Environmental Modulation and Ligand-Induced Stabilization of β2-Adrenergic Receptor

Aditi Sinha · 2026-06-01

The human β2-adrenergic receptor (β2AR) is a cornerstone of modern molecular pharmacology and the prototypical member of the Class A G protein-coupled receptor (GPCR) superfamily. Historically, structural understanding of this receptor was limited to the static snapshots provided by X-ray crystallography most notably the landmark 2.4 Å structure of the receptor bound to the inverse agonist carazolol (PDB: 2RH1). While these structures established the blueprint of the seven-transmembrane (7TM) helical architecture, they cannot capture the inherent conformational flexibility and metastable transitions that underlie GPCR function. Proteins are dynamic ensembles, and for β2AR, its role as a regulator of respiratory and cardiovascular smooth muscle relaxation is intrinsically linked to this plasticity.

In respiratory medicine, β2AR is the primary target for bronchodilators used to treat asthma and chronic obstructive pulmonary disease (COPD). Activation by endogenous catecholamines such as epinephrine triggers a well-characterised signalling cascade: the receptor couples to the heterotrimeric Gs protein, activating adenylyl cyclase, raising intracellular cyclic AMP (cAMP), and ultimately stimulating protein kinase A (PKA) leading to smooth muscle relaxation. Conversely, in cardiovascular medicine, β2AR’s inactive state is the therapeutic target of beta-blockers, which reduce heart rate and blood pressure by preventing receptor activation.

The transition from static structure to dynamic insight is enabled by molecular dynamics (MD) simulations, which track the motion of every atom in a biological system over time. Traditional MD workflows have, however, been hampered by significant barriers: complex hardware configurations, fragmented software dependencies, and a steep learning curve for non-specialists. LiteFold has emerged as an AI-native platform for drug discovery that addresses these challenges by integrating structure prediction with automated MD pipelines. This allows the seamless evaluation of receptors under varied physiological conditions pure aqueous environments, ligand-bound states, and fully solvated lipid bilayers within a unified interface.

This report presents a comprehensive evaluation of the LiteFold MD pipeline using human β2AR as a case study. Three parallel 30-ns simulations were conducted the apo receptor in water, the carazolol-bound receptor, and the receptor embedded in a POPC lipid bilayer to isolate the effects of environment and ligand occupancy on structural stability, local flexibility, and mass distribution.

Simulation Setup

The force field was selected for its established accuracy with membrane proteins and lipids, and TIP3P water was used as the solvent model. Key parameters are summarised in Table 1.

Theoretical Foundations of Structural Stability Metrics

Three complementary metrics were used to characterise the structural behaviour of β2AR across the three simulations.

Root Mean Square Deviation (RMSD)

RMSD quantifies the average displacement of a structure’s atoms relative to a reference state typically the initial crystallographic coordinates. It provides a single value per time step and is the standard measure of global structural drift over a trajectory.

Radius of Gyration (Rg)

While RMSD tracks deviation from a fixed reference point, Rg measures the physical spread of the protein’s mass around its center of mass.

It is a sensitive indicator of the compactness of the fold and the integrity of the hydrophobic core: a decreasing or stable Rg suggests that the overall architecture is being preserved, whereas an increasing Rg may signal partial unfolding or domain separation.

Root Mean Square Fluctuation (RMSF)

RMSF is calculated on a per-residue basis, averaged over the entire trajectory. Unlike RMSD which is a global, time-resolved measure RMSF provides a residue-level map of local flexibility. High RMSF values identify mobile regions such as loops and termini; low values identify rigid structural elements such as transmembrane ™ helices.

The LiteFold Architecture: Streamlining the MD Workflow

The simulations described here illustrate the efficiency gains provided by LiteFold’s integrated architecture. Traditionally, a 30-ns simulation of a membrane-embedded GPCR (approximately 100,000 atoms) would require extensive manual system preparation, membrane assembly, parameter selection, and computational resource management before production simulations can begin.

LiteFold’s “DYNAMO” engine automates this entire process through a browser-based interface, allowing researchers to prepare, launch, monitor, and analyze simulations from a unified environment. Complex setup procedures are abstracted into an intuitive workflow, reducing operational overhead and accelerating experimental iteration.

The platform leverages a modern, high-performance simulation infrastructure with GPU acceleration, enabling efficient and scalable molecular dynamics studies without the burden of software installation, dependency management, or low-level configuration. This streamlined approach allows researchers to focus on biological questions and mechanistic insights rather than computational infrastructure.

By integrating system preparation, simulation execution, trajectory analysis, and AI-assisted interpretation within a single platform, LiteFold significantly reduces the time required to move from structure to actionable insight.

Table 2 summarises the key architectural advantages relevant to this study.

Comparative Analysis of Case Studies

The three 30-ns simulations were designed to isolate distinct physical variables: the absence of any binding partner (Case A), the presence of a high-affinity inverse agonist (Case B), and the constraints imposed by a native-like lipid bilayer (Case C).

Case A: The Apo Receptor in Aqueous Solvent

In Case A, the receptor was simulated as a standalone protein in a water box, representing the unliganded baseline state. The apo β2AR exhibited the greatest structural drift of the three systems, with RMSD stabilising in the 0.4–0.7 nm range. This is consistent with the conformational selection model, which holds that unliganded GPCRs continuously sample multiple states—including active-like and inactive conformations in the absence of a stabilising chemical entity.

The Radius of Gyration remained broadly stable around 2.1 nm, but with more pronounced oscillations than in the other two cases.

These fluctuations are mirrored in the elevated RMSF peaks observed at the intracellular loops ICL2 and ICL3, regions central to G-protein and β-arrestin engagement. In the absence of a stabilising ligand, transmembrane helices TM5 and TM6 display greater lateral mobility, as they are not constrained by the inter-helical hydrogen bond network that inverse agonists such as carazolol impose.

Case B: The Holo Receptor and Ligand-Induced Stabilisation

In Case B, the inverse agonist carazolol was retained in the orthosteric binding pocket. Carazolol is a high-affinity ligand (Kₓ ∼0.1 nM) that is known experimentally to rigidify the receptor.

The LiteFold simulation captured this stabilisation clearly: global RMSD plateaued at 0.3–0.5 nm, and Rg decreased slightly to approximately 2.0 nm, reflecting a more compact and restrained structure compared to the apo state.

Case C: The Membrane-Embedded Receptor

In Case C, the receptor was embedded in a POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) lipid bilayer. The bilayer introduces a significant physical constraint on the lateral movement of TM helices. RMSD values for Case C (0.3–0.6 nm) were intermediate: greater than the ligand-bound state, but lower than the apo-in-water system.

This intermediate behaviour reflects the dual role of the membrane. On one hand, the hydrophobic lipid tails dampen vibrations within the helical bundle: RMSF values for the TM regions in Case C were the lowest of all three systems (mean < 0.05 nm for TM helices), a result that should not be conflated with the whole-protein average reported in Table 3 (0.07 nm), which includes the more mobile loop regions. On the other hand, the extracellular loops, which protrude into the aqueous phase, remain free to fluctuate and interact with solvent molecules and ions. The longer equilibration time required for Case C (~12 ns vs ~5 ns for the apo system) reflects the higher viscosity and greater system complexity of the membrane environment, rather than any instability in the simulation.

Case C also demonstrated LiteFold’s ability to handle the complex physics of protein-lipid interactions, including partial loop burial at the bilayer interface and the formation of transient hydrogen bonds between basic receptor residues and phospholipid headgroups (average 1.8 such contacts).

Quantitative Summary of MD Metrics

Table 3 summarises the key structural metrics extracted from the 30-ns production trajectories across all three environments.Table 3. Quantitative comparison of structural metrics across the three simulation conditions.

Environmental Modulation of the Conformational Ensemble

Beyond simple metrics of stability, the comparison of these three cases yields deeper insights into GPCR pharmacology. The data suggest that the physicochemical environment is not a passive medium but an active co-regulator of receptor conformation.

The Membrane as a Thermodynamic Buffer

The contrast in RMSD trajectories and equilibration times between Cases A and C illustrates that the lipid bilayer functions as a thermodynamic buffer. In water, the receptor samples conformational space rapidly but with high-amplitude fluctuations. The membrane, by contrast, imposes a higher effective viscosity that slows conformational sampling while constraining the overall fold. For computational drug discovery, this has a practical implication: docking calculations performed against a solvated protein with no membrane will not accurately reproduce the constrained binding pocket geometry that exists in a physiological context.

Ligand-Induced Rigidity and Allosteric Communication

The reduction in whole-protein RMSF from 0.09 nm (apo) to 0.06 nm (carazolol-bound) is a quantitative signature of induced-fit stabilisation. By locking β2AR into a single low-energy inactive state, the inverse agonist suppresses the conformational heterogeneity of the ensemble. Crucially, the LiteFold simulations also reveal allosteric communication across the receptor: in Case B, the binding of carazolol at the extracellular orthosteric pocket produced a measurable decrease in RMSF at the cytoplasmic ends of TM3 and TM6. This “distal dampening” is the structural signature of inverse agonism the propagation of conformational restraint from the binding site to the intracellular signalling interface, even in the absence of a G-protein.

The Role of Ordered Water in the Binding Pocket

The hydrogen bond analysis for Case B identifies an average of 5.2 persistent ligand-receptor interactions. A subset of these involve “ordered” or “intrinsic” water molecules that bridge the ligand and receptor residues at the base of the binding cavity. These water-mediated contacts are frequently missed by static docking algorithms that treat the binding site as a rigid, desolvated cavity. LiteFold’s explicit TIP3P solvation model captures these bridging interactions, providing a more physically realistic picture of ligand binding. Identifying which ordered water molecules should be displaced by a drug candidate and which should be retained is a key challenge in structure-based drug design (SBDD).

Computational Microscopy: RMSF and Domain Flexibility

A residue-level analysis of the RMSF profiles across the three systems reveals a consistent pattern of flexibility dictated by receptor architecture.

Transmembrane Helical Stability

In all three cases, the central portions of the 7TM helical bundle (roughly residues 30–340, excluding loop regions) showed RMSF values below 0.05 nm. This core rigidity is essential for maintaining the structural integrity of the 7TM fold under thermal stress. The most rigid element was TM3, which serves as the structural hub of the receptor, mediating the “ionic lock” with TM6 via the Asp130–Glu268 salt bridge a key interaction that stabilises the inactive state and must be broken upon activation.

Loop Mobilities and Functional Implications

The highest RMSF values were consistently located in the loop regions:• ICL3 (Intracellular Loop 3): Connecting TM5 and TM6, ICL3 is a major docking site for both G-proteins and β-arrestins. In the apo-in-water system (Case A), this region reached RMSF values of ~0.15 nm—reflecting the large-scale conformational sampling required for the outward movement of TM6 (~14 Å) that accompanies receptor activation.• ECL2 (Extracellular Loop 2): ECL2 forms a “lid” over the orthosteric binding pocket and contains a short α-helix. In Case B, the RMSF of ECL2 was substantially reduced relative to Case A, as carazolol forms steric and polar contacts that effectively tether the loop to the helical bundle.

The Convergence of AI and Physics-Based Simulation

The results of this case study illustrate the value of AI-native research platforms that combine machine learning with rigorous classical mechanics. By integrating neural network-based pocket detection and molecule generation with physics-based MD for stability validation, LiteFold addresses the accuracy gap that has historically limited purely computational drug discovery pipelines.

Speed and Scalability

Running three distinct 30-ns trajectories on cloud GPUs in parallel completing each in 2–4 hours represents a substantial throughput improvement over traditional workflows, where similar simulations on shared academic clusters might require days of queue time or significant capital expenditure on local infrastructure. This scalability enables research teams to move from a single reference structure to an ensemble of ligand-receptor complex simulations without hardware constraints.

Reproducibility and Standardisation

A persistent challenge in MD research is the lack of inter-laboratory reproducibility, arising from variation in force field versions, equilibration protocols, and post-processing scripts. LiteFold’s standardised pipeline enforces a consistent protocol across all simulations regardless of whether the system is in water or embedded in a membrane ensuring that differences in output metrics reflect genuine physical differences between systems rather than methodological artefacts. This reproducibility is also a prerequisite for generating the large-scale, physics-based datasets needed to train the next generation of machine learning models for drug discovery.

Conclusion

The human β2-adrenergic receptor is not a static structure but a dynamic molecular machine whose conformational landscape is shaped by its physicochemical environment and the ligands it binds. The three-way comparison presented here apo in water, carazolol-bound, and membrane-embedded demonstrates that each condition produces a quantitatively and qualitatively distinct structural ensemble, faithfully captured by the RMSD, Rg, and RMSF metrics.

The ligand-induced stabilisation by carazolol, the thermodynamic buffering effect of the POPC bilayer, and the intrinsic conformational sampling of the apo receptor are all reflected in the simulation output in ways that are physically interpretable and pharmacologically meaningful. These results validate the technical capabilities of the LiteFold DYNAMO engine and demonstrate its utility as a high-throughput tool for rational drug design.

As drug discovery shifts increasingly toward in-silico-first workflows, the ability to rapidly and reproducibly assess receptor behaviour across diverse environments will become a core competency. Platforms that bridge the gap between AI-based prediction and classical physics-based validation are well positioned to accelerate the development of more selective and efficacious therapies for respiratory and cardiovascular diseases.