From Sea to Synapse: Molecular Docking and Dynamics Study of Aplysinopsin Analogues as Selective 5-HT₂C Ligands

Abstract

Background:

Selective modulation of the serotonin 5-HT₂C receptor is a promising strategy for treating conditions, such as obesity and neuropsychiatric disorders. Aplysinopsins, a class of marine indole alkaloids, have emerged as potential 5-HT₂C-selective scaffolds.

Objective:

This work aims to assess the potential of aplysinopsin-based analogues as selective 5-HT₂C ligands.

Methods:

Here, we present conducted an in silico study of a library of aplysinopsin analogues and related indole alkaloids using induced-fit molecular docking, molecular dynamics (MD) simulations, and Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) profiling.

Results:

Dozens of analogues were docked into a serotonin receptor model, revealing sub-micromolar predicted binding affinities for several compounds. Top-ranked ligands, such as tubastrindole C, achieved docking scores around –12.7 kcal/mol and maintained stable binding poses in 100-ns MD simulations. Key ligand–receptor interactions included hydrogen bonding to the conserved Asp residue in the binding pocket and extensive π–π contacts with aromatic side chains. Notably, certain analogues with modest docking scores (e.g., N-propionyl-aplysinopsin) exhibited very favorable molecular mechanics generalized born surface area (MMGBSA) binding free energies, suggesting significant induced-fit effects upon binding. Predicted pharmacokinetic properties of the lead compounds were encouraging: all hits obeyed drug-likeness rules (0–1 Lipinski’s rule violations) and showed high oral absorption prospects. While polar analogues had limited blood–brain barrier permeability, several top candidates displayed moderate central nervous system (CNS) penetration scores.

Conclusion:

These results highlight aplysinopsin-based analogues as attractive selective 5-HT₂C ligand candidates and provide molecular insights to guide future optimization and experimental validation.

INTRODUCTION

The serotonin 5-HT₂C receptor is a G protein-coupled receptor (GPCR) expressed in the central nervous system and implicated in appetite regulation, mood, and addiction.[1,2] Selective 5-HT₂C agonists have demonstrated clinical potential—for example, lorcaserin, a 5-HT₂C-selective agonist, showed high affinity (K_i ≈ 15 nM) and preferential activation of 5-HT₂C over 5-HT₂A and 5-HT₂B receptors (≈18-fold and 104-fold selectivity, respectively),[3] leading to its use (now withdrawn) as an anti-obesity medication.[3,4] However, designing ligands that discriminate between the 5-HT₂C and the closely related 5-HT₂A subtype has proven challenging due to their high sequence homology and similar orthosteric binding sites.[5] Even subtle structural differences in ligands can drastically shift subtype selectivity.[5]

Marine natural products have provided unconventional scaffolds to probe 5-HT receptors. Aplysinopsins—indole-imidazolone alkaloids originally isolated from the sponge Thorecta aplysinopsis and later from various corals and anemones—are one such example.[6] Early pharmacological studies revealed that certain aplysinopsin derivatives bind to human 5-HT receptors with moderate affinity.[5] Remarkably, a naturally occurring aplysinopsin (6-bromo-2′-de-N-methylaplysinopsin) was >40-fold selective for 5-HT₂C over 5-HT₂A, although its 5-HT₂C affinity was in the low micromolar range (~2.3 μM).[5] This indicated that the aplysinopsin pharmacophore could be tuned for subtype selectivity.[5] Subsequent structure–activity relationship works on synthetic analogues achieved dramatic improvements: For instance, incorporating dual halogens on the indole ring and N-alkylation of the imidazolone yielded compounds with sub-50 nM affinity and ~2100-fold selectivity for 5-HT₂C.[5] Notably, 5,6-dichloro substitution and concurrent methylation at N¹/N³ of the imidazolidinone were identified as key features conferring 5-HT₂C selectivity.[5] These findings underscore the promise of aplysinopsin-based analogues as selective 5-HT₂C ligands. These findings underscore the promise of aplysinopsin-based analogues as selective 5-HT₂C ligands. Despite this progress, the binding modes and molecular interactions underpinning the selectivity of aplysinopsin analogues remain incompletely understood. At the time many of these analogues were discovered, high-resolution structures of 5-HT₂C were not yet available, complicating receptor-based design. In recent years, several serotonin receptor structures have been solved, enabling structure-guided investigations.[6,7] Here, we employed computational approaches to elucidate how aplysinopsin analogues interact with the 5-HT₂C binding site and to identify new promising analogues. We performed induced-fit docking of a diverse set of aplysinopsin derivatives, including halogenated and N-alkylated analogues, as well as related bis-indole alkaloids (e.g. tubastrindoles) into a 5-HT receptor model, followed by molecular dynamics (MD) simulations to refine binding poses. In parallel, we predicted ADMET properties to evaluate drug-likeness and brain penetration potential of the top hits. Our integrated in silico strategy provides insight into the ligand–receptor interactions that drive 5-HT₂C selectivity and highlights lead candidates for future synthesis and experimental testing.

METHODS

Computational receptor model

In the absence of a crystal structure of the agonist-bound 5-HT₂C receptor at the initiation of this work, a closely related serotonin receptor structure was used as the docking model. Specifically, the cryo-EM structure of human 5-HT₁B bound to the agonist donitriptan (PDB ID: 6G79) was chosen as a template, given the high homology among 5-HT receptor subtypes and the conserved architecture of their orthosteric binding sites.[8] The protein structure was prepared using Schrödinger’s Protein Preparation Wizard[9] (Schrödinger Release 2025-1), assigning protonation states appropriate for pH 7.0 and optimizing hydrogen-bonding networks.[10] All water molecules and ions were removed except those mediating key crystal contacts. To approximate the 5-HT₂C binding pocket, we retained the side-chain conformations from this active-state 5-HT₁B structure, reasoning that the agonist-bound conformation would accommodate ligands in a similar manner to 5-HT₂C. The co-crystallized ligand (donitriptan) was extracted to define the binding site.[11] A grid encompassing the orthosteric site was centered on the ligand and defined with a 10 Å radius.

Ligand preparation

A library of 55 aplysinopsin-derived compounds was compiled from literature and natural product databases. This set included aplysinopsin itself and various natural analogues (e.g., 6-bromoaplysinopsin, 2′-demethylaplysinopsin, oxoaplysinopsins) as well as dimeric or cyclized derivatives like isoplysins and tubastrindoles. Two-dimensional structures were converted to 3D using LigPrep (Schrödinger 2025-1). Protonation and tautomeric states were generated for pH 7 ± 0.5; notably, aplysinopsin analogues can exist in an equilibrium between 2-amino- and 2-imino-imidazolone forms, so both were considered during preparation. Low-energy conformers were generated for each ligand.

Induced-fit docking

Docking simulations were performed with the Schrödinger induced-fit docking (IFD) protocol, which accounts for receptor flexibility upon ligand binding.[12] In the first stage, ligands were docked into the rigid receptor using Glide[13] (Schrödinger 2025) with softened van der Waals potentials (0.5 scaling for nonpolar atoms) to allow initial close contacts. Poses within 5 kcal/mol of the best GlideScore for each ligand were subjected to receptor refinement: residues within 5 Å of the ligand were flexibly optimized using Prime (Schrödinger) to accommodate the ligand (including side-chain rotamer adjustments). In the final stage, each ligand was re-docked into its induced-fit receptor structure and scored with the Glide XP scoring function. The IFDScore—a composite scoring metric combining ligand pose energy, receptor strain energy, and docking score—was recorded for each complex. For comparative purposes, the co-crystal ligand donitriptan was also docked into the binding site (after re-extraction) to verify that the protocol could reproduce its native pose.

Binding free energy estimation

For each top-ranking induced-fit pose, the binding free energy was estimated using the molecular mechanics generalized Born surface area (MM-GBSA) method. The Prime MM-GBSA module was applied to compute the ligand binding ΔG (ΔG_bind) in kcal/mol based on the docked complex (implicit solvent, VSGB 2.1 model).[14] This provides an end-point free energy estimation that complements the docking score.

Molecular dynamics simulations setup

MD simulations were carried out for selected ligand–receptor complexes. Two representative complexes were chosen: one with a high GlideScore (tubastrindole C, top docked hit) and one with a high MM-GBSA binding affinity (N-propionyl-aplysinopsin, top-ranked by ΔG_bind). Each complex was embedded in an explicit membrane and solvent environment using Desmond (D. E. Shaw Research) with the OPLS4 force field.[15] The receptor was oriented in a 1-Palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) lipid bilayer using an alignment from the OPM database for 5-HT receptors. The system was solvated with TIP3P water and 0.15 M NaCl, yielding simulation cells of ~100×100×120 Å.[3] After an initial minimization and slow heating (0–300 K over 50 ps) with positional restraints on the protein and ligand, 100 ns production simulations were run for each complex in the NPT ensemble (310 K, 1 bar) with a Nosé–Hoover thermostat and Martyna–Tuckerman–Klein barostat. Trajectories were saved every 100 ps. Root-mean-square deviation (RMSD) of protein Cα atoms and ligand heavy atoms was monitored to assess equilibration. Key intermolecular interactions (hydrogen bonds, salt bridges, π–π stacking) were tracked over time using the simulation event analysis module in Maestro.

ADMET predictions

Drug-likeness and ADMET properties of the compounds were predicted with QikProp (Schrödinger 2025-1). Key descriptors included molecular weight (MW), lipophilicity (QPlog P_o/w), hydrogen bond donors/acceptors, polar surface area (PSA), permeability (Caco-2 cell permeability, QPPCaco), blood–brain barrier penetration (QPlogBB), predicted CNS activity category (CNS, scale –2 to + 2), and overall rule-of-five compliance (#stars, where 0 indicates full compliance). These in silico ADMET profiles were used to flag any obvious developability issues among the top candidates.

RESULTS

Docking outcomes

IFD identified several aplysinopsin analogues with excellent predicted binding to the 5-HT₂C receptor model. The docking scores (Glide XP) ranged from approximately –12.7 to –4.4 kcal/mol across the library, indicating a wide span of binding affinities. As a validation, the native agonist donitriptan (co-crystallized in the 5-HT₁B template) was re-docked and scored –12.35 kcal/mol, with an RMSD < 1 Å from the crystal pose, confirming the reliability of the docking protocol.[16]

Notably, the top-ranked compound by docking score was tubastrindole C, a dimeric indole alkaloid, with a GlideScore of –12.68 kcal/mol. This suggests tubastrindole C can be accommodated snugly in the binding site [Figure 1]. Other tubastrindole analogues also scored highly (e.g., tubastrindole A and B around –11.9 kcal/mol), likely due to extensive π–π and hydrophobic contacts afforded by their larger polyaromatic framework. Among monomeric aplysinopsins, the best docking scores were obtained for 6-methoxycycloaplysinopsin B (–12.03 kcal/mol) and cycloaplysinopsin C (–11.51 kcal/mol), which are constrained analogues where the indole and imidazolone rings form a bicyclic system.[17] This rigidity may favor a pre-organized binding conformation. Standard aplysinopsin derivatives with halogen substitution also performed well; for instance, 6-bromo-1′-ethoxy-1′,8-dihydroaplysinopsin scored –10.34 kcal/mol. In general, compounds with halogens at the indole 5- or 6-position and those with alkylation at the imidazolone N¹ or N³ tended to dock with better scores than unsubstituted analogues, consistent with the notion that these modifications enhance fit and interactions in the largely hydrophobic pocket.

Figure 1.

Docking complex of PDB ID 6G79 with tubastrindole C (Compound 51). (a) 2D ligand–protein interaction map for the best-ranked pose. The ligand is anchored by hydrogen bonds to Met337 (indole N–H), Ser212 (terminal amine), Thr213 (phenolic O), and Asp^3.32 (which interacts with the basic amine/indolic NH). In addition, a cation–π contact with Phe330 is observed. The polycyclic scaffold is embraced by a largely hydrophobic pocket lined by Leu126, Ile130, Val201, Pro184, Pro338, Phe330, Phe331, Phe351, Trp327, and Cys133. (b) 3D view of the same pose (protein in grey cartoon/sticks; ligand in green sticks) showing the hydrogen-bond network (notably to Leu126, Ser212, Thr213, and Asp^3.32). Aromatic contacts are also observed, including π–π stacking with Phe330 and proximity to Phe351 (dashed lines)

The induced-fit protocol also provided refined poses for each ligand and corresponding Prime energies and MMGBSA ΔG_bind values. Interestingly, the docking rank did not always correlate with binding free energy predictions. For example, N-propionyl-aplysinopsin (compound 13) achieved only a moderate docking score (–8.06 kcal/mol, outside the top 15 by GlideScore), yet it had one of the most favorable MMGBSA binding energies (ΔG_bind ≈ –49.1 kcal/mol, second only to donitriptan). This large discrepancy suggests that compound 13 induces substantial receptor conformational adjustments that benefit binding – effects captured in the induced-fit and MMGBSA calculations but not fully reflected in the initial GlideScore. In its induced-fit pose, N-propionyl-aplysinopsin formed a bidentate hydrogen-bond network with the conserved Asp in TM3 and a neighboring Thr/Ser, facilitated by its polar propionyl group, which likely contributed to the strong MMGBSA score. Similarly, 1-N-methylaplysinopsin (Compound 12) showed a moderately good docking score (–6.67 kcal/mol) but a very favorable binding energy (–47.7 kcal/mol). In contrast, tubastrindole C, despite its excellent docking score and sizable contact surface, had a less negative ΔG_bind (–29.9 kcal/mol). This could reflect desolvation penalties or partial exposure of tubastrindole C to solvent in the binding site due to its bulk. Overall, a subset of aplysinopsin analogues (notably compounds 12 and 13, and several oxoaplysinopsins) achieved ΔG_bind values in the –45 to –50 kcal/mol range, comparable to or exceeding those of the top-scoring tubastrindoles, highlighting them as strong candidates.

In addition to GlideScore and MMGBSA ΔG_bind, the IFDScore and Glide eModel terms were considered in pose selection. In most cases, the top GlideScore pose for each ligand was retained as it also yielded a low IFDScore. All top hits formed at least one hydrogen bond to the receptor (commonly to Asp^3.32, the conserved Asp in helix 3). Hydrophobic interactions with aromatic residues in helices V, VI, and VII were also prominent—for example, indole ring stacking against Phe and Trp in the pocket. Halogenated ligands like 6-bromoaplysinopsin (compound 15) likely engage in favorable halogen–π or halogen bond interactions with residues such as Phe³·²⁹ or Thr⁶·⁵⁸ (5-HT₂C numbering), consistent with experimental observations that 6-bromo analogues prefer the 5-HT₂C subtype. The docking results thus suggest that multiple aplysinopsin-based compounds can achieve high-affinity binding to 5-HT₂C, either through a snug fit (as in the case of rigid polycyclic analogues) or through induced-fit optimization of binding interactions (as seen with more flexible, polar analogues).

Molecular dynamics simulation

To verify the stability of the docked complexes, MD simulations were carried out for tubastrindole C (Compound 51) and N-propionyl-aplysinopsin (Compound 13) in complex with the receptor. Both systems equilibrated rapidly: the protein Cα RMSD stabilized within 10–20 ns to fluctuations of ~ 2 Å from the starting (docked) structure, indicating that the binding pocket did not undergo large conformational changes beyond the induced-fit adjustments. The ligand heavy-atom RMSD also plateaued (oscillating within 1–2 Å of the docked pose for the majority of the trajectory), suggesting that the ligands remained bound in a stable conformation.

For tubastrindole C, the MD trajectory showed the ligand nestled in a primarily hydrophobic cavity formed by residues on transmembranes 3, 5, 6, and 7. Its two indole moieties engaged in stacking interactions with aromatic residues. Notably, one indole ring of tubastrindole C π-stacked against a conserved phenylalanine in TM6, while the other ring was oriented toward TM5, occupying a sub-pocket akin to the extended binding pocket seen in some GPCR–ergoline complexes. The linker of tubastrindole C (connecting the two indoles) maintained a hydrogen bond to Asp^3.32 via an NH group, which persisted ~ 70% of the simulation time. This interaction anchors the ligand, albeit not as strongly as a formal salt bridge (tubastrindole C is uncharged). The bulky ligand induced slight side-chain rearrangements (for instance, a rotamer shift in a glutamine on TM7 to accommodate the second indole), but overall the binding mode remained close to the docked pose. By the end of 100 ns, tubastrindole C remained securely bound, with an average ligand RMSD ~ 1.5 Å and no signs of dissociation.

For N-propionyl-aplysinopsin (13), the MD simulation supported an even more robust binding mode. Compound 13 is smaller and more polar, and it formed a stable bifurcated hydrogen-bond network: the imidazolone carbonyl of 13 hydrogen-bonded with Asp^3.32 (persisting > 90% of the trajectory), while its N-propionyl amide NH formed a hydrogen bond with a serine on TM5. These polar contacts compensated for the ligand’s lack of a permanent positive charge, essentially anchoring it in place of the prototypical protonated amine of serotonergic ligands. Additionally, the indole ring of 13 remained buried in an aromatic pocket, π-interacting with a tryptophan in TM6. Throughout the 100 ns run, compound 13 showed minimal movement; its average position shifted by <1 Å from the starting induced-fit pose, and crucial interactions were maintained. No water molecules displaced the ligand, indicating a well-packed complex. The stable trajectory of 13 correlates with its very favorable MMGBSA energy, suggesting that once receptor adjustments occur to accommodate this ligand, the complex is energetically highly favorable.

Comparing the two simulations, compound 13 demonstrated a somewhat tighter binding profile than tubastrindole C, evidenced by lower RMSD fluctuations and higher interaction occupancy. This may reflect the better complementarity (shape and electrostatics) of the smaller ligand to the binding site after induced-fit, whereas tubastrindole C, due to its size, leaves parts of the pocket more solvent-exposed and is limited by steric clashes unless the protein undergoes larger conformational changes. Nevertheless, both ligands remained stably bound, lending confidence that the docking poses represent realistic binding modes. Importantly, neither simulation showed any tendency for the ligands to prefer an alternate binding orientation or to egress from the pocket, reinforcing the validity of the predicted poses.

ADMET and drug-likeness

The pharmacokinetic profiles of the candidate ligands were generally favorable. All top compounds adhered to Lipinski’s rule-of-five, with zero or one parameter outside the recommended range (#stars ≤ 1). For instance, tubastrindole C (the largest analogue, MW ~510) incurred one rule violation (molecular weight slightly above the preferred 500 threshold), whereas simpler analogues, like aplysinopsin and its N-alkyl derivatives, had MW ~250–310 and zero violations. Predicted aqueous solubility (QPlogS) values were in an acceptable range (not shown), and none of the top hits bore reactive or pan-assay interference (PAINS) substructures.

One important consideration for centrally acting drugs is CNS penetrance. The predicted blood–brain barrier (BBB) permeability varied notably across the analogues. Aplysinopsin itself and its N-methyl and N-propionyl analogues showed modest log BB values (e.g., QPlogBB ≈ –0.4 to –0.8, unitless), corresponding to partial brain availability. For example, N-propionyl-aplysinopsin (13) had QPlogBB –0.77, similar to that of tubastrindole C (–0.84). In contrast, a reference serotonergic drug, Lysergic acid diethylamide (LSD), had QPlogBB +0.31 (indicating efficient BBB crossing), reflected in its CNS category score of 1 (on a scale from –2 (minimal) to + 2 (high CNS likelihood)). Most aplysinopsin analogues received CNS scores of –1 (low-to-moderate), suggesting some limitation in brain penetration relative to highly CNS-active compounds like LSD. The outlier was the very polar (–)-oxoaplysinopsin C (40), which had a high polar surface area (133 Ų) and correspondingly poor BBB permeation (QPlogBB –1.39, CNS score –2). Such polar analogues may require prodrug approaches or chemical modification to improve CNS delivery if they are to be developed for central effects. By contrast, analogues with intermediate polarity (PSA~50–80 Ų), such as 1-N-methylaplysinopsin (12) and N-propionyl-aplysinopsin (13), achieved high predicted human oral absorption (>90% absorbed) and moderate BBB penetration. For instance, compound 12 was estimated to have ~100% oral absorption and a CNS score of 0, indicating a balance of permeability that could allow some CNS activity while maintaining good systemic bioavailability.

In terms of metabolic and safety profiles, the QikProp predictions did not flag any major issues for the lead compounds. All top candidates had low predicted hERG liability (QPlogHERG above –5), suggesting minimal risk of cardiotoxicity via hERG channel blockade. The compounds also had clean profiles for metabolic stability (the number of predicted metabolites was within normal ranges, 5–8). No structural alerts for toxicity were present. While these in silico ADMET assessments are preliminary, they support the drug-likeness of aplysinopsin analogues, especially the smaller, monomeric ones. In particular, the high oral absorption forecasts and lack of significant rule-of-five violations mean these molecules—unlike many complex natural products—could be orally bioavailable. The principal concern might be moderate brain penetration; nonetheless, given that 5-HT₂C is located in the CNS, optimizing CNS exposure will be a key consideration. Medicinal chemistry efforts may focus on increasing lipophilicity or reducing PSA slightly (e.g., via prodrug strategies or substituent changes) for those analogues that currently straddle the BBB permeability threshold.

Taken together, the ADMET analysis highlights that several of the computational hits have physicochemical properties within the range of known CNS drugs. For example, 1-N-methylaplysinopsin (12) has MW 268, cLogP ~3.0, PSA ~56 Ų, and zero rule violations—a profile reminiscent of drug-like 5-HT₂C ligands such as lorcaserin (MW 195, cLogP 2.7, PSA 12 Ų). Although the aplysinopsin analogues carry an indole motif (which can be susceptible to rapid metabolism), substitutions like halogens or N-methyl groups can enhance metabolic stability. The presence of a cyclic urea/imide in these structures offers multiple sites for potential hydrogen bonding; balancing this polarity has been key, as seen in our top analogues that incorporate small N-alkyl groups to moderate polarity. Overall, no insurmountable ADMET barriers were evident for the lead analogues, reinforcing their promise as candidates for further development as selective 5-HT₂C modulators.

DISCUSSION

In this study, we applied a comprehensive computational workflow to explore the 5-HT₂C binding potential of aplysinopsin analogues, yielding both expected trends and novel insights. The docking and MD results strongly support earlier experimental observations that halogenation of the indole ring and alkylation of the imidazolone ring substantially influence affinity and subtype selectivity. For instance, our highest-scoring analogues included several brominated derivatives and N-alkyl analogues, which echoes the findings of Cummings et al.[5] that 6-bromo substituents and N¹, N³-dimethylation synergistically improved 5-HT₂C selectivity and potency. The docking poses indicate that a 6-bromo group on the indole is oriented toward a hydrophobic sub-pocket where it can form van der Waals contacts and possibly a halogen bond with a polar residue (e.g., a conserved Thr in TM5). This specific interaction may be absent in the 5-HT₂A receptor if the analogous position is less accommodating (as seen with 6-fluoro analogues preferring 5-HT₂A), thus rationalizing the subtype discrimination. Our data thus reinforce the hypothesis that the 5-HT₂C binding pocket has “hotspots” for certain substituents, notably a halogen-binding pocket around the five- and six-positions of the indole ring, that can be exploited to enhance selectivity.

Another notable insight is the role of ligand protonation state and basicity in binding. Classic serotonergic agonists (e.g., tryptamines, ergolines) carry a protonated amine that forms a salt bridge with Asp^3.32, a critical anchor point in all 5-HT receptors. Aplysinopsin analogues lack a comparable basic amine; instead, they rely on polar amidine/urea functionalities to engage Asp^3.32 via hydrogen bonding. This difference may be translated to altered signaling profiles or subtype preferences. Interestingly, “atypical” 5-HT ligands with lower basicity have been noted to sometimes confer biased signaling or improved selectivity.[6] Our MD simulations confirmed that hydrogen bonds can indeed substitute for the canonical ionic lock in maintaining ligand binding (as seen with compound 13). The relatively modest penalty for losing the ionic interaction (reflected in mid-nanomolar predicted affinities for aplysinopsins vs. typically sub-nanomolar for the best protonated ligands) suggests that the aplysinopsin scaffold can achieve good potency while potentially avoiding off-target effects associated with highly basic compounds. Furthermore, subtype differences (e.g., an Ala vs. Ser at 5-HT₂A residue 5.46, or other local polarity differences) might make a neutral ligand more favorable at 5-HT₂C than at 5-HT₂A, whereas a protonated ligand might bind both similarly. This could be an underlying factor in the selective binding of aplysinopsins to 5-HT₂C observed experimentally. In essence, our findings highlight low-basicity hydrogen-bonding as an alternative binding mode that can impart selectivity—a principle that could be leveraged in designing new ligands for GPCRs where traditional amine binding leads to promiscuity.

Without experimental functional assays, we can only speculate, but the docking poses (all in the same active-state model) do not penalize potential antagonists. Future work could involve docking to an inactive 5-HT₂C model (such as a homology model based on the 5-HT₂B or an inverse agonist-bound structure) to see if certain analogues preferentially bind the inactive state. Such an approach could predict which analogues are agonists vs. antagonists. Nevertheless, given that our interest is largely in agonists for therapeutic reasons (e.g., appetite suppression via 5-HT₂C activation), the identified top hits, like N-propionyl-aplysinopsin, are encouraging, as their interaction pattern (anchoring Asp^3.32 engaging conserved aromatic residues) is consistent with agonism.

From a drug development perspective, the ADMET results were heartening. Natural product-derived leads often suffer from poor drug-likeness, but many aplysinopsin analogues are relatively small (MW 250–350) and of moderate lipophilicity (cLogP ~ 2–4). They also contain multiple heteroatoms, which kept PSA somewhat high; still, several analogues struck a good balance, achieving high predicted oral absorption. One potential issue is CNS penetration—some analogues may need structural modification to ensure sufficient brain levels for 5-HT₂C engagement in vivo. For instance, the oxoaplysinopsins (which had among the best docking energies) were quite polar; reducing one hydrogen bond donor or adding a more hydrophobic substituent might improve their log BB. Encouragingly, the more hydrophobic end of our series (e.g., tubastrindoles) had no trouble crossing the BBB in predictions, but those might have other liabilities (metabolic stability, synthetic accessibility). Meanwhile, analogues like 1-N-methyl-6-bromoaplysinopsin (Compound 19) could be a sweet spot: it has a single bromine to boost affinity, one N-methyl to reduce PSA, and it still preserves the core scaffold. Such compounds were predicted to have reasonable CNS penetration and may warrant prioritization. It is worth noting that lorcaserin, the exemplar 5-HT₂C agonist, has a very simple structure and minimal polar functionality. Aplysinopsin analogues are more complex and polar, but their extra functionality could confer selectivity and unique pharmacology, as discussed. The key will be ensuring enough of the compound reaches the brain—something medicinal chemistry optimization can likely achieve by tweaking substituents (for example, converting the urea to a carbamate or adding a lipophilic group to mask a polar moiety).

In comparing our computational findings with the known SAR from literature, it is satisfying to see strong concordance. Compounds identified here as top contenders (such as 6-bromo analogues, Compound 13, Compound 19, and N,N-dimethyl analogues) overlap with those potent and selective in in vitro binding assays by Schetz and colleagues.[18] Moreover, our work extends those insights by suggesting novel analogues like tubastrindoles as potential 5-HT₂C binders—these were not examined in prior SAR studies, as tubastrindoles were originally isolated for their antimicrobial or cytotoxic properties rather than CNS activity. If these larger bis-indole molecules can be shown to bind 5-HT₂C, they would represent a new chemotype for this receptor. Their sizable structure might also confer biased signaling or atypical pharmacokinetics (for instance, possibly slower receptor dissociation, akin to certain bitopic GPCR ligands). This invites experimental evaluation. Another novel finding is the significant receptor relaxation energy associated with certain ligands (e.g., compound 13). This suggests that induced-fit effects are important—something docking alone might miss. In practical terms, this implies that in vitro assays might find some compounds (like 13) more potent than expected from a static model, because the receptor can adapt. High flexibility at the binding site could also mean that covalent or bitopic ligand strategies might lock the receptor in beneficial conformations; aplysinopsins could be used as warheads in such hybrid designs.

It should be acknowledged that our study has limitations common to computational approaches. The homology receptor model (5-HT₁B-based) may not capture all the nuanced differences of 5-HT₂C, although it is a reasonable proxy. Any divergence in key residues (especially in the binding pocket’s extended regions) could affect actual binding affinities and selectivity—for example, 5-HT₂C has an Asn in ECL2 where 5-HT₁B has Asp, which could alter ligand recognition of extracellular loops. We partially mitigated this by focusing on conserved core interactions, but the absolute affinity predictions (docking scores, ΔG_bind) should be interpreted relatively, not as exact values. Another limitation is the single-state modeling; as mentioned, we did not explicitly model inactive-state binding, which is relevant for antagonist design. Additionally, the MD simulations, while of decent length (100 ns), cannot fully explore slow conformational transitions or rare binding/unbinding events. Experimental validation—receptor binding assays and functional tests—will ultimately be needed to confirm the computational hits. Encouragingly, many compounds highlighted here either have been synthesized or can be readily prepared following known routes (for instance, condensation of indole-3-carboxaldehydes with creatinine derivatives). We anticipate that testing these compounds will reveal how accurately the in silico predictions translate to reality.

The discovery of potent, selective 5-HT₂C agonists is of considerable interest for developing treatments for obesity, depression, and substance use disorders, given 5-HT₂C’s role in regulating appetite and mood. The aplysinopsin scaffold offers an alternative to classic phenethylamine-based 5-HT₂C agonists, potentially with reduced off-target activity at 5-HT₂A (and 5-HT₂B, whose activation is linked to valvular heart disease). Our findings lay the groundwork for rational optimization of this scaffold. Future work could involve structure-based design such as fragment growing or docking-guided substituent exploration. For example, our model suggests that extending the N-propionyl group of compound 13 into the extracellular vestibule could engage additional contacts, possibly enhancing both affinity and selectivity. Such modifications could be explored computationally and then synthesized. Furthermore, co-crystallization or cryo-EM of 5-HT₂C with a high-affinity aplysinopsin analogue (once available) would be invaluable to validate the predicted binding mode. On the computational front, free energy perturbation (FEP) calculations could be employed to more quantitatively rank the analogues, now that plausible binding poses are known. This may improve the accuracy of affinity predictions, guiding which analogues to prioritize for synthesis.

CONCLUSION

Aplysinopsin analogues from marine sources constitute a compelling chemotype for selective serotonin 5-HT₂C receptor ligands. Through IFD and MD, we showed that diverse analogues can stably engage the 5-HT₂C binding pocket, often mirroring key interactions of known agonists. Halogenation and ring N-alkylation were confirmed as beneficial modifications, enhancing binding affinity and subtype selectivity. Several top candidates (e.g., N-alkyl-6-bromoaplysinopsins) combine high predicted potency with drug-like properties, meriting further investigation. While computational models have limitations, our findings align well with experimental SAR, lending credence to the predicted ligand–receptor complexes. This work sets the stage for the experimental validation of aplysinopsin-inspired compounds and illustrates the value of in silico methods in natural product-based drug discovery for CNS targets. Ultimately, refining these analogues could yield novel therapeutics that harness the specificity of 5-HT₂C modulation for treating human disorders.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

The Deanship of Scientific Research (DSR) at King Abdulaziz University (KAU), Jeddah, Saudi Arabia, has funded this project, under grant no. (RG-12-166-43).