A cholesterol analog stabilizes the human β₂-adrenergic receptor nonlinearly with temperature

Abstract

In cell membranes, G protein–coupled receptors (GPCRs) interact with cholesterol, which modulates their assembly, stability, and conformation. Previous studies have shown how cholesterol modulates the structural properties of GPCRs at ambient temperature. Here, we characterized the mechanical, kinetic, and energetic properties of the human β₂-adrenergic receptor (β₂AR) in the presence and absence of the cholesterol analog cholesteryl hemisuccinate (CHS) at room temperature (25°C), at physiological temperature (37°C), and at high temperature (42°C). We found that CHS stabilized various structural regions of β₂AR differentially, which changed nonlinearly with temperature. Thereby, the strongest effects were observed for structural regions that are important for receptor signaling. Moreover, at 37°C, but not at 25° or 42°C, CHS caused β₂AR to increase and stabilize conformational substates to adopt to basal activity. These findings indicate that the nonlinear, temperature-dependent action of CHS in modulating the structural and functional properties of this GPCR is optimized for 37°C.

INTRODUCTION

G protein–coupled receptors (GPCRs) constitute one of the largest families of human transmembrane proteins (1, 2) that sense extracellular signals and transduce this information into the cell to initiate cellular responses (3, 4). Members of the GPCR family modulate responses to hormones and neurotransmitters, and play crucial roles in vision, smell, taste, the immune response, and inflammation. One of the structurally and functionally best studied GPCRs is the human β₂-adrenergic receptor (β₂AR), which serves as a model of the largest class, class A, of the GPCR family (5, 6). Because β₂AR binds to adrenaline in bronchial tissue, promoting smooth muscle relaxation, and to noradrenaline in cardiac tissues, playing a role in heart function and failure, it is a promising therapeutic target (7-9). β₂AR and other GPCRs coexist in multiple conformational states of which certain states become more prominent depending on the GPCR assembly, membrane lipid composition, and ligand binding (10-13). Detailed insight into how GPCRs interact with the membrane environment has been provided by structural models obtained by x-ray crystallography, nuclear magnetic resonance (NMR), cryo–electron microscopy (EM), native mass spectrometry, and molecular dynamics (MD) simulations (5, 6, 12-19). On the basis of this progress, we are now starting to understand how the lipid composition of the cellular membrane modulates the functional state of GPCRs (17, 20, 21). Among the lipids of the cell membrane, cholesterol has prominent roles in modulating the functional state and stability of GPCRs (5, 22). However, how and to which extent cholesterol affects the conformational stability of GPCRs remains a topic of investigation.

Cholesterol, a major component of mammalian plasma membranes, ranges in physiological concentration from 10 to 45 mole percent (mol %) depending on the cell type (23-25) and has crucial roles in modulating the structure-function relationship of transmembrane proteins (22, 26). For example, cholesterol and its more water-soluble analog cholesteryl hemisuccinate (CHS) modulate the ligand binding and functional states of GPCRs in an allosteric manner and increase the stability of GPCRs (27-29). On the other hand, the structure and function of GPCRs are often characterized at nonphysiological temperatures, although the temperature dependencies of membrane receptor assembly, stability, and function have been observed frequently (27, 30). Thus, which of the structural properties of GPCRs are modulated by cholesterol and CHS, and how this modulation depends on the physiologically relevant temperature remain to be understood.

High-resolution structures of several GPCRs have revealed cholesterol bound to the receptors (5, 27, 31, 32). X-ray structures of β₂AR show cholesterol-binding sites between transmembrane α helices TMH1, TMH2, TMH3, and TMH4 (5, 27). Structures of close family members of β₂AR show cholesterol- or CHS-binding sites between TMH2 and TMH4 and between TMH3 and TMH5 for the β₁AR, and between TMH2 and TMH3 and between TMH5 and TMH6 for the adenosine A2a receptor (A_2aAR) (33, 34). Complementary to such static structural models, MD simulations of β₂AR embedded in lipid membranes support the idea that cholesterol preferentially occupies specific interaction sites of the receptor (35-37). Full-atomistic MD simulations spanning time ranges of up to ~100 μs show several high-affinity cholesterol-binding sites of which one is located in a cleft formed at the intracellular surface of TMH1 to TMH4, another at a cleft between TMH5 and TMH6 at the intracellular side, and two closely spaced cholesterol hotspots located at the extracellular region of TMH5-TMH6-E3-TMH7 (36). Although structural models and MD simulations can unveil static and transient interactions of β₂AR and cholesterol, they cannot be used to quantify whether and to what extent cholesterol modulates the mechanical, kinetic, and energetic properties of the receptor in physiologically relevant time ranges.

Atomic force microscopy (AFM)–based single-molecule force spectroscopy (SMFS) has been introduced to characterize the structural folding and properties of transmembrane proteins, including transporters, ion channels, and GPCRs (38-42). When applied to the human β₂AR in the presence of CHS at room temperature (29), dynamic SMFS (DFS) has revealed the changes that occur to the mechanical, kinetic, and energetic properties of β₂AR upon ligand binding (38). However, the understanding of how sensitive such structural properties of GPCRs are to physiological temperature ranges is limited. Here, we applied DFS experiments and atomistic MD simulations to study the role of CHS in modulating the mechanical, kinetic, and energetic properties of β₂AR at room temperature (25°C), at physiological temperature (37°C), and at high temperature (42°C), which leads to cellular damage in humans (43, 44). The insights gained, which defined the structural regions of β₂AR that are affected by CHS and quantified how their properties depend on temperature, highlight the mechanisms by which sterols optimize the structure-function relationship of β₂AR to physiological temperature.

RESULTS

Mapping the mechanical stability of human β₂AR

To characterize how temperature variations within the physiological range influence the mechanical stability of β₂AR, we reconstituted human β₂AR in 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) liposomes without or with CHS (DOPC:CHS ratio 10:1 v/v) (fig. S1A). The proteoliposomes were adsorbed in buffer to carbon grids or mica for ~1 hour and imaged by transmission EM (TEM) or AFM, respectively, where they opened up as single-layered membrane patches (fig. S1, B to F) To acterize whether the appearance of the membranes changed with temperature, we imaged the samples at 25°, 37°, and 42°C by AFM (fig. S2). The membranes and β₂AR assemblies protruding from the membrane did not change appearance, thus indicating that the lipids did not undergo phase separations and that β₂ARs did not change their assembly. After imaging the opened-up proteoliposomes by AFM in buffer at room temperature (25°C), the stylus of the AFM cantilever was pushed onto β₂AR embedded in a DOPC membrane at ~700 pN for 0.5 s (Fig. 1A). In ~0.1% of all cases (n > 1,000,000), the mechanical contact promoted the nonspecific attachment of the N-terminal end of β₂AR to the AFM stylus (29, 38,45). Subsequently, the AFM cantilever was retracted from the membrane at constant speed. During retraction, the N-terminal end of the receptor was mechanically stretched, and the extraction and unfolding of a single β₂AR was recorded with a force-distance curve (Fig. 1B). As previously reported (29, 38), the force-distance curve showed force peaks extending to distances of ~80 nm, which corresponds to the contour length of the fully unfolded and stretched β₂AR polypeptide. The superimposition of 100 force-distance curves, each recorded upon mechanically unfolding a single β₂AR, revealed the common unfolding pattern of human β₂AR with eight dominant force peaks (Fig. 1C). Every force peak of the force-distance curve was fitted by the worm-like chain (WLC) model to estimate the contour length of the stepwise unfolded and stretched β₂AR polypeptide (45). The contour lengths of the eight force peaks were then used to localize the beginning and the end of the eight structural segments (S1 to S8) unfolded by β₂AR (Fig. 1D). The mechanical stability of each structural segment, which is typically composed of α helices and polypeptide loops (29, 38), is described by the magnitude of the respective force peak (45).

Fig. 1. SMFS analysis of the human β₂AR reveals a common unfolding peak pattern.

(A) Schematic representation of the SMFS experiment. The stylus of the AFM cantilever approaches to the surface of β₂AR-containing proteoliposomes until reaching a contact force of 700 pN to nonspecifically attach to the N terminus of individual receptors (38). After 0.5 s, the cantilever is retracted at constant speed and stretches the polypeptide linking the stylus and β₂AR. At sufficient mechanical force, β₂AR unfolds stepwise until the polypeptide is completely extracted from the membrane. During this process, a force-distance curve is recorded. (B) Representative force-distance curves recorded upon mechanical unfolding of individual β₂ARs from the N-terminal end at 25°C (38). (C) Superimposition of force-distance curves recorded upon unfolding of β₂AR from a DOPC membrane. The density plot reveals a highly reproducible unfolding fingerprint pattern of eight force peaks. Each force peak is fitted by the WLC model (cyan lines; see Materials Methods) to approximate the contour length of the unfolded and stretched polypeptide. The mean contour lengths given at the top of each WLC curve in amino acids indicate where a structural segment starts unfolding and the unfolding of the previous structural segment ends. The gray scale bar of the density plot indicates the number of data points superimposed per bin. n gives the number of superimposed force-distance curves, which were recorded from more than 10 independent experiments. (D) The eight structural segments S1 to S8 mapped to the secondary structure of β₂AR. Helices are numbered with Roman numerals. Extracellular (E1 to E3) and cytosolic (C1 to C3) loops are labeled. The linearized secondary structure was taken from the study of Zocher et al. (38) accordingly to tertiary structure model from Protein Data Bank (PDB) ID code 3D4S (57).

CHS protects β₂AR from destabilization at 37°C

The superimposition of force-distance curves recorded for β₂AR and other GPCRs is highly reproducible and sensitive to experimental conditions (29, 38, 42, 46). To characterize how temperature influences the stability of the human β₂AR, we used SMFS to record many force-distance curves at 25°, 37°, and 42°C (Fig. 2). For β₂AR embedded in DOPC membranes, the force-distance curves revealed considerable temperature dependency (Fig. 2, A to C). The force peaks at contour lengths ≤135 amino acid residues (≤40 nm) were largely reduced in magnitude and distributed randomly upon increasing the temperature from 25° to 37°C (Fig. 2, A and B). The substantial change of the first force peaks of the force-distance curves suggests the destabilization of the TMHs and polypeptide loops of the N-terminal half of β₂AR (Fig. 1D). Upon further increasing the temperature to 42°C, all force peaks had reduced magnitude and were distributed randomly, thus indicating that the entire secondary structure of β₂AR was destabilized (Fig. 2, G to I). These results show that human β₂AR embedded in DOPC membranes destabilizes considerably at 37°C and higher temperatures.

Fig. 2. The unfolding fingerprint pattern of human β₂AR is sensitive to temperature.

(A to F) Superimposed force-distance curves recorded upon the unfolding of β₂AR from DOPC (A to C) and DOPC:CHS (D to F) lipid membranes at the indicated temperatures. Every density plot was compiled from force-distance curves recorded at six different unfolding speeds ranging from 300 to 5000 nm/s (see Materials and Methods). n gives the number of force-distance curves superimposed. (G to I) Contour length histograms mapping the occurrence of unfolding force peaks. The contour length of every unfolding force peak of every force-distance curve was determined as outlined in Fig. 1. The SMFS data were recorded upon the unfolding of β₂AR embedded in DOPC (pink) and DOPC:CHS (green) membranes at 25°C (G), 37°C (H), and 42°C (I). For each experimental condition, n gives the number of superimposed force-distance curves, which were recorded from more than 60 independent experiments. aa, amino acids.

Next, we characterized the temperature-dependent stability of human β₂AR embedded in DOPC membranes supplemented with the cholesterol analog CHS (Fig. 2, D to I). Force-distance curves recorded at 25°C for β₂AR in DOPC:CHS resembled those recorded in DOPC, thus indicating that CHS had little effect on the number and position of the unfolding force peaks (Fig. 2G). The observation that the unfolding force peaks remained unchanged in position suggests that β₂AR stabilizes the same structural segments and thus exposes the same native fold in DOPC and DOPC:CHS membranes (29). However, the magnitude of the unfolding force peaks was slightly greater in the presence of CHS, which suggests that the mechanical stability of β₂AR was increased (Fig. 2, A and D, and fig. S3). At 37°C, the SMFS spectrum recorded in DOPC:CHS showed patterns very similar to those recorded at 25°C, thus revealing no considerable changes of β₂AR (Fig. 2, D and E). This finding indicates that the structural stability of β₂AR embedded in DOPC:CHS remains largely unaffected by increasing the temperature from 25° to 37°C. Comparing the substantial change in force peaks assigned to the N-terminal region of β₂AR in DOPC with the force peaks detected for β₂AR in DOPC:CHS at 25°C (Fig. 2G) and 37°C (Fig. 2H) suggests that CHS protects this structural region against thermal destabilization.

Further increasing the temperature to 42°C changed the force-distance curves recorded for β₂AR in DOPC:CHS in a similar way as was observed for β₂AR in DOPC at 42°C (Fig. 2, C and F). In particular, the eight force peaks had reduced magnitudes and were distributed randomly, as was also indicated in the probability histogram of the force peaks (Fig. 2I). We can thus conclude that the presence of CHS in the phospholipid membrane moderately increases the mechanical stability of β₂AR at 25°C and protects β₂AR from partially destabilizing at 37°C. However, CHS could not protect β₂AR from destabilizing at 42°C. This thermal destabilization particularly affected the N-terminal structural region of β₂AR, which contains TMH1, TMH2, TMH3, and TMH4. Because these TMHs represent the consensus cholesterol-binding motif that predicts cholesterol binding for 44% of human class A GPCRs (27), these results suggest that the cholesterol analog CHS interacts with and stabilizes this structural region.

CHS modulates the free-energy landscape of β₂AR nonlinearly with temperature

The unfolding force of proteins depends on the speed (loading rate) at which they are mechanically stressed (47, 48). Thus, their mechanical, energetic, and kinetic properties are better described by the concept of a free-energy landscape (47-52). In the absence of any detailed information, the free-energy landscape stabilizing each structural segment of β₂AR against mechanical unfolding is best described by a simple two-state model having a single unfolding barrier (Fig. 3A) (49, 52). To determine the parameters of the free-energy landscape of every structural segment of β₂AR in DOPC and DOPC:CHS membranes, we measured their unfolding forces as a function of loading rate at 25°, 37°, and 42°C by DFS (figs. S3 to S5). The loading rate–dependent unfolding force of every structural segment of β₂AR was fitted with the Bell-Evans model (49, 52, 53) to reveal their free-energy landscape parameters (Fig. 3), including the following: (i) the distance between the folded and the transition state $x_{\mathrm{u}}$, which describes the conformational flexibility (or variability) needed to adopt conformational substates; (ii) the transition rate $k_0$ or lifetime $1∕k_0$; (iii) the stabilizing free-energy barrier ΔG; and (iv) the free-energy landscape roughness $\epsilon$ describing the energetic stability of conformational substates.

Fig. 3. Unfolding free-energy barrier stabilizing the structural segments of β₂AR.

(A) According to the Bell-Evans model (52, 53), a natively folded structure resides in a free-energy valley, which is separated from the unfolded state by a free-energy barrier (black line). Unfolding the structure is initiated by overcoming the transition state of the free-energy barrier. $x_{\mathrm{u}}$ is the distance between the folded and transition states, $k_0$ is the transition rate to overcome the free-energy barrier $\Delta G$ in the absence of an external force (thermal equilibrium), and $\epsilon$ is the free-energy landscape roughness. Externally applying a mechanical force F reduces the thermal likelihood of the structure to reach the transition state. The free-energy projection along the reaction coordinate (pulling direction of the externally applied force) is tilted by the applied mechanical energy −F(cosΘ)x (red dashed line), where Θ is the angle between reaction coordinate and the externally applied force. This tilt decreases the free-energy barrier height (red line), thereby increasing the probability of the folded structure to overcome the transition state toward unfolding. (B) Distance between the folded and transition state ($x_{\mathrm{u}}$) of the structural segments S1 to S8 of β₂AR embedded in DOPC and DOPC:CHS. (C) Transition rates ($k_0$) of the structural segments S1 to S8 of β₂AR embedded in DOPC and DOPC:CHS. (D) Free-energy barrier height stabilizing each structural segment of β₂AR embedded in DOPC and DOPC:CHS. The values of parameters characterizing the free-energy landscape were obtained from fitting the Bell-Evans model to DFS plots recorded at 25°, 37°, and 42°C in the absence (pink) and presence (green) of CHS (see Materials and Methods and figs. S3 to S5). The raw data of the SMFS experiments, the number of force-distance curves analyzed, and their analysis are given in Figs. 2 and 3. More than 300 independent experiments were performed over the time course of 5 years to acquire the necessary data for statistical analysis (Materials and Methods). Statistical significance was determined by comparing the slopes and intercepts of linear regressions of the DFS plots in the presence and absence of CHS ( figs. S3 to S5) by a method equivalent to ANCOVA (70). Differences were considered significant if P values approached *P < 0.1 or **P < 0.05 and if experimentally determined values did not overlap within their SDs (A and C) or their SDs after logarithmical transformation of skewed $k_0$ distribution (B).

Cholesterol increases the conformational flexibility of β₂AR at physiological temperature

At 25°C, the distance of the transition state $x_{\mathrm{u}}$ of each structural segment of β₂AR showed small differences in the absence and presence of CHS (Fig. 3B). In the presence of CHS, the two structural segments S1 and S2 increased $x_{\mathrm{u}}$ moderately from 0.12 to 0.25 nm and 0.15 to 0.25 nm, respectively. At 37°C, however, six of eight structural segments had decreased $x_{\mathrm{u}}$ values in DOPC, whereas S1, S2, S4, S5, and S6 increased their $x_{\mathrm{u}}$ values substantially in the presence of CHS. Although the remaining segments S3, S7, and S8 also showed a tendency to increase their $x_{\mathrm{u}}$ values in CHS, the increase was not statistically significant. Markedly, the $x_{\mathrm{u}}$ values from the structural segments S1, S2, S5, and S6 ranged from 0.15 to 0.21 nm in DOPC and from 0.49 to 1.01 nm in DOPC:CHS. The biggest difference was observed for segments S1 and S5, which, in the presence of CHS, increased $x_{\mathrm{u}}$ by about sixfold to 1.01 and 0.9 nm, respectively. This observation shows that in the presence of CHS at 37°C, five of eight structural segments of β₂AR increased their transition state $x_{\mathrm{u}}$ and thus their conformational flexibility (Fig. 3B and fig. S4). Upon further increasing the temperature to 42°C, the structural segments had substantially decreased $x_{\mathrm{u}}$ values (≤0.2 nm), which was independent of the presence of CHS. At 42°C, no significant differences in $x_{\mathrm{u}}$ values were observed among the structural segments in DOPC and DOPC:CHS.

Cholesterol modulates the functional states of various GPCRs (21, 54, 55), whereas the depletion of cholesterol from the membrane of neonatal cardiac myocytes alters the signaling behavior of endogenous β₂AR (56). NMR and double-electron resonance studies revealed substantial conformational heterogeneity of β₂AR in the apo state (10). Such heterogeneity gives rise to conformational substates between which β₂AR must transition to exhibit basal activity (10). Our results show that in the absence of CHS, β₂AR is exposed to a narrowed conformational space, which implies that the receptor cannot adopt all of the conformational substates available to β₂AR exposed to cholesterol. Such a reduction in the multistate free-energy landscapes was previously predicted to illustrate how GPCRs lose function (51). The observed structural flexibility of segments S1, S2, S5, and S6 in the presence of CHS at 37°C may be interpreted in terms of their relevance for the β₂AR structure-function relationship. For example, segments S5 and S6 stabilize TM5 and TM6, and β₂AR transiting from the inactive to the active state shows large structural outward movements of TMH6 and TMH5, whereas TMH5 extends its α helix toward the cytoplasm (57). On the other hand, TMH1, whose properties are described by segments S1 and S2, is a part of binding pocket for allosteric ligands, which modulate the activity of neurotransmitters and hormones (58). Our results thus suggest that cholesterol assists β₂AR in maintaining a wide conformational space at 37°C, as is required for receptor function (51).

Cholesterol increases the lifetime of β₂AR

At 25°C in DOPC, the first three N-terminal structural segments S1, S2, and S3 showed transition rates $k_0$ ranging from 1.6 to 7.2 s⁻¹. They hence exposed considerably lower kinetic stabilities compared to segments S4 to S8, whose transition rates ranged from 10⁻⁴ to 10⁻¹ s⁻¹ (Fig. 3C). In DOPC:CHS, the kinetic stability of all structural segments exhibited $k_0$ values ranging from 10⁻⁴ to 10⁻¹ s⁻¹, thus indicating that the presence of CHS increased the kinetic stability of structural segments S1, S2, and S3 of β₂AR. At 37°C, most structural segments of β₂AR embedded in DOPC exhibited substantial increases in their $k_0$ values, with structural segments S2, S3, S4, and S5 showing the highest values. This result indicates that β₂AR in DOPC has reduced kinetic stability at 37°C compared to that at 25°C. In the presence of CHS, however, the transition rates of all structural segments remained considerably low, ranging from 10⁻¹¹ to 10⁻¹ s⁻¹, which indicates that CHS kinetically stabilized β₂AR. The highest kinetic stabilization was observed for segments S1, S2, S5, and S6. At 42°C, all structural regions of the receptor exhibited substantial increases in their transition rates regardless of whether the receptor was embedded in DOPC or DOPC:CHS. Together, the results show that CHS kinetically stabilizes β₂AR at 25°C and protects the receptor from kinetic destabilization at 37°C. However, this stabilizing effect of CHS is lost at 42°C.

The observation that the transition rates of the structural segments of β₂AR increase with temperature is similar to what has been described for bacteriorhodopsin (59). Because β₂AR shows structural homologies to bacteriorhodopsin and class A GPCRs, the results suggest that increasing the temperature within the physiological range can kinetically destabilize GPCRs. Although this effect may not hold for the structurally and functionally thermostable transmembrane protein bacteriorhodopsin (60), it is obviously prominent for the structure-function relationship of GPCRs, which are sensitive to much smaller temperature variations (28, 30, 61). For example, the melting temperature of the apelin receptor (APJ), a class A GPCR, lies at ~43°C in the absence of CHS (30). In addition, isothermal denaturation assays in the presence of the strong denaturant GnHCl revealed an eightfold increased stability of β₂AR at 35°C in the presence of CHS (27). Our finding that CHS kinetically stabilizes the β₂AR structure is in general agreement with such findings. However, our results indicate that the action of CHS on β₂AR is likely tailored to the physiological temperature of 37°C. In particular, the properties of individual structural segments of β₂AR embedded in CHS-containing membranes change nonlinearly with temperature.

Cholesterol increases the energetic stability of β₂AR

The free-energy barriers stabilizing individual structural segments of β₂AR at 25°C ranged from 16.4 to 26.0 k_BT in DOPC and from 19.0 to 26.6 k_BT in DOPC:CHS (Fig. 3D). At room temperature, the presence of CHS had no considerable effect in energetically stabilizing the structural regions of β₂AR. However, at 37°C, the free-energy barriers of most of the structural segments were reduced in DOPC, ranging from 16.9 to 19.9 k_BT, whereas they increased in DOPC:CHS to values ranging from 19.0 to 41.7 k_BT. In particular, the energetic stability of segments S1, S2, S5, and S6 increased considerably. At 42°C, the stabilizing effect of CHS was lost and the free-energy barriers were reduced to values ranging from 16.0 to 17.8 k_BT in both DOPC and DOPC:CHS, which supports our initial observation that β₂AR became mechanically destabilized at 42°C (Fig. 2). In summary, these results show that CHS had no substantial effect in energetically stabilizing β₂AR at 25° and 42°C. However, CHS stabilized the receptor at 37°C. One may thus speculate that the functional impairment of β₂AR in the absence of cholesterol (54, 56, 62) may be due to the energetic destabilization of its structure.

Free-energy landscape roughness of the β₂AR structure

Using the free-energy landscape parameters approximated at 25° and 37°C, we calculated the free-energy landscape roughness of each structural segment, $\epsilon$, stabilizing β₂AR (Materials and Methods and Eq. (3) in the absence and presence of CHS (Table 1). This roughness is frequently used to describe the local minima of the free-energy landscape, which trap smaller conformational substates of protein structures (Fig. 3A). Thereby, rougher free-energy landscapes suggest that these smaller conformational substates exhibit higher energetic stabilities (59). The roughness of most structural segments in the absence and presence of CHS ranged from 3.4 to 5.6 k_BT. However, in the presence of CHS, the structural segments S2 and S4 showed a rougher (difference > 1 k_BT) free-energy landscape of 4.6 versus 3.4 k_BT and 5.6 versus 4.5 k_BT, respectively.

Table 1. Energy landscape roughness $\epsilon$ of the structural segments of β₂AR in the absence and presence of CHS.

$\epsilon$ values and SD values were calculated by applying Eq. (3) (Materials and Methods) to the DFS data recorded at 25° and 37°C. Structural segments changing $\epsilon >1$ k_BT are marked by asterisks.

Structural segment	DOPC (kBT)	DOPC:CHS (kBT)
S1	3.9 ± 2.5	4.5 ± 1.0
S2*	3.4 ± 11.6	4.6 ± 1.6
S3	4.6 ± 1.9	4.0 ± 6.0
S4*	4.5 ± 1.4	5.6 ± 11.1
S5	4.8 ± 1.1	4.7 ± 0.6
S6	5.1 ± 5.9	4.5 ± 1.0
S7	4.8 ± 1.5	4.6 ± 19.6
S8	5.4 ± 2.9	5.5 ± 4.0
Mean ± SD of S1 to S8	4.6 ± 3.6	4.8 ± 5.6

The roughness of the free-energy landscape stabilizing the structural segments of β₂AR (~3 to 6 k_BT) was similar to that reported for segments of bacteriorhodopsin (~4 to 6 k_BT) (59). In the presence of CHS, the β₂AR segments S2 and S4, which represent the cytoplasmic end of TMH1, cytoplasmic loop C1, TMH2, and extracellular loop E1, exhibited substantial increases in their free-energy landscape roughness. This increase in roughness suggests that CHS energetically stabilizes the conformational substates of segments S2 and S4. This structural region of β₂AR, which is part of the consensus cholesterol-binding motif of human class A GPCRs, increases the packing constraints of GPCRs upon binding cholesterol (27). Similarly, TMH2, together with TMH4, increases helical packing in the presence of cholesterol, which serves as a bridge between both TMHs (27). Moreover, because TMH4 is considered to have the weakest fold in β₂AR (27), it can be assumed that the cholesterol bridging TMH2 and TMH4 contributes to the overall stabilization of β₂AR (Fig. 3).

MD simulations show segment-dependent interactions of CHS with β₂AR

In our SMFS experiments, the mechanical unfolding of a single β₂AR takes from ~13 to ~265 ms depending on the cantilever speed. Although currently available computing resources in national supercomputing centers are exceptional, they do not enable us to atomistically simulate complex systems such as those studied here over the course of several milliseconds. This implies that simulating the unfolding processes observed experimentally by SMFS is not possible unless one simulates the process at an accelerated speed, which may cause artifactual results because highly accelerated speeds unfold proteins far from thermal equilibrium (63). Given this point, we decided not to use atomistic simulations to explore the unfolding process itself. Instead, we used multimicrosecond simulations to gain insight into the molecular-scale mechanisms by which CHS affected the thermal stability and conformations of β₂AR. Such simulations can capture the spontaneous interaction events (associations and dissociations) of CHS and β₂AR, which enabled us to interpret the CHS-induced effects observed in our SMFS experiments. We thus performed multimicrosecond atomistic MD simulations of β₂AR embedded in DOPC and DOPC:CHS membranes (fig. S6) and calculated the time-correlation functions for the interaction of CHS with the structural segments (fig. S7).

The MD simulation models accounted for seven of the structural segments (S1 to S7) of β₂AR, because the C-terminal end, which, in our experiments, is described as segment S8, has structurally not been solved (27, 36). The volumetric maps showed CHS to preferentially interact with β₂AR at specific sites located at well-defined structural segments (Fig. 4A). In particular, CHS strongly interacted with structural segment S6, which consists of the intracellular parts of TMH5 and TMH6, and of their connecting loop C3. Note that structural segment S6 matches the position of the cholesterol hotspot IC2 as predicted by our previous simulations (36). There is also a high density of CHS observed between two structural segments S1 and S4, which overlaps with the low-affinity interaction site EC3 that was predicted for cholesterol (36). In addition, CHS also interacts with structural segment S7 and, to some extent, with structural segment S5, which confirms our previous work that the EC1 interaction site of cholesterol lies at S7 and partly at S5 (36).

Fig. 4. Atomistic MD simulations of CHS interacting with structural segments of the human β₂AR.

(A) Volumetric maps of the density of CHS (green surface) interacting with human β₂AR at 25°, 37°, and 42°C. At each temperature, the data were averaged from all independent trajectories (table S1). The structural segments of β₂AR are color-coded according to Fig. 1D. (B) Interaction energies resulting from the electrostatic (top) and van der Waals (bottom) interactions of CHS with the indicated structural segments of the human β₂AR at 25°, 37°, and 42°C. Bars and error bars of each structural segment show means ± SD (n ≥ 3, table S1). (C) Structural fluctuations of the indicated structural segments of β₂AR projected along two major principal components. Bars describe the fluctuations of stable structural segments in the absence (pink) and presence (green) of CHS. The error (SD) of the spread estimation is <20%. For a description of the PCA used, see the Supplementary Materials. Differences were considered significant if P values approached *P < 0.1 (calculated from a two-tailed t test) and if the determined values did not overlap within their errors.

The interaction times of CHS with β₂AR depend on the structural segment (Table 2 and fig. S7). In particular, the interaction time of CHS was greatest for segments S1, S4, S5, S6, and S7, which correlates with specific CHS interaction sites revealed by volumetric maps of the CHS density (Fig. 4A). Moreover, the interaction energies revealed strong interactions of CHS with the structural segments S1, S4, S5, S6, and S7 of β₂AR (Fig. 4B). These pronounced interactions of CHS with certain structural segments of the receptor may explain why, in our SMFS experiments, we observed them to be mechanically, kinetically, and energetically stabilized. One exception was segment S2, for which the MD simulations could not explain the experimentally observed stabilization by CHS. We hence speculate that segment S2 may be stabilized by indirect effects of cholesterol on the lipid bilayer properties, such as the modulation of the mechanical properties of the membrane. Together, the MD simulations suggest how membrane-embedded CHS may interact preferentially with certain functionally important regions of β₂AR.

Table 2. Interaction times of CHS with structural segments of β₂AR as revealed from MD simulations.

The average interaction times and SD values were calculated from the decay of time-correlation functions (fig. S7) by identifying the times when the time-correlation function was reduced to a value that was either 10 or 1% of the value at time zero. For the analysis shown, the data used were averaged over the six simulation repeats and the three simulation temperatures to maximize sampling and thus render the segment-based trend as clearly as possible.

Structural segment	Interaction time (10%) (ns)	Interaction time (1%) (ns)
S1	81.4 ± 21.5	284.5 ± 48.3
S2	17.7 ± 10.4	140.5 ± 65.0
S3	31.4 ± 31.0	231.8 ± 182.7
S4	111.8 ± 67.6	295.2 ± 88.8
S5	183.8 ± 52.8	410.9 ± 58.8
S6	129.3 ± 57.9	406.8 ± 203.1
S7	134.1 ± 66.1	492.8 ± 103.4
Mean ± SD of S1 to S7	98.5 ± 69.3	317.7 ± 157.1

To learn about the conformational space explored by β₂AR, we performed a principal components analysis (PCA) of the fluctuations of each structural segment for each given temperature (fig. S8 and Fig. 4C). The spread of the fluctuations was calculated by the two major principal components, PC1 and PC2. At 25°C, β₂AR exhibits similar structural flexibilities in the presence and absence of CHS. Similarly, at 42°C, CHS does not affect the structural flexibility of β₂AR significantly. However, at 37°C, the presence of CHS increases the structural flexibility of segment S6, and partially also segments S5 and S7, which supports our experimental findings.

General overview of temperature-dependent structural properties of β₂AR

To provide a general overview of the temperature-dependent β₂AR stabilization by CHS, the structural properties of β₂AR quantified by SMFS in the presence and absence of CHS (Fig. 3) were mapped onto the β₂AR structure (Fig. 5). The conformational variability, kinetic, and energetic stability detected at 25°, 37°, and 42°C bring our above-described findings together, which will be to discussed in the following section.

Fig. 5. Temperature-dependent kinetic and energetic properties of human β₂AR in the presence and absence of CHS.

(A) Eight structural segments (taken from Fig. 1) mapped to the structure of human β₂AR [PDB ID: 3D4S; (57)]. (B) Distance between the folded and transition states, $x_{\mathrm{u}}$. (C and D) Transition rate $k_0$ (C) and height of the free-energy barrier $\Delta G$ (D) of the structural segments of β₂AR. The individual parameters are explained in Fig. 3, and their values were obtained from SMFS data recorded for β₂AR embedded in DOPC membranes in the absence and presence of CHS at 25°, 37°, and 42°C (Fig. 3).

DISCUSSION

Upon increasing temperature in the absence of CHS, the experimentally determined unfolding pattern of β₂AR reduced force and became noisy (Fig. 2). Both effects describe the destabilization of various structural regions of β₂AR. At 37°C, mostly the structural segments of the N-terminal region of β₂AR (TMH1 to TMH5) were destabilized and exhibited low kinetic stabilities (Fig. 5). In the presence of CHS, this structural region of β₂AR, which contains the consensus cholesterol-binding motif of human class A GPCRs (27), was substantially stabilized. However, because cholesterol also modulates the mechanical properties of the membrane, one may speculate that these mechanical properties may modulate the β₂AR stability as well (64). In our SMFS experiments, we characterized the outcomes of both effects, the direct effects resulting from physicochemical interactions and the indirect effects resulting from the mechanical properties of the lipid membrane on the stability of β₂AR.

In the absence of CHS, the conformational flexibility of the structural segments of β₂AR decreased with increasing temperature from 25° to 37° and 42°C (Fig. 5). The decreasing flexibility indicates that the number of conformational states of the receptor was reduced, which is consistent with the picture of kinetic and energetic destabilization of β₂AR (Fig. 5, C and D). However, in the presence of CHS, the conformational flexibility of β₂AR increased upon increasing temperature from 25° to 37°C. Simultaneously, the kinetic and energetic stabilities of the receptor increased. At the further increased temperature of 42°C, all three parameters—the conformational flexibility, kinetic stability, and energetic stability of β₂AR exposed to CHS—decreased substantially to values that were observed for the destabilized receptor in the absence of CHS. Thus, CHS modulates the structural properties of β₂AR specifically at 37°C. At all three temperatures investigated, MD simulations showed similar probability distributions for the interactions of CHS with β₂AR (Fig. 4A). However, CHS interacted with the structural segments S1, S5, S6, and S7 for greater times (Table 2 and fig. S7). Overall, the atomistic MD simulations support the experimentally observed nonlinear stabilization of β₂AR by CHS.

Our experiments and simulations indicate that CHS stabilizes specific structural regions of human β₂AR. The largest CHS effects on energetic and kinetic stabilization and conformational space were observed for structural regions that are of importance for β₂AR signaling (TMH5, intracellular loop 3, and TMH6) and the binding of allosteric ligands (TMH1, TMH2, TMH6, and TMH7) (58). The stabilization of these structural regions was particularly effective at 37°C, which is physiologically the most relevant temperature β₂AR. At 42°C, the protective function of CHS was lost and β₂AR became destabilized. Whereas this type of destabilization may not necessarily lead to immediate thermal denaturation of β₂AR, it most probably affects the structure-function relationship of the receptor. This nonlinearity suggests that our understanding of how CHS modulates the structural properties of a GPCR at 25°C cannot be necessarily projected to describe how CHS modulates the receptor at 37° or 42°C. More generally, it is essential to characterize the action of cholesterol on GPCRs at physiologically relevant temperatures. Given the structural and functional homology of GPCRs (65), we suggest that our findings about how temperature and cholesterol modulate the structural properties of β₂AR nonlinearly may be generalized to other class A GPCRs.

In the future, the combination of experimental SMFS studies and theoretical MD simulations presented here may be applied to systematically quantify how functionally or disease-related point mutations modulate the CHS-promoted temperature-dependent (de-)stabilization of specific structural regions of β₂AR. Hereto, point mutations of the key residues in the consensus cholesterol-binding motif and in functionally important structural regions may represent promising targets to start with.

MATERIALS AND METHODS

β₂AR subcloning, expression, purification, and reconstitution

Human β₂AR was expressed and purified as described previously (66). Briefly, β₂AR with a truncated C-terminal end amino acids) and an N-terminal FLAG epitope followed by a tobacco etch virus protease cleavage site was expressed in Spodoptera frugiperda (Sf9) insect cells for ~48 hours. To facilitate protein expression, we used a C-terminal truncated β₂AR construct, which does not affect β₂AR signaling (57). Successful expression of β₂AR was evaluated by immunofluorescence. Cells expressing β₂AR were harvested by centrifugation at 5000g for 15 min and stored at −80°C. To purify β₂AR from Sf9 cells, a three-step purification procedure was used. β₂AR was reconstituted in DOPC:CHS (10:1, v/v) or DOPC lipid vesicles as described previously (66). Briefly, a lipid:β₂AR mixture was mixed with reconstitution buffer and kept on ice for 2 hours. The lipid-to-protein ratio was 1000:1 (mol/mol). Detergent was removed with a 25 cm–by–0.8 cm Sephadex G-50 (fine) column and reconstitution buffer [100 mM NaCl, 20 mM Hepes (pH 7.5)].

Single-molecule force spectroscopy

β₂AR samples were aliquoted and flash-frozen in liquid nitrogen. For each SMFS experiment, a new sample was thawed freshly and used once. Proteoliposomes of β₂AR reconstituted in DOPC or DOPC:CHS (10:1, w/w) were adsorbed to the freshly cleaved surface of mica in SMFS buffer solution [300 mM NaCl, 25 mM MgCl₂, 25 mM tris (pH 7.0)] for 1 hour. After the adsorption time passed, the samples were washed several times with SMFS buffer to remove weakly attached membrane patches. SMFS was performed at 25°, 37°, and 42°C using fully automated AFM (ForceRobot 300, JPK Instruments) and a temperature-controlled sample holder (High Temperature Heating Stage, JPK Instruments). Every freshly thawed β₂AR sample was characterized by SMFS for a maximum of 5 hours after preparation and thereafter discarded. Data for different temperatures and lipid compositions were recorded in random order. Sixty-micrometer-long silicon nitride cantilevers (A-BioLever, BL-RC150VB, Olympus, Japan) were calibrated in SMFS buffer before and after each SMFS experiment with the equipartition theorem. SMFS was recorded at different speeds of cantilever retraction (300, 600, 900, 1200, 2500, and 5000 nm s⁻¹). Data at 2500 and 5000 nm s⁻¹ were recorded with an external 16-bit data acquisition card (NI PCI-6221, National Instruments).

SMFS data analysis

In total, ~6 × 10⁶ force-distance curves from more than 300 independent SMFS experiments were recorded for this work over the time course of 5 years. A mechanically fully unfolded and stretched β₂AR extends to distances of 70 to 90 nm (29, 38). Thus, only force-distance curves showing a force peak pattern extending to distances >70 nm were selected for analysis. Consistent with previous results (29, 38), we observed the N-terminal end of β₂AR to predominantly attach to the AFM stylus compared to the C terminus (72%, n = 210). For statistical reasons, we hence focused on the analysis of force-distance curves, which recorded the mechanical unfolding of β₂AR from the N-terminal end. About 100 force-distance curves recording the mechanical unfolding from the N-terminal end were analyzed for each unfolding speed to minimize the SEM. Every selected force-distance curve was then fitted by applying the WLC model, with a persistence length of 0.4 nm and a contour length of 0.36 nm per amino acid residue (67). This WLC fit of each unfolding force peak provided the rupture force required to mechanically unfold a structural segment, as well as the contour length (in amino acid residues) of the unfolded and stretched polypeptide. All contour lengths and rupture forces of all unfolding force peaks were grouped and analyzed for each experimental condition. Each unfolding force peak assigns the beginning and end of a structural segment stabilizing the unfolding β₂AR structure. The eight main force peaks detected were mapped to the secondary structure of β₂AR as described previously (29, 38). To locate the force peaks within or on the opposite side of the lipid membrane relatively to the pulling AFM cantilever a membrane thickness of ~4 nm, which corresponds to the contour length of an ~11–amino acid residue–long polypeptide stretch, has been considered to approximate the total contour length of the unfolded and stretched polypeptide (45, 68, 69). In this so-called “membrane compensation,” the contour length of the unfolded and stretched polypeptide such as was estimated by fitting the force peak with the WLC model is extended by the contour length corresponding to the membrane thickness. Furthermore, at the relatively low unfolding forces applied in our study, the highly conserved disulfide bridge between residues Cys¹⁰⁶ and Cys¹⁹¹ of β₂AR remains intact (38). Because the polypeptide stretch held together by the much stronger disulfide bridge did not unfold in our SMFS experiments, it was not accounted to localize the contour length of the unfolded and stretched polypeptide to a structural segment of β₂AR.

Quantifying free-energy landscape parameters

According to the Bell-Evans model (53), the most probable unfolding force $F^{\ast }$ is a function of the loading rate $r$

$$F^{\ast }=\frac{k_{\mathrm{B}}T}{x_{\mathrm{u}}}\text{ln}\left(\frac{x_{\mathrm{u}}r}{k_{\mathrm{B}}T{k}_0}\right)$$ (1)

where $k_{\mathrm{B}}$ is the Boltzmann constant and $T$ is the absolute temperature. The most probable unfolding force and loading rate for every force peak (structural segment) at every speed were calculated, taking the maximum of the Gaussian fit of the rupture force and loading rate distributions. Applying Eq. (1), we then calculated $x_{\mathrm{u}}$ and $k_0$ for each structural segment depending on the lipid composition and temperature. The height of the free-energy barrier $\Delta G^{‡}$ was calculated by applying Eq. (2)

$$\Delta G^{‡}=-k_{\mathrm{B}}T\text{ln}({\tau }_{\mathrm{A}}{k}_0)$$ (2)

Using the DFS parameters, we applied Eq. (3) to quantify the average roughness $\epsilon$ of the free-energy landscape in the presence and absence of CHS

$$\begin{gathered} {\epsilon }^2\approx \frac{x_{\mathrm{u}}(T_2)k_{\mathrm{B}}{T}_2{x}_{\mathrm{u}}(T_1)k_{\mathrm{B}}{T}_1}{x_{\mathrm{u}}(T_2)k_{\mathrm{B}}{T}_2-x_{\mathrm{u}}(T_1)k_{\mathrm{B}}{T}_1}\left[\Delta G{{}_{\mathrm{u}}}^{\ast }\left(\frac{1}{x_{\mathrm{u}}(T_1)}-\frac{1}{x_{\mathrm{u}}(T_2)}\right)+ \\ \frac{k_{\mathrm{B}}{T}_1}{x_{\mathrm{u}}(T_1)}\text{ln}\left(\frac{r_f(T_1)x_{\mathrm{u}}(T_1)}{k_0(T_1)k_{\mathrm{B}}{T}_1}\right)-\frac{k_{\mathrm{B}}{T}_2}{x_{\mathrm{u}}(T_2)}\text{ln}\left(\frac{r_f(T_2)x_{\mathrm{u}}(T_2)}{k_0(T_2)k_{\mathrm{B}}{T}_2}\right)\right] \end{gathered}$$ (3)

Statistical data analysis

Overall, more than 300 independent SMFS experiments were performed to acquire the necessary statistics for each of the six experimental conditions (covering both temperature and lipid composition). To access the statistical significance of the differences observed in the presence and absence of CHS, the slopes and intercepts of linear DFS fits were compared (figs. S3 to S5). A method equivalent to analysis of covariance (ANCOVA) (70) and implemented in GraphPad Prism was used. Comparison of the slopes was used to access statistical significance for differences in $x_{\mathrm{u}}$, and comparison of the intercepts was used to access statistical significance for differences in $k_0$. Differences were considered significant when $P$ values approached *P < 0.1 and **P < 0.05 and mean values did not overlap within their SDs for $x_{\mathrm{u}}$ and $\Delta G$, and their SDs after logarithmical transformation for $k_0$, because the distribution of $k_0$ is right-skewed (Fig. 3, B to D).

All-atom MD simulations

We performed atomistic MD simulations of β₂AR embedded in a DOPC lipid bilayer in the presence and absence of 10 mol % of CHS, at 25°, 37°, and 42°C (Fig. 5 and table S1). Systems explored in this work contained one β₂AR placed in a membrane composed of 202 to 338 lipid molecules. All systems were explicitly solvated by water together with counter ions added to achieve electroneutrality with 150 mM NaCl. Each system was first energy-minimized to remove bad contacts. After energy minimization, we simulated the systems for 25 to 50 ns with position restraints on the receptor heavy atoms and then for another 25 to 50 ns with position restraints only on the receptor backbone atoms. Subsequently, all restraints were released, and every system was subjected to 2.5 μs of simulation, with three simulation repeats for DOPC systems and with six simulation repeats for DOPC:CHS systems (table S1). All simulations were performed with the GROMACS 5.0.4 package (71) using the all-atom optimized potentials for liquid simulations (OPLS-AA) force field (72). Parameters for CHS were as described previously (73). As used previously for DOPC (36), we used torsional and Lennard-Jones parameters for saturated (74) and unsaturated hydrocarbons (73, 75) and the torsional potential for the glycerol backbone and the phospholipid head group (74). Water molecules were modeled with the TIP3P water model, which is compatible with OPLS parameterization (76). All simulations were performed in the isothermal-isobaric (NpT) ensemble. The v-rescale (stochastic velocity rescaling) thermostat (77), with a time constant of 0.1 ps, was used to maintain the simulation temperature. The temperatures of β₂AR, lipids, and solvent (water and ions) were controlled independently. The pressure of the systems was maintained at 1 bar with the Parrinello-Rahman barostat (78) with a time constant of 1 ps. A semi-isotropic scheme was used for pressure control. Simulations were performed with a time step of 2.0 fs. To preserve the lengths of covalent hydrogen bonds, the LINCS algorithm (79) was applied. Periodic boundary conditions were applied in all three directions. Van der Waals interactions were treated using the Lennard-Jones potential with a cutoff distance of 1.0 nm. Long-range electrostatic interactions were evaluated with the particle mesh Ewald algorithm (80) using a real-space cutoff of 1.0 nm, a β-spline interpolation (order of 6), and a direct sum tolerance of 10⁻⁶. The simulation model used was validated in an exhaustive series of >100-μs atomistic simulations (36).