New approaches towards the understanding of integral membrane proteins: A structural perspective on G protein‐coupled receptors

Abstract

Three‐dimensional structure determination of integral membrane proteins has advanced in unprecedented detail our understanding of mechanistic events of how ion channels, transporters, receptors, and enzymes function. This exciting progress required a tremendous amount of methods development, as exemplified here with G protein‐coupled receptors (GPCRs): Optimizing the production of GPCRs in recombinant hosts; increasing the probability of crystal formation using high‐affinity ligands, nanobodies, and minimal G proteins for co‐crystallization, thus stabilizing receptors into one conformation; using the T4 lysozyme technology and other fusion partners to promote crystal contacts; advancing crystallization methods including the development of novel detergents, and miniaturization and automation of the lipidic cubic phase crystallization method; the concept of conformational thermostabilization of GPCRs; and developing microfocus X‐ray synchrotron technologies to analyze small GPCR crystals. However, despite immense progress to explain how GPCRs function, many receptors pose intractable hurdles to structure determination at this time. Three emerging methods, serial femtosecond crystallography, micro electron diffraction, and single particle electron cryo‐microscopy, hold promise to overcome current limitations in structural membrane biology.

Introduction

Determining three‐dimensional structures of integral membrane proteins was largely deemed impossible until the 1970s. In 1975, a seminal paper by Richard Henderson and Nigel Unwin1 presented the first structural information for an integral membrane protein, a 7 Å resolution structural model of bacteriorhodopsin, a light‐driven proton pump from the archaebacterium Halobacterium salinarium. This structure was obtained from two‐dimensional crystals imaged by room temperature electron microscopy, but it took another 15 years of methods development before an atomic model of bacteriorhodopsin was published in 1990, which was the first structure of any protein determined by electron microscopy.2 In the intervening period, the first structure of an integral membrane protein, a bacterial photosynthetic reaction center, was determined by X‐ray crystallography,3 and in 1988, Hartmut Michel, Johann Deisenhofer, and Robert Huber were awarded the Nobel Prize in Chemistry for this work. These accomplishments launched the era of membrane protein structural biology.

During the last three decades, breathtaking progress has been made in elucidating how membrane proteins function through determining their three‐dimensional structures and utilizing information from nonstructural experimental approaches. For this to happen, strategies, reagents, and tools for recombinant expression, purification, stabilization, crystallization and data collection needed to be developed4 because many membrane protein targets, especially eukaryotic membrane proteins, are not abundant naturally and are unstable in detergent solution owing to their conformational flexibility. This review summarizes concepts that have advanced the field of membrane protein structure determination, using GPCRs as an example, and discusses emerging avenues of investigation and new technological developments with the potential to overcome current limitations in membrane protein structural biology.

G protein‐coupled receptors

GPCRs are integral membrane proteins that reside in the plasma membrane of eukaryotic cells. They are central to transmitting signals from the extracellular milieu to the inside of the cell. Ligand binding on the extracellular surface of GPCRs leads to conformational changes within the receptor, triggering the activation of G proteins and the interaction with kinases and arrestin molecules in the cytosol.5 With more than 800 members in the human genome, GPCRs constitute the largest protein family involved in signal transduction, and are major targets for drug development.

GPCRs exist in an equilibrium between inactive, nonsignaling states, and active states that can interact with signaling partners within the cell. In contrast to rhodopsin, many GPCRs exhibit basal activity that has been correlated with structural instability.6 Ligands range from inverse agonists, which inhibit basal activity, to full agonists, which maximally activate the receptor. The conformational dynamics of GPCRs provide the basis for the complexity and fine‐tuning of hormone signaling, but also render GPCRs difficult to crystallize because of their conformational heterogeneity and instability.

In the year 2000, the first crystal structure was determined of a GPCR, the visual pigment rhodopsin.7 The second crystal structure, the β₂‐adrenergic receptor (β₂AR), was published only in 2007.8, 9, 10 10 years later, crystal structures of 38 unique GPCRs have been reported.11 Extraordinary progress towards understanding the activation mechanism of GPCRs was achieved by the crystallization of receptors in an active state, coupled to a cytoplasmic heterotrimeric G protein,12 an engineered mini G protein,13 G protein mimicking nanobodies (Nbs),14, 15, 16, 17 or arrestin18 (Fig. 1). This exciting progress required a tremendous amount of methods development, with the most notable achievements described below. The foundation of today's successes was the decades of research by groups throughout the world who were convinced that structures of GPCRs will be solved through sufficient effort and the development of novel approaches. Indeed, it took two decades of hard work after the cloning of the respective cDNAs19, 20, 21, 22, 23 before the structures of the β₂AR,8, 9, 10 the β₁‐adrenergic receptor (β₁AR),24 the adenosine A_2a receptor (A_2aR),25, 26, 27, 28 and the neurotensin receptor (NTSR1)29, 30, 31, 32 were determined.

Figure 1.

Milestones of GPCR structure determination: PDB codes are given in parentheses. (A) Rhodopsin (1F88), (B) β₂‐adrenergic receptor (2RH1), (C) β₂AR with nanobody (3P0G), (D) β₂AR with heterotrimeric G protein (3SN6), (E) Rhodopsin with arrestin (4ZWJ), (F) Adenosine A_2a receptor with mini G protein (5G53), (G) Calcitonin receptor (CTR) with heterotrimeric G protein (5UZ7). Structures shown in panels A–F were solved by X‐ray crystallography, whereas the structure shown in panel G was solved by single particle cryo‐EM. The T4 lysozyme fusion partner facilitating crystal formation is indicated as T4L. Individual protein chains of receptor complexes are labeled and shown in different colors. Publication years are given at bottom of figure.

What are the challenges in solving membrane protein structures?

This section addresses key considerations for membrane protein crystallization and serves as a guide to navigate through the different aspects discussed in this review. Hurdles for obtaining high‐resolution structures may present themselves at virtually every step, from recombinant expression to crystal formation and diffraction data acquisition.

Despite the miniaturization of the actual crystallization process by robotics, the target GPCR needs to be produced in yeast, insect cells, or mammalian cells at milligram scale. These production methods are preferred because only very few GPCRs can be expressed in functional form in bacteria.29, 33, 34, 35 Owing to space limitations, the reader is referred to reviews that cover the topic of recombinant expression of membrane proteins in detail [e.g., Refs. 36, 37, 38].

GPCRs are highly dynamic molecules and their inherent flexibility is the underlying cause for their often‐observed biochemical instability. Successful structure determination requires that detergent‐solubilized GPCRs are maintained in a biologically relevant form, as aggregated or denatured protein will not form crystals. Bovine rhodopsin was the first GPCR to be crystallized,7, 39 reflecting its high stability in many detergents as long as the receptor was kept in its inactive, dark state. However, most other GPCRs are labile in the presence of detergent40 owing to their highly dynamic properties. Remedies to stabilize receptors include addition of a high‐affinity ligand to stabilize a particular signaling conformation thus reducing conformational flexibility and increasing receptor stability;41 conformational thermostabilization of GPCRs42 as a means to lock the receptor in a particular ligand‐dependent state; or complexation with signaling mimetics or partners such as nanobodies,43 mini G proteins44, 45 or G protein12 to improve stability. Continued efforts to develop novel mild detergents such as bicyclic alkylmaltosides,46 facial amphiphiles,47 and maltose‐neopentylglycols48, 49 have been instrumental in receptor stabilization and concomitant formation of well‐ordered crystals.12

Protein crystals are materials in which molecules are arranged in a highly ordered microscopic structure characterized by symmetry. Thus, any flexible entity of a receptor, such as poorly structured termini and loops, or dynamic features of the transmembrane core, counteract the growth of highly diffracting crystals. Unstructured receptor regions have been removed by protein engineering in many receptor constructs optimized for crystallization. Replacement of loops by crystallization modules such as T4 lysozyme (T4L) has restricted the movement of adjacent transmembrane helices.10 Conformational thermostabilization42 and co‐crystallization with conformation‐specific antibodies,9, 50, 51 nanobodies,14, 15, 16, 17 or G protein12, 13 have resulted in more homogenous protein preparations, and diffracting crystals.

Crystallogenesis requires that protein molecules engage in crystal lattice contacts. As for water‐soluble proteins, polar interactions contribute to the stabilization of crystals of membrane proteins.52 The polar surfaces of many GPCRs are small, and using stabilizing, large detergents such as dodecyl‐maltoside, results in occlusion of the hydrophilic receptor regions essential to form crystal contacts. Small detergents such as octyl‐glucoside are in theory better suited for crystallization52 but are often highly denaturing precluding their use for most GPCR targets. Two approaches for promoting receptor crystal formation have evolved: Increasing the hydrophilic area by engineered fusion proteins such as T4L,10, 53 antibodies,9, 50, 51 or nanobodies,43 so that mild detergents with a large micelle can be used; or thermostabilizing receptors to make them withstand harsh, short‐chain detergents.42

Fusion partners and antibody fragments

Obtaining high‐resolution structures of GPCRs other than rhodopsin has been challenging because of their inherent structural instability. In addition, many GPCRs contain relatively small hydrophilic loops, and the scarcity of hydrophilic regions available to form potential crystal contacts is a major impediment to obtaining diffraction quality GPCR crystals. Replacement of inner loop 3 (IL3) with T4L, a well‐folded protein that adds a polar surface and restricts the movement of the adjacent transmembrane helices, proved critical for obtaining the well‐ordered crystals of the β₂AR.10 An expanded set of fusion partners for crystallizing GPCRs has subsequently been developed, most notably a thermostabilized apocytochrome b₅₆₂ (BRIL),53 rubredoxin,53 and the catalytic domain of Pyrococcus abysii glycogen synthase (PGS).54 These fusion partners have been used in various ways to replace IL3 (β₂AR‐T4L,10 chemokine receptor CCR5‐rubredoxin,55 orexin receptor OX2‐PGS54), or IL2 (metabotropic glutamate receptor mGlu5‐T4L),56 or positioned at the receptor N‐terminus (T4L‐β₂AR,12, 57 BRIL‐nociceptin/orphanin FQ receptor58).

Antibody fragment‐mediated crystallization59 has also been successful through reducing the conformational dynamics of the receptor and extending the hydrophilic surface area available to make crystal contacts. Receptors crystallized in the presence of mouse monoclonal antibody fragments include β₂AR bound to Fab5, which recognizes a three‐dimensional epitope on the intracellular side of the β₂AR,9, 50 and A_2aR bound to Fab2838, which binds to the intracellular surface of A_2AR where it acts as an allosteric antagonist.51

Nanobodies and mini G proteins

The formation of highly ordered 3D crystals necessitates that the receptor is preferentially in one particular conformation. This is achieved by the use of high‐affinity inverse agonist or full agonist ligands, and in the case of active‐state receptors, by complexation with a signaling partner or mimetic. To date, only one structure of a GPCR‐arrestin complex (rhodopsin‐arrestin18) and only one structure of a GPCR‐G protein complex (β₂AR bound to the adenylate cyclase stimulating G protein G_s 12) have been reported, providing the first insight into the arrangement of those complexes. In addition, receptors have been crystallized in an active state coupled to peptide fragments of the Gα subunit60, 61 or arrestin.62 Crystallizing GPCR‐arrestin and GPCR‐G protein complexes remains very challenging, but the development of G protein mimicking nanobodies63 and most recently minimal G proteins44 facilitated crystallization of several active‐state receptors (β₂AR‐Nb80,16 β₂AR‐Nb6B9,17 muscarinic receptor M₂R‐Nb9–8,15 μ‐opioid receptor‐Nb39,14 A_2AR‐mini G_s 13).

Nanobodies43 are the recombinant variable domains of camelid heavy‐chain antibodies with full antigen‐binding capability, and are selected from peripheral blood lymphocytes of an immunized llama through combinatorial methods such as phage display, yeast display, or ribosome display.43, 63 Nanobodies have evolved as alternatives to conventional antibodies because of their elongated shape and convex antigen binding site, allowing access to cavities on proteins that are difficult to access by conventional antibodies. Nanobodies have been instrumental in obtaining structures of conformationally stabilized active‐state GPCRs.

Mini G proteins are engineered, small soluble proteins composed of a truncated form of the GTPase domain of the G protein, including point mutations to stabilize the protein in the absence of Gβγ and in the presence of detergent. Three truncations eliminate the switch III region, part of the N terminus, and the α‐helical domain of the Gα protein.44 Mini G proteins efficiently couple to GPCRs in the absence of Gβγ subunits.44 A growing number of mini G proteins provide now the tools to study GPCRs in their active conformation and facilitate complex formation of GPCRs for crystallization experiments.45

In surfo and in meso crystallization

Common and successful approaches to membrane protein crystallization are methods that make use of detergent‐solubilized membrane protein preparations in crystallization trials (in surfo: vapor diffusion, dialysis, microbatch) [see Ref. 64]. While protein–protein interactions dominate the crystallization of soluble proteins, both detergent–detergent and protein–detergent interactions are additional considerations for the membrane protein crystallization, and choosing the right detergent has been key to successful membrane protein crystallization efforts. GPCRs (except rhodopsin) show poor stability in detergent solutions, especially in harsh short‐chain detergents, preferred for crystal contacts to occur. Conformational thermostabilization is one possible approach to make the protein more tolerant to short‐chain detergents, and this approach has been successfully used to determine GPCR structures in surfo.29, 42, 65 Alternatively, receptors have been prepared in mild, stabilizing detergents such as dodecylmaltoside or neopentylglycol detergents. However, the hydrophilic regions of the membrane protein, essential to form crystal contacts, are then often occluded by the detergent micelle, preventing crystal growth. Extending the hydrophilic surface area by using a monoclonal F_ab antibody fragment51 or replacing IL3 with BRIL66 allowed the structures of antagonist‐bound A_2AR to be determined from crystals grown by vapor diffusion and using large‐micelle detergent.

Lipidic cubic phase (LCP) is a membrane‐mimetic matrix suitable for stabilization and crystallization of membrane proteins in a lipidic environment (in meso crystallogenesis).64 This methodology was introduced by Ehud Landau and Jürg Rosenbusch back in 1996,67 and optimized for miniaturization and automation by Martin Caffrey and Vadim Cherezov.64 The success of using LCP for crystallization can be attributed to the stabilizing membrane‐like environment of LCP and type I crystal packing, where crystal contacts are formed between both polar and nonpolar parts of the protein, contributing to better order and diffraction. LCP crystallization accommodates membrane proteins prepared in mild long‐chain detergents. Almost all in meso GPCR structures have utilized receptors modified by fusion partners, although no fusion partner was used for the chemokine CCR9 receptor68 and the high‐resolution structure of β₁AR.69 LCP crystallization is not limited to GPCRs, but structures of other membrane proteins, both α‐helical and β‐barrel, including single and multisubunit proteins, have been reported [see Ref. 70]. Crystals grown in meso are often small, and the development of microfocus X‐ray synchrotron technologies such as the minibeam71 combined with rasterization capabilities72 and the development of fast detector technologies have been indispensable for data collection.

Two alternative crystallization techniques, in which protein is crystallized from a lipid membrane environment, are the bicelle crystallization method9, 73 and crystallization by vesicle fusion.74 These methods have had some successes with a variety of membrane proteins.

Conformational thermostabilization

GPCRs are highly dynamic molecules with low thermal barriers between the inactive and active signaling states, displaying in many cases basal activity. Furthermore, GPCRs show poor stability in detergent solution. Therefore, conformational flexibility and instability, once extracted from the membrane, impede the growth of well‐ordered crystals.42 Two strategies have been used to address these shortcomings: Use of high‐affinity ligands, with slow off‐rates to allow full occupancy during the crystallization process, lock the wild‐type receptor into a single conformation, which increases the receptor stability in detergent solution and therefore the probability of crystal formation. Alternatively, improving the thermostability of the GPCR itself makes the protein more tolerant to short‐chain detergents, preferred in vapor diffusion crystallization experiments. The latter approach allows co‐crystallization with low‐affinity ligands, because the enhanced stability is encoded within the receptor protein itself.42 To isolate thermostabilizing mutations in a receptor, mutations are introduced throughout the receptors by either systematic scanning mutagenesis [see Refs. 65, 75] or random mutagenesis as part of an evolution‐based strategy.76, 77 In systematic scanning mutagenesis, the most stabilizing single mutations are then combined to generate a mutant receptor with an augmented melting temperature. The combination of mutagenesis with a specific selection strategy was used to stabilize agonist78, 79 and antagonist65, 78 conformations of GPCRs.

Crystal structures of agonist‐bound GPCRs have relied on the use of either exceptionally high‐affinity agonists or receptor stabilization by mutagenesis. However, many endogenous ligands bind with low affinity and are structurally not tractable in complex with a receptor. Using directed evolution, Ring and colleagues17 used yeast surface display together with a conformationally specific selection strategy to engineer a high‐affinity nanobody that minimizes receptor heterogeneity and stabilizes the active state of the β₂AR to determine the crystal structure of the activated receptor bound to the endogenous agonist adrenaline.

In situ data collection

Membrane protein crystals are often tiny and fragile, and thus pose challenges for handling such as mounting the crystals or soaking in cryoprotectants. Recently published methods on in situ room temperature data collection have made a major contribution to address these issues. Axford et al.80 mounted an entire 96‐well sitting‐drop crystallization plate at a synchrotron beam line and successfully collected a data set from many membrane protein crystals. Huang and colleagues81 implemented an in situ approach with in meso crystallization, termed the “in meso in situ serial crystallography” (IMISX) method. Both methodologies pave new avenues to data collection from crystals that are too fragile and too small to mount without losing diffraction quality through mechanical shear‐induced stress.

Techniques that complement structural work

The focus of this review is on approaches yielding high resolution structural information. By nature, crystal structures provide static rather than dynamic views and typically capture the lowest energy state within an ensemble of conformations that may differ from other signaling conformations of membrane‐inserted receptors, not observed in crystal structures. Therefore, techniques that complement structural work are essential to understand how proteins function. For example, molecular dynamics simulations sample interactions of a membrane protein with lipid and partner proteins, and their internal dynamics;30, 82, 83 single molecule spectroscopy and super‐resolution imaging allows the study of macromolecular function;84 fluorescence spectroscopy probes conformational changes within a receptor,85, 86 and native mass spectrometry has been used to characterize noncovalent protein–protein and protein–ligand complexes and lipid interactions.87 Nuclear magnetic resonance (NMR) techniques and site‐directed electron paramagnetic resonance (EPR) spectroscopy88, 89, 90 are powerful tools for examining intramolecular dynamics and distance distributions between domains of a protein.91

Nuclear magnetic resonance (NMR)

Two main methods have been employed to study the structure and dynamics of membrane proteins, namely solution NMR and solid‐state NMR. Solution NMR methods require that the sample undergoes fast rotational diffusion and this therefore poses limitations on the size of the target membrane protein and its preparation (detergent‐solubilized and lipid micelles, small bicelles, nanodiscs).92 Solid‐state NMR methods have principally no size limitations because they are used on static samples, or samples rotated at high speed, and thus allow membrane protein samples to be prepared in lipid bilayers mimicking the protein's natural environment. However, solid‐state NMR methods still meet technical barriers, because solid‐state NMR spectra are complex, leading to spectral crowding of uniformly labeled samples of large proteins, restricting the size of structures that can be determined.92

Solving the complete structure of a membrane protein has been very challenging and has been impaired by substantial difficulties of sample preparation of sufficient quantities of isotopically 15N‐ and 13C‐labeled protein, limited thermal stability, and sample heterogeneity. Nevertheless, the backbone assignment of the archaeal seven‐helix transmembrane receptor sensory rhodopsin II93 has been accomplished by solution NMR. Furthermore, the backbone structure of a phospholipid bilayer embedded GPCR, the chemokine receptor CXCR1, has been reported by using solid‐state NMR techniques.94 For this, CXCR1 was expressed and isotopically labeled in E. coli as inclusion bodies, purified, and refolded into DMPC proteoliposomes.

In contrast, selective labeling of residues with 19F, 13C, or 15N has been used to provide evidence for ligand‐ and G protein‐dependent conformational states and energy landscapes of GPCRs, and their local dynamics by solution NMR methods.91 Chemical conjugation with isotope‐labeled compounds, either alkylation of cysteine residues or dimethylation of lysine residues, has been used to gather site‐specific information and to avoid presently intractable spectral overlap of a uniformly labeled sample. Chemical labeling is limited by solvent accessibility and the efficiency of the chemical reactions. Despite this limitation, chemical labeling has the advantage that highly sensitive nuclei, such as 19F, can be incorporated, thus leading to excellent signal/noise ratios. Selective isotopic labeling during protein biosynthesis accesses any labeling sites, even those in the hydrophobic environment of the membrane protein. Site‐specific labeling is an excellent tool for fingerprinting the range of potential states of a receptor, but a major challenge remains of relating these observations to the structural changes that give rise to these phenomena.

19F NMR probes have been introduced into detergent‐purified receptors91, 95, 96 at naturally occurring cysteine residues or at a selected position mutated to a cysteine96 and provide exquisite sensitivity to solvent exposure or side chain packing. To monitor a unique receptor environment associated with specific conformational changes requires use of minimal cysteine receptor versions,91 which have to be thoroughly tested for ligand binding and G protein activation to assure integrity of the mutated receptor. Specific labeling of methyl groups, such as 13CH₃ε‐methionines, provide ideal NMR probes for high molecular weight proteins because the length and flexibility of the methionine side chain compensates for the slow tumbling, improving the resolution and sensitivity of the spectra. Methionine residues are dispersed throughout the receptor and many are found in the transmembrane region, not accessible for study by fluorine NMR, which requires the addition of small‐molecule probes to surface‐exposed cysteines.97 To study specific receptor regions, methionines that face the exterior of the receptor may be mutated thus simplifying the spectrum and eliminating signals from 13CH₃ε‐methionine residues that do not undergo structural changes when comparing active and inactive structures.97 The labeling of methionine residues is achieved by growing for example insect cells in methionine‐deficient media in the presence of 13C methyl‐labeled methionine. Alternatively, 13C labels can be introduced into detergent‐purified GPCRs by reductive methylation of, for example lysine residues, with13C‐formaldehyde.98 The above NMR methods have provided data of functional relevance on selected side‐chains. A recent investigation by Isogai and colleagues99 studied receptor motions followed at backbone sites by selective labeling with 15N‐valine, accomplished by growing insect cells in medium devoid of unlabeled valine in the presence of 15N‐valine.99

In elegant work, Smith and colleagues100 have characterized the activation triggers of the visual pigment rhodopsin using solid‐state NMR approaches. The opsin is produced in HEK293S cells and reconstituted with retinal. The information is obtained through measurements of chemical shift and dipolar couplings. This does not yield complete 3D protein structural information because of the expense of uniform 13C‐, 15N‐labeling of proteins using eukaryotic cell lines. Rather, selective labeling of specific amino acid types or different ligands allows the structural features of the ligand or protein of interest to be studied in exquisite detail. For this, stable tetracycline‐inducible suspension HEK293S cell lines containing the opsin gene and its mutants were used to express rhodopsin101 in medium supplemented with specific 13C‐labeled amino acids.

The conformational and dynamic heterogeneity observed in these and other recent NMR studies provides new insights into membrane‐protein function that would have been impossible to obtain through crystallography alone.

New Approaches

Despite tremendous progress to elucidate the structural biology of GPCRs, many receptors pose unsurmountable obstacles to structure determination owing to a multitude of reasons such as very low expression levels, high instability even in mild detergents, extensive post‐translational modifications, formation of microcrystals not suitable for use with current synchrotron mini X‐ray beams, or resistance to crystallization. Three emerging methods, serial femtosecond crystallography, micro electron diffraction, and single particle electron cryo‐microscopy, hold promise in addressing some of these limitations. The first two approaches relate to micro‐crystals, whereas the latter does not require crystals at all.

Serial femtosecond crystallography (SFX)

SFX102 is an emerging technique that relies on extremely bright and ultrashort pulses produced by X‐ray free‐electron lasers (XFELs), permitting the collection of high‐resolution diffraction intensities from very small crystals at room temperature with minimal radiation damage, using the principle of “diffraction before destruction”.103 A single diffraction pattern is collected before each crystal is destroyed by the powerful XFEL pulse. To obtain a single data set, diffraction patterns from thousands of randomly oriented crystals need to be collected. The crystals are therefore delivered into the beam in a continuous hydrated stream, and diffraction images are recorded by a detector operating at the XFEL pulse repetition rate.103 Technical improvements related to the delivery, detector technology, data acquisition, and analysis104 have transformed macromolecular crystallography. It is now possible to determine structures using XFEL using less than a milligram of protein.105 In 2011, a low‐resolution electron density map from crystals of photosystem I demonstrated proof‐of‐principle103 and now structures have been determined at sub‐2 Å resolution.106 Several structures of GPCRs at high resolution have also been determined.18, 107, 108, 109, 110, 111 Most excitingly, time‐resolved serial femtosecond crystallography at an X‐ray free electron laser has allowed the visualization of conformational changes in bacteriorhodopsin from nanoseconds to milliseconds following photoactivation112 and the light‐induced structural changes for oxygen evolution by photosystem II.113 However, access to XFELs for SFX is currently very limited and not a routine approach for most membrane protein crystallographers.

Micro electron diffraction (MicroED)

MicroED114 is a newly developed technique that records electron diffraction patterns from very small crystals in the electron microscope. Extremely tiny 3D crystals are required as electrons cannot penetrate thick samples. Samples are vitrified, as hydration is critical. The specimen is rotated and multiple diffraction patters are taken from a single crystal, resulting in a series of diffraction patterns that can be indexed and processed with available X‐ray crystallography software. The continuous rotation of the sample requires direct electron detectors. The extremely low electron dose rate results in greatly reduced radiation damage of biological samples that are dose sensitive, but this still produces quality diffraction patterns of high resolution. Proof of principle of the MicroED approach was achieved using lysozyme, yielding a structure to 2.9 Å resolution from 3D microcrystals in cryogenic conditions.115 Subsequently, several protein structures have been solved using improved techniques,114 but the structure of a membrane protein by MicroED is yet to be determined.

Single particle electron cryo‐microscopy (Cryo‐EM)

To date, the majority of membrane protein structures has been determined by X‐ray crystallography and by electron crystallography. These approaches necessitate the availability of well‐ordered, three‐dimensional, or two‐dimensional crystals that are a major road block for many membrane protein targets.116 In contrast, single particle cryo‐EM sidesteps the requirement of well‐ordered crystals for structure determination, thus providing an attractive alternative to crystallography. Major technological advances in the last few years have improved detection (direct electron detection camera), classification of images of heterogeneous sample conformations, correction for beam‐induced motion blurring of images, and dose‐fractionation to use the best subframes for data processing.117, 118 This has resulted in the ‘resolution revolution’119 and an increasing rate of high‐resolution structures deposited in the EM Database.120 Indeed, single particle cryo‐EM has already been used extensively to characterize soluble proteins in multiple conformations at high resolution121 and is being used increasingly on membrane proteins.122, 123, 124, 125, 126, 127, 128 The recent structure determination of hemoglobin (molecular weight 64 kDa) at 3.2 Å resolution is a milestone in cryo‐EM,129 which has had an immediate impact in the GPCR field; the identical methodology was recently used to determine the structure of the calcitonin receptor coupled to heterotrimeric G _s.130 Further structures of GPCRs in complex with other binding partners will no doubt follow swiftly, which will result in a revolution in our understanding of the active states of GPCRs.

Conclusions

Membrane protein structure determination has advanced beyond all expectations our insight into mechanistic aspects of how they function. New arising concepts and technological developments hold promise to overcome some of the remaining experimental limitations, propelling us into an exciting time, where structures of integral membrane proteins are no longer regarded as insurmountable challenges.