Visualizing lipid membrane structure with cryo-EM: Past, present, and future

Abstract

The development of electron cryomicroscopy (cryo-EM) has evolved immensely in the last several decades and is now well-established in the analysis of protein structure both in isolation and in their cellular context. This review focuses on the history and application of cryo-EM to the analysis of membrane architecture. Parallels between the levels of organization of protein structure are useful in organizing the discussion of the unique parameters that influence the collective behavior that influences membrane structure and function. Importantly, the timescales of lipid motion in bilayers with respect to the timescales of sample vitrification is discussed and reveals what types of membrane structure can be reliably extracted in cryo-EM images of vitrified samples. Appreciating these limitations, a review of the application of cryo-EM to examine the lateral organization of ordered and disordered domains in reconstituted and biologically derived membranes is provided. Finally, a brief outlook for further development and application of cryo-EM to the analysis of membrane architecture is provided.

Introduction

Stmcture-function relationships exist at all levels of biological organization, from individual molecules to ecosystems. A key level in this structural hierarchy is the supramolecular organization of the four primary molecular building blocks of life: amino acids, nucleotides, sugars, and lipids. The 20th century saw major advances in our understanding of the function of proteins and nucleic acids owing to the development of powerful X-ray and neutron diffraction techniques capable of resolving structure at the atomic level (1). More recently, advances in cryogenic electron microscopy (cryo-EM) combined with ever-increasing computational resources have super-charged the structural revolution, yielding new insights into classes of proteins that are difficult to crystallize and thus recalcitrant to diffraction-based methods (2,3). It is also now possible to cryo-preserve macromolecular complexes and even intact cells, further leading the field into high-resolution analysis of macromolecular complexes as they exist in the native cellular environment (2–4).

Unique among biological molecules, lipids self-assemble mainly through weak, non-covalent interactions, forming a staggering variety of architectures including droplets, micelles, lamellar (monolayer or bilayer), and non-lamellar structures. Of primary importance in biology are the lipid bilayers that form the outer barrier (i.e., the plasma membrane) of all living cells as well as the enclosing membranes of internal organelles in eukaryotes. While the structural biology revolution has focused mainly on proteins, it has become increasingly clear that a detailed understanding of the structure and organization of lipid membranes is no less important for elucidating their function. As with proteins, lipid bilayers exhibit a hierarchy of organization: a) different biological membranes have characteristic lipid compositions (5); b) these compositions can exist in different lamellar phases (6); and c) domains of different phases can coexist within the same membrane (7,8), a useful analogy can be drawn between these three levels of membrane organization and the primary, secondary, and tertiary structure of proteins.

While much is now known about the primary and secondary structure of biological membranes through advanced lipidomics (9) and biophysical techniques (especially NMR (10) and scattering (11,12)), significant gaps remain in our knowledge of tertiary structure, the most prominent example of which are lipid rafts (13). Rafts arise from non-ideal interactions between different classes of lipids found in the plasma membrane that, if sufficiently strong and unfavorable, result in lipid segregation and a laterally heterogeneous membrane (14). Enriched in sphingomyelin and cholesterol, rafts are thicker, more rigid, and less compressible than the disordered sea of unsaturated lipids that surround them (13). Their functionality in part derives from an ability to spatially organize membrane proteins based on a preference for raft vs. non-raft environments (15). In resting cells, rafts are also small, with consensus estimates converging at tens of nanometers (16). Their structure has thus been difficult to characterize with conventional optical microscopy (17).

As a direct imaging technique, cryo-EM holds great promise for answering questions about plasma membrane tertiary structure—that is, the size, shape, and connectivity of membrane nanodomains—that will likely lead to novel functional insights. In this review, we focus on the historical development of cryo-EM for obtaining information about membrane structure with a focus on synthetic liposomes that are models for cell membranes.

Lipid bilayer phases

Lipid bilayers are polymorphic, undergoing dramatic structural transitions as a function of temperature, pressure, hydration, and lipid composition; here, we focus on the structure of the biologically relevant bilayer phases. Membranes composed of a single lipid type have a characteristic main transition temperature T_M. As shown in Fig. 1a, under typical biological conditions of full hydration and atmospheric pressure, bilayers are in a disordered fluid state above T_M and an ordered, solid-like state at temperatures well below T_M. The disordered fluid phase (termed L_α, or L_d when the bilayer is composed of a mixture of lipids) is characterized by fast lateral lipid diffusion and highly disordered acyl chains (rapid trans-gauche isomerization). In contrast, lipids in a gel phase bilayer diffuse much more slowly, and their hydrocarbon chains are in a highly ordered, all-trans configuration and tilted with respect to the bilayer normal (L_β′ phase).

Figure 1. Lipid bilayer phases and phase transitions.

a, bilayers composed of PC lipids are in a gel state (L_β′) well below their main transition temperature and a fluid state (L_α) above their main transition temperature. Addition of cholesterol converts both gel and fluid bilayers to a liquid-ordered phase (L_o). b, bilayers can undergo structural changes including phase changes during vitrification due to rapid gauche-trans isomerization. The extent of vitrification-induced chain ordering depends on the types of lipids in the bilayer (i.e., saturated vs. unsaturated, with or without cholesterol) as indicated. c, correlation times of various lipid motions relevant to the vitrification timescale as discussed in the text. Lateral diffusion correlation times are calculated from published diffusion coefficients (18) as the time required for a mean squared displacement of 1 nm (i.e., a distance corresponding to nearest neighbor lipid exchange). Additional references: trans-gauche isomerization (19) wobble and axial rotation (20) and flip-flop (21).

In addition to sphingolipids and glycerophospholipids, the plasma membrane of eukaryotic cells also contains an abundance of cholesterol (30-50 mol%) that serves a critical role in modifying the phase behavior. Figure 1a demonstrates the influence of cholesterol on lipid phase behavior. When added to saturated lipid bilayers at concentrations of a few mole percent, cholesterol forces nearby lipids to align parallel with the bilayer normal, thus converting the L_β′ phase to an untilted L_β gel phase (22). At higher concentrations, cholesterol breaks up the long-range crystalline order that is characteristic of gel phases, allowing the lipids to diffuse rapidly while maintaining their high chain conformational order (23). When added to lipids in the disordered L_α or L_d phases, cholesterol increases the probability of trans conformers, thus straightening the hydrocarbon chains. The ability of cholesterol to create and maintain high lipid chain conformational order while allowing for fast lipid lateral diffusion results in a unique state termed the liquid-ordered (L_o) phase (24). Raft phenomena in cell membranes are thought to be a manifestation of L_d + L_o phase coexistence (25).

Secondary bilayer structure from cryo-EM images

Membrane secondary structure—i.e., the phase state of a bilayer and its associated chemical and physical properties—has been characterized as a function of temperature and cholesterol concentration for many glycerophospholipids and sphingolipids. As mentioned previously, NMR and scattering techniques have played a prominent role in determining the detailed structure of the various bilayer phases. Among the structural parameters obtained from these techniques are bilayer thickness, average area per lipid, and hydrocarbon chain order parameter profiles, while lateral diffusion can be measured with fluorescence correlation spectroscopy and fluorescence recovery after photobleaching.

To our knowledge, Lepault and coworkers (26) were the first to focus a cryo-EM study entirely on lipid bilayer structure. They prepared unilamellar liposomes from PC lipids by controlled detergent dialysis and vitrified the samples with liquid nitrogen-cooled propane. Three saturated lipids—DLPC, DMPC, and DPPC—were investigated, covering a wide range of main chain transition temperatures (T_M = −2, 24, and 42°C, respectively). The unsaturated lipid DOPC (T_M = −17°C) was also studied to determine the influence of a cis double bond in the lipid chains. Cryo-EM projection images showed quasi-spherical vesicles characterized by two high density, roughly concentric rings that were attributed to the electron-dense polar headgroups of the inner and outer bilayer leaflets. The bilayer thickness, estimated as the distance between the high-density lines, was found to be sensitive to sample temperature immediately prior to vitrification: vesicles showed a uniform thickness if vitrified from a temperature well above T_M but a faceted appearance with variable thickness if vitrified from well below T_M. For all bilayers, the measured thickness was 40 ± 5 Å regardless of vitrification temperature.

Further investigation with electron diffraction measurements revealed characteristic reflections of an L_β′ phase for bilayers vitrified from below T_M, suggesting that gel phase bilayers are structurally preserved during cryo-preservation. Remarkably, DMPC and DLPC bilayers vitrified from well above T_M (i.e., in the L_α phase) showed reflections consistent with an untilted gel (L_β) phase, rather than the single broad reflection characteristic of L_α. In contrast, the electron diffraction pattern of vitrified DOPC samples was consistent with an L_α phase, although the 4.4 Å chain-chain spacing was smaller than expected, indicating slightly tighter chain packing. This suggests that the kink caused by the cis double bond is sufficient to prevent conformational reorganization of the chains into a gel phase during vitrification, although some degree of ordering does occur. Figure 1b summarizes these vitrification-induced structural changes. This important study raises a caveat to interpreting cryo-EM images of lipid membranes due to the relative timescales of lipid motions with respect to cryo-preservation, discussed in more detail below.

Tahara et al. (27) were the first to demonstrate the sensitivity of cryo-EM to sub-nanometer changes in average bilayer thickness using vesicles composed of PC lipids vitrified from room temperature. For each composition, thickness histograms were constructed by measuring the distance between the high-density lines at 4 nm intervals around the projected vesicle circumference. Vesicles composed of the unsaturated lipid DOPC were visually uniform with a correspondingly narrow thickness distribution. In contrast, thickness distributions for the saturated acyl chain species DLPC, DMPC, and DPPC were broad, reflecting variability within individual vesicles that was clearly visible in images. The observations of (1) smooth DOPC vesicles and (2) faceted saturated lipid vesicles were consistent with the findings of Lepault et al. a decade earlier and suggest either an incomplete transition to L_β or a full transition to L_β′ for DLPC and DMPC during vitrification (the lack of electron diffraction data precludes a definitive interpretation). Though not commented on by the authors, vesicles of the mixed-chain lipid POPC were visually heterogeneous, and the width of the POPC thickness distribution was intermediate between that of DOPC and the saturated lipids.

Although the thickness distributions for saturated lipids were broad, their averages showed a linear relationship between acyl chain length and thickness (Fig. 2a, black squares) that is consistent with trends in the bilayer headgroup-to-headgroup distance measured by X-ray scattering (28). These comparative data are shown in Fig. 2a for saturated lipids in the L_β′ phase (blue squares) and L_α phase (red squares). Interestingly, the addition of 30 mol% cholesterol to each of the saturated lipids produced visually uniform vesicles and narrowed the thickness distributions. This observation is consistent with the known influence of cholesterol on saturated lipid bilayers above their T_M: cholesterol orders the chains, converting the L_α phase to an L_o phase in which the chains are in an all-trans conformation and therefore would not be sensitive to further straightening during vitrification (Fig. 1b, lower). The average thickness of cholesterol containing DLPC and DMPC bilayers was slightly larger than the non-cholesterol containing bilayers, reinforcing that the non-cholesterol containing bilayers undergo an ordering transition during vitrification. More recently, a cryo-EM study by our group (29) also found a linear trend between chain length and thickness for unsaturated PC lipids (Fig. 2b, solid black circles) in agreement with the trend reported from a joint analysis of SANS and SAXS data (Fig. 2b, red circles) (30).

Figure 2. Cryo-EM is sensitive to bilayer thickness.

Bilayer thicknesses measured by cryo-EM compared to scattering techniques for saturated (a, (27) and unsaturated (b, (27,29) PC lipids. For scattering-derived thicknesses of saturated lipids, 12:0-PC and 14:0-PC in the L_α phase were measured at 30°C (31), 14:0-PC in the L_β′ phase was measured at 10°C (28), and all other lipids were measured in the L_β′ phase at 25°C (32). For scattering-derived thicknesses of unsaturated lipids, thickness values in the L_α phase were measured at 30°C (30)

Despite similar thickness trends, the data compiled in Fig. 2 reveal significant differences in absolute thicknesses measured by cryo-EM and scattering methods. Even though the physical origin of both measurements is electron density variation across the bilayer, a direct comparison is inherently problematic for several reasons. In the case of X-ray scattering, the bilayer’s electron density profile is directly connected to the reciprocal space scattering intensity through a 1D Fourier transform and is thus relatively straightforward to extract through a modeling procedure. In contrast, the electron density imaged with cryo-EM is transformed by 2D projection and corrupted with a contrast transfer function (CTF) that itself depends on parameters such as the electron energy and chosen defocus length; changing these parameters can change the apparent thickness. In principle, these complications can be computationally reversed to recover the true electron density profile (and thus, a thickness that is directly comparable to values measured with scattering), but this relies on sufficiently spherical vesicles and an accurate estimation of the CTF. In practice, recovered density profiles are broader (33) than those derived from scattering experiments and become comparable only after smearing the latter with a Gaussian of 5-6 Å standard deviation, implying a substantially lower resolution of cryo-EM-derived profiles and a loss of fine structural features. More importantly, even if high-resolution, de-smeared profiles could be obtained from cryo-EM data, the vitrification-induced chain ordering described above would preclude an exact correspondence between transverse structural parameters (including thicknesses) derived from cryo-EM and scattering data.

Despite these caveats, cryo-EM is clearly sensitive to composition-dependent differences in membrane thickness at the level of a few angstroms. In the following section, we describe recent efforts that exploit this sensitivity as a contrast mechanism for detecting tertiary structure in vesicles composed of lipid mixtures.

Timescales of vitrification and lipid motions

The results presented in the previous section highlight the importance of the timescales of various lipid motions relative to the vitrification time for cryo-preserving samples, demonstrated schematically in Fig. 1c. The temperature of a thin sample plunged into liquid ethane drops at a rate of 10⁵ – 10⁶ K/s (34). Accounting for the exponential decrease in lipid motional rates with decreasing temperature (18), motions with μs or faster correlation times may occur to an extent that would measurably alter membrane structure as cooling proceeds. Among the fastest motions are trans-gauche isomerization in the lipid chains that occur with a characteristic timescale of 10-100 ps at biological temperatures. Indeed, the Lepault (26) and Tahara (27) results for saturated lipids above T_M suggest that the cooling rate, while fast, nevertheless allows for ordering of lipid chains through gauche-to-trans isomerization that drives the bilayer into the gel phase. This raises the question: why was the untilted L_β phase observed with electron diffraction, rather than the tilted L_β′ phase that forms when these bilayers are slowly cooled through T_M? Though the authors did not speculate, a possible answer is that the slower dynamics of axial rotation and wobble that would be required to convert the nascent L_β bilayer to L_β′ are effectively frozen out or at least substantially hindered early in the cooling process.

While the factors discussed above can lead to artifacts in membrane secondary structural features, a different set of dynamics must be considered for understanding the potential influence of vitrification on membrane tertiary structure, where the timescale of lipid lateral diffusion is key. Potential vitrification-induced artifacts (i.e., cooling-induced demixing or other changes in domain composition or structure) would require sufficient time for lipids to sample their translational degrees of freedom. The diffusion distance of a lipid during vitrification can be estimated from published values of the temperature dependence of lipid diffusion coefficients and vitrification cooling rates noted previously. Such estimates produce ranges of 5-20 nm depending on composition and phase state of the bilayer immediately prior to vitrification. One inference is that a composition that is initially uniformly mixed but near a phase boundary in temperature could demix locally at the onset of cooling. The maximum cluster size that could result from this is half the diffusion distance estimate, or ~ 2.5-10 nm. This suggests that the lateral organization of lipids in a mixture should be essentially preserved in a vitrified sample, possibly with local lipid reorganization at domain boundaries. It also suggests that the lateral structure should be visible in a cryo-EM image provided there is sufficient difference either in thickness or electron density of coexisting membrane phases.

Tertiary bilayer structure from cryo-EM images

The first studies to provide visual evidence of membrane phase coexistence were published simultaneously in 2020. In one of these studies, our group (29) used cryo-EM for direct probe-free imaging of domains in vitrified biomimetic and biological membranes. To probe lateral heterogeneity, bilayer thickness was measured at 5 nm increments around the projected vesicle circumference and binned to produce thickness histograms similar to the analysis of Tahara et al. (27). As noted above, a consistent increase in average bilayer thickness with increasing chain length was found for single-component bilayers of unsaturated lipids. The same analysis was then applied to vesicles of a canonical domain-forming model membrane mixture, DPPC/DOPC/cholesterol (4/4/2), revealing a bimodal thickness distribution consistent with phase separation that contrasted sharply with the uniform thickness distributions of single-component vesicles (Fig. 3). The thickness difference between the two peaks of the distribution was 6.3 Å, consistent with the thickness difference of L_o and L_d phases measured with X-ray and neutron scattering (35,36). Importantly, similar analyses performed on giant plasma membrane derived vesicles (GPMVs) isolated from eukaryotic cells provided the first direct visualization of phase separation in bilayers derived from the complex lipid mixture of mammalian plasma membranes.

Figure 3. Lateral phase separation in cryo-EM images.

a and b, field of DOPC vesicles exhibit a uniform thickness distribution centered at 29.7 Å. c and d, vesicles composed of DPPC/DOPC/POPG/Chol 40/35/5/20 show a bimodal distribution with peaks at 28.7 Å and 36.1 Å. Scale bars are 100 nm. Figure adapted from (29).

In a parallel study, Cornell et al. (37) developed two visualization methods to characterize and analyze nano-sized domains in cryo-preserved vesicles composed of ternary mixtures of DPPC, diphytanoyl-phosphocholine, and cholesterol. These authors employed cryo-electron tomography for data acquisition to generate 3D reconstructions of isolated vesicles. They then measured the bilayer thickness through a central slice of the tomograms by analyzing pixel-by-pixel profiles of filtered and edge-detecting algorithms. Similar to the results of Heberle et al. (29), the thickness of the bilayers with L_o composition was larger compared to that of vesicles with L_d composition. In the second method, a visual tag was produced using a trimer of the fluorescent protein mCherry that was targeted to the L_d domain by binding the His-tagged mCherry construct to Ni-NTA-dioleoylglycerosuccinylimino-diacetic acid (a lipid that preferentially distributes to L_d domains). This creative approach enabled the visualization of macroscopic domain formation with fluorescence microscopy to confirm phase separation. Parallel samples processed by cryopreservation and cryo-tomography revealed a “brush” of electron density representing mCherry molecules evident along portions of the bilayers thought to represent L_d phases. This use of protein density targeted to unique domains is one of several potential ways to increase contrast in the electron microscopic images that is yet to be fully exploited. Importantly however, acknowledged by the authors, is that the probes themselves can have unpredicted impacts on lipid packing density, or their physical/chemical properties.

Outlook

In a relatively short time, cryo-EM has contributed enormously to solving high-resolution structures of isolated proteins and protein assemblies in vitro and in situ, but its application to the study of membrane structure has lagged. This is despite work decades ago, highlighted in this review, that provided tantalizing hints of its potential. After a hiatus, recent advancements in cryo-EM instrumentation and automation—combined with interest sparked by the debate over the existence of lipid rafts—have brought the analysis of membrane structure back to the forefront. Our group and others have demonstrated that cryo-EM can resolve differences in membrane thickness as small as a few angstroms in synthetic or biologically derived vesicles (29,37). Further, the thickness and contrast differences between ordered and disordered domains can be reliably discerned. These efforts have provided the first visual evidence of nanoscopic phase separation in unsupported lipid bilayers and pave the way for new investigations.

To take one example, a debate has emerged within the field of membrane biophysics regarding the physical origin of nanoscopic domains. Among the proposed explanations are curvature-induced microemulsions (38), first-order phase coexistence (39), line active lipids (40), and Ising-like critical fluctuations (41), but experimental data for testing these theories is scant as few techniques have the resolution needed to detect nanoscopic structure. The real-space images of nanodomains provided by cryo-EM should enable direct determination of correlation functions that are the signatures of the underlying physics. Establishing a robust cryo-EM workflow for characterizing these domains as a function of composition and liposome size will likely lead to new insights from synthetic vesicle studies.

Among the most significant challenges remaining is to apply these (and other) tools to the investigation of rafts in the membranes of living cells. Remarkably, the existing tools for imaging and analysis can be applied to cryo-preserved cells grown directly on EM grids. Our group has had success in imaging organelle structure in the processes of cryo-preserved cells and neurons (Fig. 4) (4). Despite the greater complexity of cellular membranes, these initial efforts clearly identify thickness differences among membranes of different organelles and the plasma membrane. Existing tools for analysis are presently being employed to quantify these differences.

Figure 4. Imaging membranes of cryopreserved cells with cryo-EM.

Example image of an area of a hippocampal neuron cryopreserved after 10d of growth from plating directly on an EM grid (see (4) for more details). The image on the left identifies the plasma membrane, a recycling vesicle, and a mitochondrion at the edge of a neuronal process. Using methods developed in (29), the thickness of the membrane was determined in each of the membranes and is displayed in the pseudo-colored overlays on the right.

Several potential advancements further brighten the outlook for cryo-EM applied to membrane structure. Data acquisition parameters continue to be refined, and existing hardware such as energy filters can reduce noise in the images that degrades contrast (42). Even larger gains may be within reach with the use of Volta phase plates that have the potential to significantly increase image contrast. There also remain significant opportunities in image restoration and analysis to extract the maximum amount of information from cryo-EM experiments. Image restoration involves post-processing steps to restore the image of the original object by correcting for artifacts introduced by phase-contrast imaging in the electron microscope. These are well-established techniques in the single particle reconstruction field (43) and can be easily applied to optimize images of vesicles or cellular membranes for subsequent analysis. Perhaps most exciting is the potential to employ machine learning algorithms to help automate analyses. Further developments in training networks to recognize contrast and thickness differences (or both) in lipid bilayers has immense potential for high throughput analyses with an associated increase in statistical power. These advancements, either separately or in tandem, should further enhance the ability to discriminate boundaries between ordered and disordered phases and refine the ability to differentiate domains on the spatial scale of nanometers.

There are also unique opportunities in sample preparation that have yet to be fully exploited. Strategically employed lipids tagged with contrast enhancing properties (Br, I, gold, or other electron-dense elements) have the potential to further increase contrast. Finally, environmentally controlled automated sample preparation has yet to be fully exploited. Control of temperature is fundamental to the study of bilayer properties and present semi-automated blotting apparati exist to provide this control. In addition, avoiding the step of blotting with filter paper to reduce sample thickness before vitrification can now be avoided. Jet vitrification of thinly painted samples in well-controlled environmental chambers has the potential to further improve sample preparation. The combination of creative sample preparation and vitrification, high-throughput data acquisition in optimally configured cryo-EM microscopes, and advances in image processing will all be employed to lead the field of membrane structure and biophysics forward.

Summary Section – Key Points.

• The relative bilayer thickness of vitrified liposomes and cell membranes can be measured with ~ 1 angstrom accuracy and a spatial resolution of 4-5 nm in cryo-EM projections.

• Liposomes composed of fully saturated lipids in the fluid phase are converted to an untilted gel phase during vitrification due to fast rotational isomerization in the hydrocarbon chains, but vitrification-induced chain ordering is less pronounced in membranes containing unsaturated lipids and/or cholesterol.

• Because of the relatively slow time scale of lipid diffusion compared to vitrification, lateral lipid organization is retained during cryo-preservation, enabling the direct imaging of nanodomains in raft-mimetic model membranes and cell-derived plasma membrane vesicles.

• Application and optimization of existing tools for cryo-EM image acquisition and image restoration and analysis should facilitate characterization of nanodomains in cryo-preserved cells.

Abbreviations:

cryo-EM

cryogenic electron microscopy

NMR

nuclear magnetic resonance

CTF

contrast transfer function

PC

phosphatidylcholine

DLPC

1,2-dilauroyl-sn-glycero-3-phosphocholine

DMPC

1,2-dimyristoyl-sn-glycero-3-phosphocholine

DPPC

1,2-dipalmitoyl-sn-glycero-3-phosphocholine

DOPC

1,2-dioleoyl-sn-glycero-3-phosphocholine

POPG

1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1’-rac-glycerol)