Hidden water’s influence on rhodopsin activation

Abstract

Structural biology relies on several powerful techniques, but these tend to be limited in their ability to characterize protein fluctuations and mobility. Overreliance on structural approaches can lead to omission of critical information regarding biological function. Currently there is a need for complementary biophysical methods to visualize these mobile aspects of protein function. Here, we review hydrostatic and osmotic pressure-based techniques to address this shortcoming for the paradigm of rhodopsin. Hydrostatic and osmotic pressure data contribute important examples, which are interpreted in terms of an energy landscape for hydration-mediated protein dynamics. We find that perturbations of rhodopsin conformational equilibria by force-based methods are not unrelated phenomena; rather they probe various hydration states involving functional proton reactions. Hydrostatic pressure acts on small numbers of strongly interacting structural or solvent-shell water molecules with relatively high energies, while osmotic pressure acts on large numbers of weakly interacting bulk-like water molecules with low energies. Local solvent fluctuations due to the hydration shell and collective water interactions affect hydrogen-bonded networks and domain motions that are explained by a hierarchical energy landscape model for protein dynamics.

Significance

Disordered water molecules are routinely omitted from structures of proteins, yet their presence and movement can have a profound influence on macromolecular function. Common structural techniques such as x-ray diffraction and cryogenic electron microscopy are unable to visualize disordered solvent molecules. Development of methods to probe these waters is crucial for a complete understanding of the energy landscapes and functional dynamics of proteins.

Introduction

The influence of solvent—including lipids for membrane proteins—is critical for protein activity (1,2,3,4), but our understanding of this interaction lags behind other aspects of structural dynamics and function (5,6,7). Water is involved in a variety of processes that govern hydrogen bonding networks, proton reactions, and hydrophobic effects (4,8,9,10,11,12,13,14), with few universal techniques available for their investigation. Some of the most common ways to obtain the structure of a protein include methods such as x-ray diffraction and cryogenic electron microscopy (cryo-EM). However, neither of these techniques investigates disordered regions due to their nonhomogeneous arrangements within individual proteins (15,16). In x-ray crystallography these disordered regions do not diffract in a way that constructively interfere, and so they do not appear in the resultant diffraction pattern. Correspondingly, in cryo-EM the disordered regions will appear in different locations for each image, and when they are combined in class averages these regions fail to add together and remain undetected (16).

Indeed, one of the key advances that allowed for high-resolution cryo-EM of biomolecular structures was flash-freezing samples to prevent crystallization of the solvent and instead produce vitreous water, specifically to avoid water diffraction from interfering with the resulting images (17). This breakthrough allowed for the modern field of cryo-EM microscopy, which has produced exciting new structures of membrane proteins (18). Yet such a development also precludes cryo-EM from being able to study the structural and dynamic effects of these same water molecules. The loss of information about disordered regions is a fairly well-known limitation of these techniques, and there is evidence to suggest that disordered regions of proteins can play an important role in their functionality (19). This drawback is often only considered in the context of the conformation of the peptide chain itself. Much of the environment surrounding the protein is disordered as well, meaning these techniques also lose information about the aqueous solvent and the membrane lipids (Fig. 1) (15). Increasingly there are results suggesting that the solvent is fundamentally entangled with macromolecular function and that studying proteins in isolation presents challenges to understanding their roles in living systems (20,21,22,23).

Figure 1.

Protein interactions with soft matter are potentially hidden in static structures created by x-ray crystallography or cryogenic electron microscopy. (A) Representative cryo-EM structure of visual rhodopsin used as a membrane protein archetype (26) (PDB: 6OFJ). Cryogenic temperatures or crystal lattice forces can restrain conformational fluctuations that occur in solution, within a membrane lipid bilayer, or a detergent micelle. (B) Rhodopsin interactions with soft matter (lipids, detergents, water) include conformational transitions associated with changes in hydration that involve water (blue) influx into the protein core (5,7). A comprehensive picture of protein dynamics includes both equilibrium structure and fluctuations. Adapted with permission from (5). Copyright 2014 American Chemical Society.

To begin to understand the influences of this disordered environment, a method to manipulate the solvent in a controlled way is currently needed. Two of the most fundamental thermodynamic state variables are temperature and pressure (24), and secondary to these are the isothermal compressibility and isobaric thermal expansion coefficients. Temperature is often chosen as the independent variable because it can be easily manipulated, as explored previously by Frauenfelder et al. (25). Even then, simple changes in temperature are insufficient to selectively probe the disordered environment of a protein. Accordingly, we propose to examine this question using pressure as the independent variable, and here we focus on its effects together with the isothermal compressibility. For this purpose, we turn to the force-based techniques of applying hydrostatic and osmotic pressure. Both are capable of probing water interactions in the surroundings of a protein, allowing for experiments to measure the impact of changes arising from the solvent. Here, we review and discuss how these techniques can probe this disordered solvent environment using rhodopsin as the archetypical G-protein-coupled receptor (GPCR) model system.

Rhodopsin as a model membrane protein

GPCRs are a class of membrane-bound proteins that are involved in many cell-signaling pathways, and include the opioid receptors, the β₂-adrenergic receptor, and numerous receptors implicated in cancer such as cysteinyl-leukotriene receptor 2 (CysLTR2) (27,28,29,30,31,32). Because of this, GPCRs are among the most commonly drugged targets, with 34% of all pharmaceuticals approved by the Food and Drug Administration acting on them (33). Once activated, these receptors interact with their effector G-proteins triggering a signaling cascade. Rhodopsin is a member of the aptly named rhodopsin-like class A GPCRs and is responsible for vision in dim light (30). As a result, it makes an ideal model protein because it can be easily activated and characterized and is readily purified from bovine retinas. Indeed, much of the underpinnings for the textbook fluid-mosaic model comes from early groundbreaking experiments with rhodopsin that showed both rotational and translational diffusion of the protein within the plane of the membrane (34,35,36). Subsequent studies of rhodopsin have led to a new membrane model called the flexible surface model (FSM), which is based on out-of-plane (curvature) proteolipid couplings (37,38). Importantly, the roles of membrane lipids and water in the activation dynamics of rhodopsin and other class A GPCRs have much in common (39,40).

Rhodopsin is bound to the cofactor retinal through a Schiff base linkage (41), which is the origin of its conformational change upon activation (40,42,43). Before activation by light 11-cis retinal acts as an antagonist, holding rhodopsin in an inactive state. Upon absorbing a photon, the 11-cis retinal isomerizes to all-trans converting the cofactor into an agonist. This isomerization propels rhodopsin through a series of excited states, which culminates in the establishment of an equilibrium between its preactive conformation, metarhodopsin-I (MI), and its active conformation, metarhodopsin-II (MII). The process can be represented as the following mechanism: Rh + hv → MI ⇌ MII_a ⇌ MII_b + H₃O⁺ ⇌ MII_bH⁺, where Rh is the dark state of rhodopsin, hv is a photon of light, MI represents the preactive state with all-trans retinal and a protonated Schiff base, MII_a is the state with a deprotonated Schiff base but in the preactive conformation, MII_b indicates the active state, and MII_bH⁺ is the active state further stabilized with Glu¹³⁴ of the E(D)RY motif protonated (23). Each of these intermediates represents a specific conformational substate of rhodopsin that is typically studied through conventional structural biology approaches without regard for solvent effects. Yet the extracellular and cytosolic faces are surrounded by water during these transitions, and there has been substantial evidence to show that such environmental water is coupled to key conformational changes within proteins (6,44,45).

Hierarchical energy landscapes describe functional hydration of membrane proteins

After absorption of a photon and subsequent isomerization of the retinal cofactor, rhodopsin adopts an equilibrium between its preactive conformation, MI, and its active conformation, MII, as described above. The equilibrium can be thought of as populating an energy landscape, where different conformations each occupy a basin with an energy barrier between them (Fig. 2) (46,47,48). Some states will be occupied more than others, with the exact occupancy governed by the thermodynamic stability of each conformation. Even uncommon conformations can be biologically relevant, such as a rare state that binds to a ligand (49). The energy landscapes of proteins correspond to a manifold that can be separated into various levels or tiers based on the barriers of the corresponding transitions, as proposed by Frauenfelder et al. (46). The transitions with the highest energy barriers correspond to movements of entire protein domains and are assigned to tier-0 and distinguish basins of the next tier (tier-1). Such basins contain all conformational substates accessible by tier-1 fluctuations. For instance, transitioning between the dark state of rhodopsin and its preactive MI conformation would constitute a tier-0 transition, with the dark state and MI constituting tier-1 basins. These tier-1 basins include the conformations of proteins that are often specifically isolated and studied. Still, these conformations are not truly static, but rather are an average of many smaller substates among which the protein fluctuates (50). These smaller fluctuations make up the lower tiers of the energy landscape. Fluctuations with the next highest energy barriers after tier-0 are the tier-1 fluctuations and correspond to interconversion between various side-chain conformations. The lowest tier that will be discussed in this paper is tier-2, which corresponds to reorganization of structures within the protein, such as helices, loops, and turns (49,51).

Figure 2.

Energy landscape model describes functional dynamics of proteins. (A) Side view of a hierarchical multidimensional energy landscape (EL) where different protein states occupy basins of attraction separated by energy barriers. The EL manifold comprises various tiers: movements of entire protein domains encompass tier-0 fluctuations; local interconversion between amino acid side-chain conformations corresponds to tier-1 fluctuations; and tier-2 fluctuations consist of movements of larger structures such as helices, loops, and turns. These fluctuations are contained within a basin of the same tier, e.g., a tier-1 basin contains multiple tier-1 fluctuations, and is separated from other tier-1 basins by tier-0 fluctuations. Top right inset: expansion of a single β-basin (synonymous with a tier-1 basin). (B) View from the top of multidimensional energy landscape, showing how a single β-basin fits into the larger energy landscape. A tier-0 basin contains all macrostates (represented as β-basins) of the protein. An individual β-basin contains numerous α-basins. Within an α-basin the protein undergoes rapid α-fluctuations and randomly samples all conformations contained in the α basin. Less frequently, the protein experiences β-fluctuations and undergoes movement among different α-basins. Previously, the larger-scale protein conformational changes were assigned to tier-1 and coupled to the bulk-like solvent (α-fluctuations); the smaller-scale changes were assigned to tier-2 and coupled to the hydration shell (β-fluctuations) (49,52). However, these assignments are now reversed (51). The smaller-scale β-fluctuations are assigned to tier-1 while larger-scale α-fluctuations are assigned to tier-2 (51) (confer figure). Although the α-fluctuations are lower in energy than β-fluctuations, they cause larger-scale conformational changes due to the greater numbers of bulk-like solvent molecules compared with the hydration shell.

Even within the same tier, the transitions can vary in their energy barriers (different peak heights), and the resulting conformational states will not have exactly the same energy (different basin depths). It follows that the energy landscape is not flat, and the transitions are not uniform, resulting in some states being favored over others. Some of these equilibrium fluctuations have been studied in detail previously, namely the α-and β-fluctuations. By analogy to glass-forming liquids the β-fluctuations are coupled (or slaved) to the solvent shell, while α-fluctuations are coupled to the bulk-like solvent (52,53,54). At ambient temperature β-fluctuations are assigned to tier-1, and α-fluctuations are assigned to tier-2 (51). For the example of rhodopsin, varying conditions such as hydration level or isomerization of its chromophore can alter the depth of these basins, and thus the observable equilibrium between different states. Understanding this energy landscape, and how it is influenced by the rhodopsin environment, is essential for deciphering the receptor function. Although conceptually useful, the thermodynamics of this energy landscape are not well characterized. Development of methods to study and characterize this complex landscape is crucial to furthering our understanding of how protein structure, dynamics, and function are related.

Force-based methods for determining protein hydration

As introduced by Parsegian et al. (55), osmotic stress is a technique that relies on using osmolytes to measure macromolecular forces and is often used to probe the hydration states of proteins (1,3,56,57,58). Application of osmotic stress can be used to quantitatively determine the number of waters involved in specific conformation changes of proteins (55), as also shown by Record and co-workers (59,60). This technique has been applied to a variety of proteins, including both globular and membrane proteins, where it has been shown that uptake or expulsion of water is involved in conformational state changes. As examples, when fully saturated with oxygen the globular protein hemoglobin takes up ∼60 water molecules (56). Conversely, upon binding of glucose, hexokinase expels ∼320 water molecules (57). With regard to nucleic acids, osmotic stress has been used to investigate preferential interactions and excluded volume effects on DNA duplex hairpins (61). Approximately 110 additional waters are taken up by the nonspecific DNA complex of the restriction nuclease EcoRI compared with the specific DNA complex (62). Similarly, osmotic stress has been used to show that water is involved in interactions of amphiphilic peptides and membrane protein conformational changes. As revealed by Bezrukov and co-workers, the ionic channel alamethicin takes up ∼100 water molecules upon transition from its closed to successive open states (63). These examples illustrate the general applicability of the approach, particularly in relation to drug design (see below).

In the case of visual rhodopsin, the transition from the preactive MI state to the active MII state causes approximately 80 water molecules to enter the interior of the protein (5,23,38,64,65). Analogously, light activation of phytochrome causes it to take up ∼80 waters, as shown by pioneering studies of Alexiev and co-workers (58). Given the ubiquitous role of the solvent in these protein transitions, it is no surprise that the significance for drug binding has been considered in previous studies. Water has been shown to be involved in key H-bonding bridges that allow for peptides to form nanopores in membranes with high specificity (66). The aqueous solvent is known to play a key role in mediating hydrogen bond networks that facilitate conformational changes of the opioid receptor upon drug binding (67,68). Simulation of water interactions with opioid receptors has allowed for more accurate interaction energies to be calculated (69). Dynamics of lipids, which solvate membrane proteins, have also been shown to block binding sites on the β₁-adrenergic receptor (70). Better understanding of these solvent-mediated interactions is necessary for interpreting various drug binding events and improving efficacy.

Application to rhodopsin

Here, we use rhodopsin as a specific example of how the osmotic stress techniques can be applied to a functionally significant membrane protein. To determine the effect of osmotic stress on the activation of rhodopsin, samples are dehydrated by addition of increasing concentrations of polymer osmolytes such as polyethylene glycol (PEG). The polymer PEG is chosen due to its prevalence in crystallography, its biomedical significance, and due to the vast range of sizes available. Activation of rhodopsin can then be assessed using UV-visible spectroscopy to measure the fraction of rhodopsin in the activated MII conformation. This measurement is facilitated by the large shift in the absorption maximum of rhodopsin upon its transition from the preactive MI state (480 nm) to the active MII state (380 nm) (71). Basis (reference) spectra of MI and MII are collected at a pH of 9.5 and 10°C or pH 5 and 21°C, respectively. The observed spectrum is thus fit as a linear combination of these two basis spectra, where the coefficients for each basis spectrum correspond directly to the fraction of the associated species (65).

The equilibrium constant is then calculated as [MII]/[MI], where it is found that, in the presence of large molar mass PEGs, the active MII fraction of rhodopsin is reduced significantly. Over a small range of osmotic pressures (∼1–5 MPa) the data can be fit to a linear formula which reads

$$\begin{array}{c}\ln K=\ln K°-\left(\frac{\Delta V°}{RT}\right)\mathrm{\Pi }\text{.}\end{array}$$ (1)

Here, K is the equilibrium constant for the transition between the MI and MII states under the measurement conditions, K° is the equilibrium constant under standard conditions, ΔV° is the change in hydrated volume, and Π is the osmotic pressure. While Eq. 1 holds for small pressure ranges, over larger pressure intervals the data become nonlinear, which can be attributed to the compressibility of the hydrated volume or to specific (quinary) interactions of the osmolyte with the protein. To model this behavior, a virial expansion of the chemical potential that leads to ln K as a function of osmotic pressure is used (7):

$$\begin{array}{c}\ln K=\ln K°-\left(\frac{\Delta V°}{RT}\right)\mathrm{\Pi }+\left(\frac{1}{2}\Delta C\right){\mathrm{\Pi }}^2\text{,}\end{array}$$ (2)

where the linear term is related to the change in hydrated volume and the quadratic term is described in terms of the C virial coefficient.

Application of hydrostatic pressure is another force-related technique that can be used to investigate protein dynamics in both membranes and solutions. While it is useful in pressure denaturation studies, it is also capable of influencing the equilibrium between protein states with different volumes as shown by Royer and coworkers (24,70,72,73,74,75,76). Accordingly, the hydrostatic pressure probes hydration of rhodopsin in a way that is complementary to osmotic pressure measurements due to fundamental differences in the perturbations, i.e., closed vs. open systems (65). The most immediate difference is that comparatively large hydrostatic pressures are needed to effect changes similar to osmotic pressure. Because the change in volume and pressure are inversely related this ensures that only small volumes of water are transferred, while the comparatively lower osmotic pressures allow for probing the movements of larger volumes of water (23). The total energy ($P\Delta V$) exerted by each technique is similar, so that the smaller volumes of water probed in hydrostatic pressure experiments also mean that the waters are more tightly associated with rhodopsin. These two factors mean that hydrostatic pressure experiments are likely to probe water movements into small cavities or involving the solvent shells of the protein, as opposed to osmotic pressure studies that probe bulk-like water movements.

The effects of hydrostatic pressure on the MI to MII equilibrium of rhodopsin have been assessed by high-pressure UV-visible spectroscopy, using a high-pressure optical bomb (77). The data collection and analysis were performed as described above for osmotic stress experiments. The result for the dependence of the MI-MII equilibrium constant on hydrostatic pressure reads:

$$\begin{array}{c}\ln K=\ln K°-\left(\frac{\Delta V°}{RT}\right)P+\left(\frac{1}{2}\Delta C\right)P^2\text{.}\end{array}$$ (3)

The above formula is similar to Eq. 2, except that the conjugate variables are now hydrostatic pressure ($P$) instead of osmotic pressure ($\Pi$) and change in protein volume ($\Delta V°$) instead of change in hydrated volume. In both Eqs. 2 and 3 the term containing the virial coefficient $\Delta C$ is related to the isothermal compressibility

$$\begin{array}{c}{\kappa }_T=-{\left(\frac{\partial \ln V}{\partial P}\right)}_T\end{array}$$ (4)

by Eq. 5 as given below:

$$\begin{array}{c}\Delta C=\frac{{\overline{V}}_0\Delta {\kappa }_T}{RT}\text{.}\end{array}$$ (5)

Here, ${\overline{V}}_0$ is the molar volume of the protein and $\Delta {\kappa }_T$ is the change in isothermal compressibility between the MI and MII states.

While the similarity of Eqs. 2 and 3 might imply that hydrostatic and osmotic pressure act similarly, it is important to note that there are also fundamental differences. Adding an osmolyte effectively dilutes the solvent, lowering the chemical potential of water wherever it is present. Since osmolytes are generally excluded from the interior of proteins, a chemical potential gradient will thus be created. As water moves down this gradient it is withdrawn from the interior of the protein. This means that osmotic stress is acting directly on the relevant volumes of water within an open thermodynamic system. By contrast, hydrostatic pressure acts on the volumes of both water and macromolecules. Under hydrostatic pressure, the chemical potential of both macromolecules and the solvent—including lipids for membrane proteins—is increased. The solvent and protein will move down the chemical potential gradient by decreasing their volume. To illustrate the difference, consider the effect of positive pressure. Applying positive hydrostatic pressure will cause the volume of the solution to be reduced, tending to move water into the solvent shell or internal cavities where its density is greater. Conversely, positive osmotic pressure will cause water to enter the protein, thereby increasing its volume. In actuality, both techniques decrease the protein volume because the addition of osmolytes creates a negative osmotic pressure. Keeping these fundamental differences in mind is important, as they allow the two techniques to interact with different sets of water and serve as complementary methods for exploring the energy landscapes of proteins.

Energy landscape for membrane protein hydration is probed by hydrostatic and osmotic pressures

We next show that osmotic and hydrostatic pressure measurements can reveal the biomembrane energy landscape due to functional changes in protein hydration, with rhodopsin as an important example. Fig. 3 shows how the equilibrium constant for the MI-MII transition varies with increasing concentrations of various sizes of PEG polymers for osmotic pressures ranging from 0 to ∼10 MPa (recall that 1 MPa = 9.87 atm). Note that an increase in the equilibrium constant corresponds to a higher fraction of MII, i.e., to more rhodopsin being activated, and vice versa. In consequence, large PEGs are found to back shift the reaction, favoring the preactive MI state. Surprisingly, all the large-molecular-weight PEGs behave virtually identically, supporting the effects being colligative in nature. However, while the data are linear over a small range of osmotic pressures, at higher osmotic pressure they become nonlinear. This behavior is represented in Eq. 2 by the quadratic term.

Figure 3.

Dependence of metarhodopsin activation equilibrium on osmotic pressure reveals back shifting to preactive MI state. Dehydration of the protein by large polymer osmolytes occurs due to nonspecific colligative properties. The logarithm of the MI-MII equilibrium constant (K = [MII]/[MI]) is plotted against osmotic pressure for a series of large polyethylene glycol polymer osmolytes (PEGs) showing an approximately second-order relation (T = 15°C). Universal colligative behavior is found for PEGs with M_r between 1000 and 6000 Da, where the linear term is proportional to the hydrated volume change. At higher osmotic pressure the second-order curvature term arises from the compressibility of the hydration volume and/or specific PEG interactions with the protein. Inset: illustration of how rhodopsin activation equilibrium is back shifted to the preactive MI (closed) state by hydrophilic polymers that are entropically excluded and dehydrate the protein. Reprinted with errors bars from (7). Copyright (2022) National Academy of Sciences.

Analogously, the MI to MII equilibrium has been investigated under hydrostatic pressure leading to a quantitatively similar trend (Fig. 4). While Figs. 3 and 4 look very similar, however, the ranges of pressures are vastly different (∼10 vs. ∼250 MPa). Fig. 4 shows how the MI-MII equilibrium constant of rhodopsin in native retinal disk membranes changes when subjected to high pressure. At high hydrostatic pressures, rhodopsin becomes increasingly back shifted toward the preactive MI state. This system also exhibits an apparent comprehensibility, with the data becoming nonlinear at high hydrostatic pressures. The inset of Fig. 4 summarizes how the equilibrium constant and the volume change are affected by various detergents. There are two trends observable in these data: first, the equilibrium constant increases in strong detergents, meaning that active MII is favored. Second, in these detergents the volume change is essentially zero. Since the volume change is related to the shifting of the equilibrium, this means that in such detergents almost no back shifting is observed.

Figure 4.

Light-activation equilibrium of visual rhodopsin is back shifted to preactive MI state by application of hydrostatic pressure. Equilibrium constant for MI-MII transition of rhodopsin in native retinal disk membranes is plotted versus hydrostatic pressure (T = 3°C). Note that relatively large hydrostatic pressures correspond to strong interaction energies of the protein with water. Inset: summary of equilibrium constant at 0.1 MPa (≈0.99 atm) and ΔV showing dependence on lipid or detergent environment of rhodopsin. Data and errors bars are replotted from (77).

An additional important feature is that, when osmotic stress experiments are performed with sufficiently small PEGs (under M_r ≈ 600 Da), a rather different behavior is observed (Fig. 5). At low concentrations these small PEGs actually forward shift the equilibrium toward MII, thereby activating rhodopsin (64). Once a critical concentration is reached, increasing the PEG concentration further resumes the back shifting of the reaction, and the behavior begins to resemble that of the large PEGs. As the size of the PEG increases this critical concentration is reached at lower osmotic pressures. The PEGs over M_r ≈ 1000 Da can be said to reach this critical concentration immediately, yielding the curves shown in Fig. 3.

Figure 5.

Dependence of metarhodopsin equilibrium on osmotic pressure for small osmolytes shows that stabilization of active MII rhodopsin occurs until saturation is reached. The equilibrium constant for the MI-MII transition is plotted versus increasing concentration (osmotic pressure) of polyethylene glycol (PEG) solutes of various molar masses (T = 15°C). Lines with positive slope indicate that the osmolyte shifts the equilibrium toward active MII state; lines with a negative slope indicate shifting of the equilibrium toward the preactive MI state. Inset: depiction of how metarhodopsin equilibrium is forward shifted to active (open) MII state by penetration of small osmolytes into the protein interior. Reprinted with errors bars from (7). Copyright (2022) National Academy of Sciences.

Rhodopsin activation entails proton transfer reactions

Examples where proton reactions are a critical component of biological pathways include cytochrome c oxidase (13), the M2 channel for viral pathogenesis (78), microbial rhodopsins (79), and GPCRs such as visual rhodopsin (80,81,82). The potential for pressure-based techniques to probe these fundamental reactions therefore makes for an exciting possibility. As shown above, for rhodopsin large PEGs back shift the reaction favoring preactive MI, which is illustrated by Fig. 3. This water movement is thought to be due to the PEG osmolytes dehydrating rhodopsin. Conversely, the small PEGs initially shift the equilibrium toward the more active MII, until they reach a critical concentration, beyond which back shifting of the reaction is resumed (Fig. 5). Several explanations are possible for the contrasting behavior of small osmolytes: rhodopsin might be less dehydrated due to their ability to penetrate into rhodopsin, reducing the concentration gradient and osmotic pressure, as observed in other systems (63,83). The lipids could also be dehydrated, which thickens the membrane thereby favoring the MII state (84). Lastly, due to their ability to penetrate into rhodopsin, the small osmolytes could dehydrate small pockets (microdomains) inside the protein, such as Glu¹³⁴ of the E(D)RY sequence motif, which favors formation of the MII state.

Movement of water into cavities is especially relevant for rhodopsin activation due to residues such as Glu¹³⁴ that, when deprotonated, hold rhodopsin in the preactive MI state through ionic interactions with Arg¹³⁵. This ionic lock heavily favors the preactive MI state, which disfavors the open binding pocket of rhodopsin (85). Ionized Glu¹³⁴ is associated with water (86) so that dehydration via small osmolytes may mediate its neutralization, forward shifting rhodopsin toward MII (87). The effect of this ionic lock on the pH titration curves of rhodopsin is illustrated by Fig. 6. Hydrostatic pressure is thought to act primarily on such pockets (73), forcing water into the protein and back shifting the equilibrium toward MI. The effect of hydrostatic pressure on rhodopsin in different systems supports this idea (inset of Fig. 4). In the native retinal disk membranes, the application of hydrostatic pressure shifts the equilibrium toward MI. Yet surprisingly this back shifting is not observed in detergents that displace nearly all of the lipid membrane from rhodopsin. Such detergents remove the pressure dependence of this equilibrium, while also forward shifting the equilibrium toward MII. Both osmotic pressure experiments and dynamics simulations predict MII to be a more hydrated state (5,7), which is consistent with detergent micelles having greater water penetration than lipid membranes. We thus propose that detergents do not seal rhodopsin in the same way as intact lipid membranes, allowing for it to be constitutively hydrated and correspondingly shifted toward the MII conformation. Since water now has greater access to large portions of rhodopsin, it is also able to readily enter and hydrate the small pockets of the protein. With these pockets already hydrated, high pressure is no longer able to force in additional water, and thus the back shifting toward MI is absent under these conditions.

Figure 6.

Protonatable groups of visual rhodopsin control its activation equilibrium. (A) Schematic pH titration curves showing fraction of MII versus pH highlighting protonation states of amino acid residues and the retinal ligand that control rhodopsin activity. Near physiological pH the protonation reaction of Glu¹³⁴ of the conserved E(D)RY sequence motif dominates the equilibrium. At acidic pH, the Schiff base linkage of the retinal cofactor begins to protonate, even in MII, thus reducing the MII fraction. At alkaline pH values the endpoint is due to MII substates that coexist with Glu¹³⁴ fully deprotonated. (B) Experimental pH titration curves from UV-visible spectroscopy show that controlled hydration governs rhodopsin activation in retinal disk membranes (T = 15°C). Applied osmotic stress stemming from large PEG osmolytes removes bulk-like water from rhodopsin, back shifting the pH titration curve and decreasing the pK_A of Glu¹³⁴ favoring the preactive MI state. Conversely small PEGs can penetrate into rhodopsin and dehydrate microdomains, increasing the pK_A of Glu^l34 and forward shifting the reaction by favoring the active MII state. Note also that an effect on the alkaline endpoint is observed: small osmolytes stabilize the open MII conformation even when Glu¹³⁴ is fully deprotonated, increasing the alkaline endpoint, while dehydrating large osmolytes reduce the deprotonated MII population and alkaline endpoint. Dehydration of rhodopsin by large PEG osmolytes reduces the pH range and hydration by small osmolytes extends the pH region over which rhodopsin is active. Adapted incuding errors bars from (7). Copyright (2022) National Academy of Sciences.

To study such movements of water and protonation reactions, molecular dynamics (MD) simulations have been used on account of their ability to visualize the comparatively small ${\mathrm{H}}_2\mathrm{O}$ and ${\mathrm{H}}_3{\mathrm{O}}^{+}$ hydronium ions (13). Such MD computer modeling provides valuable knowledge about the role of water dynamics in governing protein function. As examples of how atomistic insights complement experimental findings, multiscale simulations have revealed key features of the proton-pumping mechanism in cytochrome c oxidase (88). For rhodopsin, the formation of a continuous water pathway upon light activation has been identified (6). The mobility of retinal in the preactive MI state has been related to an influx of water (5). Other simulations have shown this influx of water to interact with several conserved residues in the retinal binding pocket, hinting at the relevance of water to the structural dynamics of GPCRs in drug design (45). The importance of solvent forces—including lipids for membrane proteins—and internal hydration have been highlighted by additional simulations (89). The effects of dehydration on lipid membrane dynamics have been explored with MD simulations, where it was found that dehydration promotes the formation of a more ordered lipid bilayer (90). Atomic-level simulations have also been used extensively to study the role of H-bonding networks in the functioning of microbial rhodopsins, where water was shown to play a critical role in mediating the H-bonds (91). Similar importance has been ascribed to water in mediating H-bonding in GPCRs, where H-bonding networks mediate the propagation of structural changes upon activation (92). Such findings extend beyond just rhodopsin (23), with similar importance of water to H-bonding networks having been found for other proteins including opioid receptors (69), pH-sensing GPCRs (23,93), and aquaporins (94).

Hydrostatic and osmotic pressures probe the dynamics of membrane proteins

Using data from the osmotic and hydrostatic pressure studies of rhodopsin, it becomes possible to draw an energy landscape for its activation (Fig. 7). Coming from Eqs. 2 and 3 it follows that the change in volume for the MI to MII transition can be calculated under conditions of either hydrostatic or osmotic pressure. This calculated ΔV will be the sum of all possible contributions, such as interstitial hydration or collapse of small cavities, solvation of hydrophobic side chains, or an increase in density of the solvation shell (75). Still, the contributions to the volume change from the solvation are likely to be the largest in most cases (77). Within this context, it is appropriate to consider the change in volume in terms of the number of water molecules by the equation:

$$\begin{array}{c}\Delta V°\approx N_w{\overline{V}}_w\text{,}\end{array}$$ (6)

where $N_w$ is the number of water molecules and ${\overline{V}}_w$ is the partial molar volume of water (95). For osmotic pressure (with large osmolytes) the change in volume corresponds to a lower limit of ∼80 water molecules that are involved in rhodopsin activation, while for hydrostatic pressure around 3 water molecules are likely to be involved (7). In addition to the experimental force-based techniques discussed here, the shift to the active MII state of rhodopsin due to such increases in hydration is theoretically supported by MD simulations (5,6,23,45,89). Although hydrostatic pressure acts on fewer water molecules, the applied pressures (∼250 MPa) far exceed the ∼10 MPa used in osmotic pressure experiments. Thus, the total energy applied ($P\Delta V$) remains roughly constant between the two techniques. On the other hand, the energy per water molecule is far greater in hydrostatic experiments. We thus propose that hydrostatic pressure experiments are acting on water molecules that are tightly associated with rhodopsin, such as those hydrogen bonded to specific residues or those in the solvent shell. Conversely, we further propose that osmotic pressure is acting on water molecules that do not form strong interactions with rhodopsin, termed bulk-like water (Fig. 8).

Figure 7.

Side view of the hierarchical multidimensional energy landscape (EL) of rhodopsin, where different protein states (MI, MII, etc.) occupy basins of attraction separated by surmountable energy barriers. The macrostates of MI and MII comprise β-basins, and light activation would correspond to tier-0 fluctuations. Bottom left inset: the energy landscape is rough; within each tier the peak heights and energy of each substate differ. Bottom right inset: collective α-fluctuations are coupled (slaved) to bulk-like water about the protein, while local β-fluctuations are coupled to the solvent shell and structural water. The α- and β-fluctuations arise from the same waters probed by osmotic and hydrostatic pressure, respectively.

Figure 8.

Schematic cartoon of how hydrostatic pressure and osmotic pressure shift the rhodopsin activation equilibrium from the active MII state (right) to the preactive MI state (left) by perturbing structural or bulk-like water. (A) Osmotic pressure removes bulk-like water (blue) surrounding the protein from the active MII state thus dehydrating rhodopsin and back shifting to the preactive MI state. (B) Hydrostatic pressure forces water (red) into the solvent shell or small internal cavities where interactions with key amino acid residues can shift the equilibrium between active MII and the preactive MI state. Influences of both hydrostatic pressure and osmotic pressure are unified by a hierarchical energy landscape mechanism (ELM).

As we have discussed above, the equilibrium fluctuations of the energy landscape have been linked to the movements of specific waters (bottom right inset of Fig. 7) (96). The α-fluctuations are coupled to the bulk-like solvent, and the β-fluctuations are coupled (slaved) to the solvent shell (51). It follows that bulk-like water is not tightly associated with rhodopsin and thus it can be displaced with relatively low energy perturbations. In contrast the solvent shell comprises water that forms relatively strong interactions—to displace it requires correspondingly higher energy perturbations. The similarity of these fluctuations to the techniques of osmotic and hydrostatic pressure is striking. Osmotic pressure is thought to act on relatively large numbers of waters applying only small forces to each individual water molecule, while hydrostatic pressure acts upon small numbers of waters but with a much higher amount of energy per water molecule. We therefore propose application of these two complementary techniques for exploring fluctuations of the energy landscape in the case of proteins and biomembranes in particular.

Compressibility changes suggest greater flexibility of rhodopsin upon light activation

It is noteworthy that, in Figs. 3, 4, and 5, the curvature of the plots can be attributed to the volumetric compressibility, although specific interactions or penetration of rhodopsin by solutes is also possible. The curvature is modeled by the quadratic terms of Eqs. 2 and 3, which allow for calculation of the apparent isothermal compressibility (${\kappa }_T$) of these systems. Because of the vastly different molar masses ($M_{\mathrm{r}}\equiv M$) and volumes of different proteins, it is convenient to use their specific compressibility ($k_T$) for comparison, where:

$$\begin{array}{c}{k}_T=\frac{V{\kappa }_T}{M}=-\frac{1}{M}{\left(\frac{\partial V}{\partial P}\right)}_T.\end{array}$$ (7)

Given that large changes of ${\kappa }_T$ are observed in even closely related biopolymers (97), we assume that relative changes in compressibility are far greater than the relative change in volume, and thus that $\Delta k_T=V\Delta {\kappa }_T/M$. Hydrostatic isothermal compressibilities for proteins are typically on the order of 1–10 × 10⁻⁶ cm³/g bar (98,99), while changes in the compressibilities of proteins undergoing partial unfolding due to hydrostatic pressure are usually on the order of 1–2 × 10⁻⁶ cm³/g bar (100). The value of 6.7 ± 0.8 × 10⁻⁷ cm³/g bar for k_T calculated using our fit of Fig. 4 is in good agreement with these values in the case of perturbation by hydrostatic pressure. Besides probing the changes in protein compressibility and compressibility of the hydrated volume, it is worth considering contributions from the compressibility of lipids as well. The specific compressibilities of lipids is 6 × 10⁻⁵ cm³/g bar (101,102,103), while the specific compressibility of water is 4.3–5.0 × 10⁻⁵ cm³/g bar. To determine whether the observed differences in the rhodopsin activation equilibrium are due to water or lipid effects, we consider the known effects of osmotic and hydrostatic pressure on lipid bilayers. In other work it has been shown that the lipid bilayer is thickened under hydrostatic pressure (104,105,106,107,108,109,110,111,112) and osmotic pressure (112,113), accompanied by a change in spontaneous curvature (112,113). According to the FSM, negative spontaneous curvature and thicker lipid bilayers both favor formation of the active MII state of rhodopsin (38). Since the equilibrium is observed to favor MI when subjected to pressure—the opposite of what is predicted for lipid effects—we attribute the observed change to water effects rather than lipid effects.

Furthermore, one can also define the specific adiabatic compressibilities ($k_S$) of proteins similarly to the isothermal compressibility, except that now entropy is held constant rather than temperature:

$$\begin{array}{c}{k}_S=\frac{V{\kappa }_S}{M}=-\frac{1}{M}{\left(\frac{\partial V}{\partial P}\right)}_S\text{.}\end{array}$$ (8)

Changes in the adiabatic compressibilities for proteins undergoing a partial unfolding due to osmotic stress are typically of a similar magnitude, ranging from 1 to 4 × 10⁻⁶ cm³/g bar (100). On the other hand, for rhodopsin the apparent isothermal compressibilities calculated from osmotic stress data are much larger: 1.5 ± 0.5 × 10⁻⁴ cm³/g bar for small osmolytes such as PEG 200, increasing with greater osmolyte size and reaching a limit of 6.1 ± 0.5 × 10⁻⁴ cm³/g bar around PEG 1000. While isothermal compressibilities will generally be somewhat larger than adiabatic compressibilities, the difference should be on the order of 10⁻⁷ cm³/g bar, i.e., roughly a 10% difference. Given that the calculated apparent compressibilities and values in the literature differ by approximately two orders of magnitude, it is possible for rhodopsin that the observed result is instead due to stabilizing interactions of PEGs with the MII conformation, similar to those with transducin. Indeed, as shown by Record and co-workers, there are previous instances where osmolytes have been observed to have specific interactions with proteins of interest that support this possibility (59,60,61,83).

Available pressure data for rhodopsin therefore suggest that the transition from MI to MII entails both a positive change in volume and apparent compressibility. We thus propose that active MII resembles a partially unfolded state, somewhat akin to intrinsically disordered proteins, and that this disorder facilitates interactions with the transducin G-protein by conformational selection. According to our interpretation, upon light activation rhodopsin adopts a more flexible MII conformation that facilitates catalytic activation of the effector G-protein transducin. By combining osmotic pressure (7,23) with hydrostatic pressure data (77,114,115), we introduce the idea of a hierarchical protein softness (65). Here the local β-compressibility is defined as the change in solvent shell or structural water volume with respect to applied pressure, while the α-compressibility is the corresponding change in bulk-like solvent volume with pressure. Osmotic stress affects the collective α-compressibility by withdrawing water from the more hydrated MII, leading to a reduction of the hydrated volume (Fig. 8 A) by removing weakly bound water. Analogously, the larger hydrostatic pressures force water into the protein interior favoring the preactive MI state, which affects the local β-compressibility (Fig. 8 B). Most notably, the application of either osmotic pressure (i.e., stress) (Fig. 8 A) or hydrostatic pressure (Fig. 8 B) yields a back shifting of the metarhodopsin equilibrium to the preactive MI state. In the forward reaction, the hydrated volume and the collective α-compressibility are both increased. (Previously we suggested that the β-compressibility was greater in MI because we had not reanalyzed the hydrostatic pressure data (77).) The conclusion of greater α- and β-compensability of the active MII state is likely significant with regard to binding and release of the transducin effector protein in the catalytic G-protein activation cycle (23). Hence, our current working model is that the active MII state is more hydrated due to an increase in weakly bound bulk-like water, together with an excess of void or cavity volumes versus preactive MI due to release of more tightly bound structural water molecules.

While here we have primarily been discussing the GPCR rhodopsin, it is important to remember that, given the highly conserved nature of GPCR activation, insights for rhodopsin activation are likely transferrable to a broad range of other GPCRs with physiological significance (116). The relevance of void spaces in proteins to physiological function—such as those involving interstitial hydration by hydrostatic pressure—has recently been highlighted for the β₁-adrenergic receptor (70). Furthermore, it has been proposed by Gruner et al. that dynamics of the lipid membrane soft matter may be responsible for the effect of general anesthesia (110). The energy landscape model has been proposed by Frauenfelder and co-workers in general terms as a framework for interpreting protein dynamics, with the higher tiers of the landscape being described and explored primarily through influences of temperature (70,117,118,119). Despite this, the lower tiers of fluctuations are not often considered, and techniques based on pressure that explore them are notably absent. In our work, we expand upon the energy landscape model by introducing the combination of force-based techniques that have the potential to explore the full energy landscape of proteins.

Implications for membrane structural biology and biophysics

The energy landscape model of protein dynamics provides a framework for interpreting biomembrane function, where force-based measurements appear as promising tools for exploring the hierarchy of protein states and substates. Solvent effects and protein dynamics are current limiting factors in our understanding of membrane protein functions. For the paradigm of rhodopsin, we propose that hydrostatic pressure acts on small numbers of tightly bound structural waters to back shift the MI to MII equilibrium, while osmotic pressure affects more weakly interacting waters beyond the solvent shell. In either case, such force-based measurements imply a greater hydration of the active rhodopsin state. Experimental validation for the importance of hydration-mediated proton reactions to the hydrostatic and osmotic pressure dependence of protein conformational equilibria is thus an important next step in developing these force-based techniques. More generally, exploring the solvation energy landscape is an exciting new direction toward the goal of more completely understanding how solvation dynamics are coupled (slaved) to protein activity. Due to the prevalence of rhodopsin-like GPCRs in critical biological signaling pathways, they are attractive targets for drug development, where incomplete knowledge of protonation reactions makes such progress challenging. Further research is needed to explore and develop osmotic and hydrostatic pressure techniques to selectively probe solvation-coupled energy landscapes and their role in membrane protein function.

Acknowledgments

We dedicate this paper to the memory of Hans Frauenfelder in recognition of his inspiring contributions to biophysics. This work was supported by the US National Science Foundation (CHE 1904125 and MCB 1817862 to M.F.B.) and US National Institutes of Health (EY026041 to M.F.B.).

Author contributions

M.F.B. developed the initial concept and Z.T.B. carried out data interpretation and analysis. Z.T.B. and M.F.B. wrote the article.

Declaration of interests

The authors declare no competing interests.

Editor: Ana-Nicoleta Bondar.