pocket. Consequently, Trp3366.48 was able to flip to its horizontal, active-like rotamer (Fig. S5, S22).

Cis-2, on the other hand, has the most pronounced agonist-like effect on the inactive receptor. In its inactive-state simulations, cis-2 maintains a similar hydrophobic contact pattern to cis-1 (e.g., with Val1563.33, Leu229, Trp3366.48, occurrences ~50–87%). These contacts keep cis-2 embedded in the orthosteric site, but interestingly, cis-2 exhibits greater mobility within the pocket than cis-1 (Fig. S2). This flexibility correlates with weaker polar interactions: cis-2’s hydrogen bonds to Ser2425.46 and Asn3436.55 are much less frequent (only ~20% and ~28% occupancy, respectively, vs. >50% for cis-1, see Fig. 2d, S1), and even the Asp1553.32 salt-bridge occupancy is slightly reduced (~80% for cis-2 vs. ~99% for cis-1, Fig. S1). The increased mobility of cis-2 induces a rotation of the Trp3366.48 toggle switch from the vertical (inactive) to the “horizontal” (active) rotameric states (Fig. 3, S5). In fact, in more than 75% of simulation frames, the indole core of cis-2 aligns closely with LSD’s ergoline core, indicating an agonist-like pose. Upon the complete loss of Ser2425.46 and Asn3436.55 H-bonds, cis-2 experiences increased mobility and a slight drift. This allows it to re-anchor to Ser2395.44, simultaneously prompting Trp3366.48 to swing downward into an active-like orientation (Fig. 3). This mechanistic shift is facilitated by the disruption of π–π stacking between Phe3406.52 and Phe2435.47, creating space for Trp3366.48 to rotate downward and establish a π–π interaction with Phe2435.47.

Fig. 3: Cis-2 triggers toggle switch rotation in the inactive receptor.
Fig. 3: Cis-2 triggers toggle switch rotation in the inactive receptor.
Full size image

Moving average (window of 20 frames) of the time evolution of four intermolecular distances. Key aromatic or polar side chains are shown as sticks and labeled by the Ballesteros-Weinstein index. Top panel, three representative snapshots extracted from the distance plot. a Initial state: the indole N1–H of cis-2 hydrogen-bonds to either Thr1603.37 or Ser2425.46 while Phe5.47 stacks with Phe3406.52, maintaining Trp6.48 in its vertical, inactive rotamer. b The ligand interacts with Trp6.48, breaking the Phe5.47–Phe3406.52 π-stack. c Trp6.48 completes its flip and establishes a π–π interaction with Phe5.47, while cis-2 anchors to Ser5.43. The behavior of the toggle switch in the other systems is reported in Fig. S5. Analysis was performed on trajectory frames extracted every 1 ns.

However, even though cis-2 clearly destabilizes the inactive receptor by weakening the ionic lock (Fig. S3, Table S2), we did not observe a full TM6 outward movement (Fig. S4, Table S2). In other words, cis-2 on its own presents the strongest agonist effect in the inactive receptor (by flipping Trp3366.48), but complete activation still likely requires the presence of the G protein or other factors.

While the binding modes described above represent the dominant conformations observed across replicas, our simulations also revealed replica-specific secondary poses detailed in Supplementary Note 2. For trans-1 in the inactive receptor, transient disruption of the Asp2315.35 interaction allowed the ligand to tilt toward TM5, forming brief contacts with Ser2395.44 and Gly2385.43. One replica of trans-2 sampled a vestibular configuration characterized by an upward shift within the binding pocket and weakening of the Asp1553.32 salt bridge. For cis-1, rotation around the diazo bond in one replica disrupted the Ser2425.46 and Asn3436.55 hydrogen bonds, inducing a reorientation of the indole ring toward Ser2395.44 or Gly2385.43. Cis-2 showed the largest conformational variability: in two replicas, the ligand transiently disengaged from the conserved Asp1553.32 anchor, drifting into the extracellular vestibule and adopting a weakly bound, solvent-stabilized pose. These reflect the inherent conformational flexibility of the receptor-ligand systems, leading to alternative configurations that reflect local rearrangements rather than substantive divergence between simulations.

Different effects on the allosteric sodium pocket on the inactive receptor

Differences in ligand binding were also reflected in the allosteric sodium (Na+) site and local water distribution, which are known to correlate with GPCR activation58. In all our inactive-state simulations, a Na+ ion remained bound to the side chain of the conserved residue Asp1202.50, located deep in the receptor core. However, the coordination and hydration of this Na+ pocket varied in a ligand-dependent manner (Fig. S6–S9).

With trans-1 bound, Na+ is trapped within a hydrophobic cage formed by Phe3326.44, which acts as a lid above the ion, while Asp1202.50 provides direct electrostatic anchoring. This coordination is further stabilized by Asn3807.49, which forms a critical hydrogen bond with Asp1202.50 that locks the sodium-binding residue in an optimal geometry (Fig. 4a). When this stabilizing network remains intact (approximately two-thirds of the simulation time, see Fig. S6), the sodium ion maintains a restricted coordination sphere with only two water molecules on average, creating a relatively dehydrated pocket characteristic of the inactive state. However, either the disruption of the Asn3807.49-Asp1202.50 hydrogen bond or an outward shift of Phe6.44 triggers a cascade of structural changes: the loss of this interaction destabilizes the Asp1202.50 side chain orientation, weakening its grip on Na+, while simultaneously breaking the Na+-Asn3807.49 and Na+-Phe3326.44 coordination bonds. This conformational change opens a continuous water channel from the extracellular vestibule to the sodium pocket, dramatically increasing the Na+ hydration shell to four water molecules on average (Fig. 4b). For trans-2, this hydration event occurs almost immediately, within the first microsecond of dynamics (Fig. S7), reflecting the ligand’s inability to stabilize the inactive sodium pocket geometry. In essence, trans-2 binding primes the sodium microswitch for activation: the ion remains bound to Asp1202.50, but the pocket is looser and populated by a higher number of water molecules, a state linked to enhanced receptor readiness to transition toward activation. Trans-1, in contrast, resists this priming: it keeps the Na⁺ pocket “closed” and dehydrated far longer (Fig. S6).

Fig. 4: Hydration of the allosteric sodium pocket in the inactive receptor.
Fig. 4: Hydration of the allosteric sodium pocket in the inactive receptor.
Full size image

a The continuous water pathway is blocked, leaving the Na+ ion (blue sphere) tightly coordinated by two water molecules, Asp2.50 and Asn7.49, and Phe6.44 sidechains. b Disruption of the Asn7.49–Asp2.50 hydrogen bond and an outward shift of Phe6.44 open a solvent pathway from the extracellular vestibule to the sodium pocket. Gray cartoons depict the seven-TM bundle; insets enlarge the Na+ site with key residues labeled.

Cis-1 shows a slight increase in water accessibility around its binding site compared to trans-1. On average, ~5 water molecules are present in the immediate vicinity of cis-1 (mostly around the azobenzene linkage), versus essentially zero water molecules for trans-1. Notably, even during these momentary perturbations, cis-1 maintains a grip on the sodium site: like trans-1, it preserves the Asn3807.49–Asp1202.50 interaction and keeps Na+ coordinated by only ~2 water molecules. Furthermore, cis-1 exhibits a unique “backup” Na+ coordination via Ser1623.39, which transiently steps in to ligate Na+ whenever the Asp1202.50-Asn3807.49-Na+ network transiently disrupts. This secondary sodium coordination prevented full hydration of the pocket and in fact results in even lower average Na⁺ hydration for cis-1 than for trans-1. Only the cleavage of both the primary (Asp–Asn) and secondary (Ser1623.39) Na⁺ coordination allows a rush of water into the pocket (Fig. S8).

In cis-2, when the Asn3807.49–Asp1202.50–Ser1623.39 triad is intact, and Phe3406.52 adopts its closed rotamer, Na+ remains tightly bound in a two-water coordination shell (Fig. S9). A first hydration event occurs when Trp3366.48 flips to its horizontal rotamer, creating a solvent path that admits one or two extra water molecules while the Asn3807.49–Asp1202.50 interaction is still strong (Fig. S9, middle panel). The most pronounced hydration, however, is seen when both the Asn3807.49 contact loosens, and Phe3406.52 flips to its open rotamer: the aromatic lid lifts, the conduit widens, and a continuous water channel from the extracellular vestibule to Asp1202.50 forms. In these frames, the Na+ hydration shell expands to 4–5 waters (Fig. 4b, S9).

Ligands binding in the active-state receptor

We next examined the behavior of each ligand in the active 5‑HT2A receptor. All four ligand isomers (cis-1, trans-1, cis-2, and trans-2) consistently stabilized the active 5-HT2A state, as can be seen from dTM and dIL (Fig. S10, S11, Table S2).

In the active complex, trans-1 toggles between its Asp231-anchored pose (39%, weakened because the receptor’s extracellular vestibule has widened) and an LSD-like orientation that restores the Ser2425.46 and Asn3436.55 H-bonds (70% and 34%, respectively, Fig. 5a, S12). Its salt bridge persists for 99% of the simulation. The ligand establishes hydrophobic contacts with Leu228 (51%), Val2355.39 (76%), Val1563.33 (74%), Ser1593.36 (93%), and Trp3366.48 (52%), contributing to the overall stability of the binding pose within the active receptor conformation. A RMSDi < 2 Å was observed for 85% of the simulation time.

Fig. 5: Hydrogen bond network analysis in the active receptor.
Fig. 5: Hydrogen bond network analysis in the active receptor.
Full size image

KDE distribution of the H-bonding interactions with Thr1603.37, Ser2425.46, Asp2315.35, and Asn3436.55 residues, together with a close-by representation of (a) trans-1, (b) trans-2, (c) cis-1, and (d) cis-2 with their main H-bonding interactions in the active receptor. Analysis was performed on trajectory frames extracted every 1 ns.

Within the active receptor conformation, trans-2 forms an H-bond with the sidechain of Ser2425.46 (70% occurrence), a key interaction for many 5-HT2A agonists (Fig. 5b)57. The ligand’s orientation within the orthosteric pocket is reinforced by sustained contacts with Val2355.39 (75%), Ser1593.36 (93%), Ser2425.46 (95%), and Tyr3707.42 (76%). Similar to trans-1, also the binding pose of trans-2 adopted in the active receptor state exhibited RMSDi < 2 Å in 88% of simulated time.

The cis-1 isomer bound to the active receptor exhibits a hydrogen bond with Ser2395.44 (Figs. 5c, S12, 54% of occurrence) and maintains frequent interactions with Asn3436.55 (49% of occurrence). As for the trans isomer, cis-1 establishes a salt bridge between the tertiary amine and Asp1553.32 (99%). This binding pose is further stabilized by persistent hydrophobic interactions with Ser1593.36 (87%), Val1563.33 (87%), Phe234 (66%), and Phe3406.52 (82.5%), reinforcing its orientation within the orthosteric pocket. In fact, the positioning of the indole moiety cis-1 closely resembles that of LSD in the active receptor state, with RMSDi values below 2.0 Å in over 92% of simulation frames. Cis-2, by contrast, penetrates the binding pocket in the active receptor. The meta-methoxy substituent of cis-2 reaches deeper into the cavity between TM4 and TM5 (where cis-1’s para substituent could not), and as a result cis-2 forms a noteworthy hydrogen bond with Thr1603.37 (on TM3) ~ 66% of the time in the active state. Cis-2’s interaction with Thr1603.37, a residue near the base of the orthosteric site, appeared to compensate for its relatively weak hydrogen bond with Ser2425.46 in the active state (only ~16% occupancy for cis-2 vs. ~62% occupancy for cis-1).

Overall, the active-state simulations reinforce that all four isomers are capable of accommodating and stabilizing the active 5-HT2A receptor. None of the ligands acts as an inverse agonist or destabilizer of the active state; rather, indicating that the efficacy differences manifest primarily in the inactive-state interactions described above. In other words, when the receptor is forced into the active state (by G protein or other factors), even trans-1 can bind in a way that does not prevent activation, but when the receptor is inactive, only certain ligand poses can trigger its activation.

Toggle switch behavior in the active receptor

Our simulations offer valuable insights into the behavior of the toggle switch residue Trp3366.48, whose reorientation is considered a key step in Class A GPCR activation. Notably, during the equilibration phase of our simulations of the active receptor, Trp3366.48 transitions from its initial horizontal orientation to a vertical conformation characteristic of the inactive receptor state in the presence of all our ligands (Fig. S13), yet the receptor was clearly active by all other measures (Fig. S10, S11). Likewise, recent high-resolution structures56 show a similar vertical orientation in the active 5-HT2A structure bound to either LSD or serotonin (Fig. S14a, b)56, even though the receptor is fully active.

Rotation of Trp3366.48, though structurally feasible with 1 and 2, does not occur during all our simulations of the active receptor. This is in line with what observed for both ergoline and tryptamine chemotypes, which share a vertical orientation and lack the bulky N-benzyl or phenoxy extensions that would displace the toggle switch (Fig. S14c). Conversely, when cis-2 is placed in the inactive receptor, we occasionally observed Trp3366.48 adopting a downward rotamer (Fig. 3) reminiscent of the orientation captured in structures of bulkier, β-arrestin-biased agonists such as 25-CN-NBOH and RS130-180, whose extended N-benzyl arms press directly on Trp3366.48 (Fig. S14c).

However, we must acknowledge that our microsecond-scale simulations may not fully capture the complete energetic landscape of toggle switch transitions. The free energy barriers separating different Trp3366.48 rotameric states and the precise transition pathways between them would require enhanced sampling methods to characterize comprehensively. Furthermore, while crystal structures provide invaluable snapshots of stable conformational states, they necessarily capture only single, low-energy conformations within what may be a broader ensemble of active states. The crystallization conditions, temperature, and crystal packing forces may also influence which conformations are observed. Our classical molecular dynamics approach, while extensive, operates within the constraints of predetermined force field parameters that may not fully account for the electronic polarization effects or subtle π-π interactions that could influence toggle switch stability. These limitations suggest that the vertical toggle switch orientation we observe may represent one of several possible active state conformations, with the full conformational heterogeneity and dynamic equilibria between states remaining to be fully elucidated through complementary experimental and computational approaches.

Discussion

Functional assays had shown that 1 (para-methoxy) behaves almost like an on/off switch25: its cis isomer activates the receptor moderately (partial agonist), whereas its trans isomer elicits minimal activity (antagonist-like). In contrast, 2 (meta-methoxy) retains substantial agonist activity in both cis and trans forms. Our simulations underscore that the vertical depth of ligand insertion within the orthosteric pocket is a critical determinant for 5-HT2A activation efficacy, acting in concert with specific residue interactions.

In the inactive receptor, trans-1’s para-methoxy group enables formation of a unique hydrogen bond with Asp2315.35 that, combined with Thr1603.37 anchoring, facilitates a deeper insertion of the ligand into the binding pocket than those of typical agonists. This over-insertion positions trans-1 below Ser2395.44 and Ser2425.46 required for activation, effectively locking the receptor in an inactive conformation (Fig. 6b).

Fig. 6: Schematic overview of depth-dependent efficacy switching.
Fig. 6: Schematic overview of depth-dependent efficacy switching.
Full size image

a In the inactive receptor (blue), trans-2 (cyan) maintains an agonist-like shallow pose while cis-2 (purple), in the active receptor (orange), achieves the deepest insertion via Thr1603.37 anchoring. b Trans-1 (cyan) in the inactive receptor inserts deeply, while cis-2 maintains a shallower pose.

Trans-2’s meta-methoxy substitution prevents this deep anchoring, allowing it to maintain an LSD-like pose at the appropriate depth for engaging activation-critical residues (Fig. 6a). While both trans isomers can bind the active receptor in agonist-compatible poses, only trans-1’s ability to adopt the deeply anchored pose in the inactive state explains its antagonist-like behavior in functional assays. The vertical depth of ligand insertion thus emerges as a key determinant of the efficacy among the two trans conformers.

Cis-1, when bound to the active receptor, loses the Asp2315.35 anchor and adopts a more elevated pose within the orthosteric pocket compared to its deeply-bound trans-1 counterpart in the inactive state (Fig. 6b): cis-1’s para-methoxy group and the distal phenyl ring of its azobenzene moiety encounter steric limitations for deep entry into the hydrophobic tunnel between TM4 and TM559 (Fig. S15). This contrasts to cis-2, whose meta-methoxy group penetrates deeply into the hydrophobic tunnel of the active receptor (Fig. 6a, S15), leading to the formation of an H-bond with Thr1603.37, and a deeper insertion in the binding site. Such observed deeper binding was also found for some high-affinity 5-HT2A agonists, which do bind in this region44. Such evidence suggests that the higher agonistic power of cis-2 with respect to cis-1 is linked to the depth of the binding pose. Like for the trans conformers, also for the cis ones, the vertical depth of ligand insertion emerges as a key determinant of the efficacy.

Our findings align with and extend recent structural insights showing that subtle variations in GPCR binding poses can produce dramatically different signaling outcomes37,60.

This insight aligns with previous structural observations showing that the specific binding depth resembling the one of LSD is associated with agonism, whereas deviations are linked to receptor inactivation and antagonism44,60. Notably, deep insertion can stabilize either active or inactive conformations depending on the receptor state, highlighting the importance of context-specific ligand positioning for functional outcomes. The observation that antagonists like risperidone and zotepine bind even deeper than trans-1 (Fig. S16) supports the notion that excessive insertion depth can hinder activation-relevant contacts and favor receptor inactivation44. This hypothesis is further supported by mutational data on the related 5-HT2B receptor60, where increasing cavity space enabled antagonist-to-agonist conversion by promoting an LSD-like binding pose. Mutational simulations of D155A, T160A, and D231P further corroborate the cooperative nature of this anchoring network: removal of individual residues destabilizes neighboring contacts rather than causing immediate ligand dissociation (Supplementary Note 4). Thus, our simulations provide a molecular rationale for the near-antagonist behavior of trans-1 and thereby offer valuable insights for rational GPCR ligand design.

In this context, guiding ligands toward specific insertion depths and interaction profiles that mimic known agonists, such as by targeting or avoiding contacts with residues like Asp231 or modulating access to binding pocket extensions like the hydrophobic tunnel, may enhance potency. Such strategies could enable the development of tailored, high-efficacy 5-HT2A receptor agonists and precisely controlled photopharmacological tools. In fact, recent studies demonstrated that modifying N-benzyl-5-methoxy-tryptamine, a partial agonist of the 5−HT2AR, by incorporating a para-methoxy-azobenzene moiety at the 3-position of its indole ring yielded a photoswitchable ligand with β-arrestin2-biased signaling properties61. The developed ligand spans through the entire orthosteric binding site; however, as observed for 1, the para-methoxy substitution impedes full engagement with the hydrophobic tunnel. Nevertheless, the introduction of an N-benzyl moiety contributes additional stabilizing interactions, thereby promoting a more constrained receptor conformational ensemble. This conformational restriction is reminiscent of the uniform structural states reported for atypical chemokine receptor 3 during β-arrestin recruitment62. Notably, a meta-substituted analog may permit deeper insertion into the hydrophobic tunnel, further stabilizing the receptor–ligand complex; however, such enhanced binding depth may come at the expense of reduced signaling efficacy, as discussed previously. This study, combined with our findings emphasize that the ability to modulate binding depth and interaction networks through strategic ligand modifications could be key to developing functionally selective 5−HT2AR ligands. Conclusively, this aligns with the broader goal in current 5−HT2A research to develop pathway-selective agonists for therapeutic benefit without inducing unwanted hallucinogenic effects63.

Methods

System preparation and simulation setup

All-atom molecular dynamics simulations were performed using the human 5-HT2A receptor in both active and inactive conformational states. The active state model was derived from the Gq protein-bound cryo-EM structure (PDB ID: 6WHA45), while the inactive state used the antagonist-bound X-ray structure (PDB ID: 6A9444).

We retained the Gq subunit in all active-state simulations, as preliminary MDs showed that its absence caused the receptor to relax towards an inactive-like conformation. While the experimental pharmacological data from Gerwe et al.25, were obtained using β-arrestin recruitment assays, the orthosteric binding pocket conformations are largely conserved between G protein- and arrestin-coupled states, with major differences occurring primarily at the intracellular face64,65. Therefore, our analysis of ligand binding poses in the Gq-stabilized active state provides relevant insights into the molecular determinants of efficacy.

Missing loops and residues in the crystal structures were modeled using MODELER 10.566 with templates from related GPCR structures. Protonation states of titratable residues at pH 7.0 were assigned using MCCE (Multi-Conformation Continuum Electrostatics)67.

Each receptor system was embedded in a HEK cell-mimetic lipid bilayer to provide a physiologically relevant environment. The bilayer composition is: 26% cholesterol, 30% POPC, 31% POPE, 5% POPS, 2% POPG, and 6% SSM. Bilayers were constructed using the CHARMM-GUI membrane builder68 with dimensions of 95 × 95 Å2 in the x-y plane and a minimum water distance of 18 Å above and below the membrane. The systems were solvated with OPC water molecules69 and neutralized with 0.15 M NaCl to match physiological conditions.

The AMBER ff19SB70 and the SLipids forcefields71 were used to treat the protein and the lipid components, respectively.

Initial binding poses of 1 and 2 were generated using Glide SP docking72 with constraints based on the 5-HT2A-LSD X-ray crystal structure (PDB ID: 6WGT45). Key constraints included: (i) maintenance of the Asp1553.32 salt bridge, essential for tryptamine binding; (ii) indole ring overlay with LSD (RMSD tolerance 2.0 Å) to ensure reasonable starting orientations; (iii) exclusion of poses with severe steric clashes. The top-scoring pose for each ligand-receptor combination was selected based on GlideScore and visual inspection. 1 and 2 were described with a modified version of the GAFF force field73 validated against quantum mechanical calculations at the B3LYP-D3/6-31 G(d) level of theory (see Supplementary Note 3). Partial charges were derived with the multiconformational restrained electrostatic potential (RESP) protocol74 as implemented in PyResp based on B3LYP/6-311 G(d,p) electrostatic potential calculations performed with Gaussian 09.

Molecular dynamics simulation

A comprehensive set of eight different simulation systems was constructed to explore the conformational landscape of 5-HT2A under various conditions.

Each system underwent energy minimization using a combination of steepest descent and conjugate gradient algorithms. This was followed by a gradual heating phase from 0 to 310 K. The systems were then equilibrated in multiple stages with progressively decreasing harmonic restraints on protein backbone atoms. Production MD simulations were conducted in the NPT ensemble at 310 K and 1 atm without restraints. For each of the eight systems, three independent replicas were simulated in the microsecond timescale, yielding a cumulative simulation time of approximately 80 µs. The length of each MD is detailed in Table S1. Temperature was maintained using the Nosé-Hoover thermostat75 with a coupling constant of 1.0 ps, and pressure was controlled using the Parrinello-Rahman barostat76 with a coupling constant of 2.0 ps and a compressibility of 4.5 × 10−5 bar−1. The LINCS algorithm77 was used to constrain bonds involving hydrogen atoms, allowing for a 2 fs integration time step. Non-bonded interactions were calculated using a cutoff of 12 Å, and long-range electrostatic interactions were treated with the Particle Mesh Ewald (PME) method.

All MD simulations were performed using the GROMACS software package (version 2024.3)78.