Dynamic pattern generation in cell membranes: current insights into membrane organization

Abstract

It has been two decades since the lipid raft hypothesis was first presented. Even today, whether these nanoscale cholesterol-rich domains are present in cell membranes is not completely resolved. However, especially in the last few years, a rich body of literature has demonstrated both the presence and the importance of non-random distribution of biomolecules on the membrane, which is the focus of this review. These new developments have pushed the experimental limits of detection and have brought us closer to observing lipid domains in the plasma membrane of live cells. Characterization of biomolecules associated with lipid rafts has revealed a deep connection between biological regulation and function and membrane compositional heterogeneities. Finally, tantalizing new developments in the field have demonstrated that lipid domains might not just be associated with the plasma membrane of eukaryotes but could potentially be a ubiquitous membrane-organizing principle in several other biological systems.

Introduction¹

Spatiotemporal organization of molecules in cells is increasingly thought to play an important role in biological regulation [1]. A ubiquitous method by which cells utilize such spatiotemporal organization to bring about regulation is through clustering of biomolecules [2]. One of the best studied examples of how cells exploit clustering as a mechanism to functionally organize biomolecules is by patterning membrane proteins and lipids to form micro- and nano-domains in cell membranes [3–5]. It is now well recognized that cell membranes contain a variety of such domains and that these domains impose a local structure that prevents free mixing of the lipid and protein components of the membrane [6].

A major paradigm that has shaped much of our thinking about domains in cell membranes is the lipid (membrane) raft model [7]. Given the diversity in the interpretation of membrane rafts, we have utilized a traditional meaning of raft as cholesterol-rich, nanoscale heterogeneity in cell membranes formed as a consequence of liquid-liquid phase separation of lipids as the working definition in this review. This model, while recognized as over-simplified [8, 9], has provided a useful testbed for generating and testing hypothesis regarding the underlying mechanisms that control the size and shape of domains in the two-dimensional environment of cell membranes. Given the notorious difficulty in detecting raft domains in cells, the raft model has also driven advances in technology to develop techniques capable of imaging biomolecules with increasing spatial and temporal resolution, as well as new methodologies to label those molecules to enable their detection.

In this review, we highlight recent advances in the study of rafts as a pattern generation mechanism in cell membranes. First, we provide an overview of current optical microscopy-based methods and probes used to detect rafts in cells. Next, we describe new conceptual advances regarding the functional consequences of such patterning. Specific examples we discuss include how rafts function to compartmentalize cellular signaling, mechanisms responsible for targeting transmembrane proteins to rafts, and how pathogens exploit rafts to gain entry into cells. Finally, we discuss newly discovered sites where membrane domains form and areas for future exploration such as the relationship between membrane geometry and compositional heterogeneity.

Technical advances in optically detecting membrane nanodomains in cell membranes

Several important properties of rafts can be studied in model membrane systems using conventional microscopy-based approaches. For example, the phase preference of membrane proteins and lipids can be directly imaged in giant plasma membrane vesicles (GPMVs), which form micron-scale coexisting raft and non-raft domains [10]. However, current wisdom holds that in living cells rafts are highly dynamic and have sizes on the orders of tens of nanometers [11]. These spatial and temporal constraints have made it difficult to study them using conventional microscopy methods. In the last decade, there have been several developments in experimental techniques that have enabled a better understanding of native membrane organization by addressing both these bottlenecks.

Stochastic localization and single molecule techniques

The raft model predicts that raft-associated proteins and lipids should cluster together to form nanodomains that exclude non-raft proteins. However, since these domains are diffraction-limited, they cannot be resolved using regular fluorescence imaging methods. Techniques such as Förster Resonance Energy Transfer (FRET) have been used previously to probe the distribution of raft-associated proteins and lipids over distances of 100Å or less [12]. However, a limitation of FRET-based approaches is that they report on distances and not distributions of molecules per se, necessitating the use of theoretical modeling to interpret their meaning [12].

In recent years, super-resolution methods have advanced our understanding of several biological systems including plasma membranes organization by their ability to probe past the diffraction limit. One approach to super-resolution microscopy, the use of stochastic localization of fluorophores, has become a powerful tool to probe for the presence of different types of local heterogeneities in cell membranes (Figure 1). In these techniques, the different fluorophores which are spatially localized together within a diffraction limited spot are individually isolated temporally and various reconstruction methods are used to fit isolated emitters [13–15]. Typically, these methods can achieve localization to within tens of nanometers while imaging live cells [16]. These super-resolution methods have been used to describe aspects of several membrane organization including how raft associated immune-receptors cluster and signal [17–19], viral protein-membrane interaction [20, 21] and growth factor receptor organization [22, 23]. Importantly, several analysis methods are being simultaneously developed to understand the organization in a reconstructed image [24–27]. Clustering tools are now increasingly complex and can describe even the three-dimensional organization of molecules in a cell and not just in the membrane [28, 29]. Further, a number of analysis tools for super-resolution are now available free to users [28, 30–32]. Despite these advances in spatial prowess, a major drawback in these experiments is that the temporal resolution of the technique is still a few seconds, which tends to average many transient clustering events [24].

Figure 1. Examples of optical methods to probe membrane heterogeneity in cells.

(Upper left) High-speed single particle tracking is commonly used to detect regions in membrane where anomalous diffusion occurs. (Upper right) Spectral imaging of membrane order utilizes lipid probes whose fluorescence emission are sensitive to the local membrane environment. A typical spectral probe is red-shifted in the disordered phase and is blue-shifted in ordered phase. Thus, by spectrally delineating the emission spectra, it is possible to detect ordered raft-like region in cell membranes. (Bottom left) In stochastic localization methods, the membrane clustering of the biomolecule of interest is determined relative to the distribution of a known (dis)ordered phase preferring probe. If the molecule co-clusters with the ordered phase probe, it is raft associated. (Bottom right) STED-FCS measures the diffusion of membrane-associated proteins and lipids through a sub-diffraction focal volume. In the presence of membrane heterogeneity, their diffusion time through the beam varies as a function of the beam waist diameter. By determining this relationship, the underlying molecular architecture of the membrane can be deciphered.

Ameliorating these deficiencies in super-resolution methods are complementary developments in live cell single molecule methods. These high-speed experiments have been used to study of diffusion of lipid probes in membrane domains and confinement of molecules in membrane in the millisecond timescale [33–35]. These approaches allow for the interrogation of clustering events that occur over very fast timescales as well as the identification of underlying features of membrane architecture that slow or transiently halt the diffusion of proteins and lipids in the plane of the membrane (Figure 1). Advances in this area have been closely coupled to the development of fluorescent analogs of physiologically relevant lipids. This has represented a significant challenge to the field since the presence of fluorophores can modify the properties of lipids [36]. In recent years, the development of high-quality lipid probes has facilitated high spatiotemporal studies in lipid and protein dynamics [36–40]. These studies have provided information on the residency and co-diffusion timescales of different biomolecules in membrane. Further, the trajectories of single molecule measurements reveal the underlying characteristics of molecular motion. For instance, the trajectory of a molecule undergoing hop-diffusion is different from the trajectory of a molecule undergoing confined diffusion [41].

Of specific importance in understanding membrane organization has been the development of novel sphingomyelin and ganglioside probes used to measure lipid-lipid and lipid-protein interactions in diverse model systems from artificial membranes to high-speed live cell experiments [36, 37, 39]. These experiments have demonstrated evidence of transient recruitment of sphingomyelin, gangliosides to CD59-rich plasma membrane domains in cells. These studies have shown that the co-diffusion timescales are on the order of hundreds of milliseconds, demonstrating the importance of the sensitivity of the experimental technique. Understanding dynamics of lipids is still a nascent field and combining these single-molecule tracking experiments with other complementary super-resolution techniques will be a powerful tool to understand lipid dynamics and behavior.

Fluorescent correlation measurements

Another general approach for studying diffusion characteristics of fluorescent molecules and how membrane heterogeneities affect diffusion is fluorescence correlation spectroscopy. Fluorescence correlation measurements determine how the fluorescence in a diffraction limited spot or volume fluctuates [42, 43]. The fluctuations in the signal are dependent on the diffusion of fluorescent molecules in and out of the observation volume and are used to generate correlation functions which are then related to the diffusion of the molecule, usually through analytical or semi-analytical methods. While correlation experiments are relatively old techniques with fluorescent correlation spectroscopy (FCS) being developed close to half a century ago (reviewed in [44]), new developments in the field have enabled application of this technique to the study of lipid rafts in cells. The advantage compared to other techniques such as STORM is that the temporal resolution is in the milliseconds range. Though typically performed as a spectroscopic technique in a confocal volume, correlation measurements are often combined with other methods to extend the capabilities of the system. Variants of FCS techniques such as brightness analysis methods, Total internal Reflection FCS, scanning FCS, Fluorescence cross correlation spectroscopy, and inverse FCS, have all been developed on the backbone of traditional FCS methods and have extended its capabilities [45–49].

One of the most exciting development in this field is the development of spot variation FCS (svFCS) [50]. Here, diffusion coefficients are measured for different illumination volumes. While the principle of svFCS was established over 10 years ago [42, 51], the development of stimulated emission depletion (STED) sub-resolution illumination methods have now moved svFCS past the diffraction limit, thereby tremendously improving its use [52]. In STED, by combining two lasers, one to excite and other to deplete, and with clever spatial arrangement of the two laser pulses, it is possible to selectively excite a sub-diffraction volume. The variation of the diffusion coefficient as a function of beam diameter is characteristic of membrane confinement behavior [53]. Thus, using STED-FCS it is potentially possible to distinguish the underlying biological mechanism behind confinement of biomolecules (Figure 1). STED-FCS has been used to show that sphingolipids and GPI-anchored proteins do not diffuse freely in the plasma membrane, while phosphoglycerol lipids do [54–56]. Further, it has been used to demonstrate that both actin and lipid nanodomains contribute to the compartmentalization of lipids in the plasma membrane [57].

An important caution while interpreting FCS measurements is to understand the caveats in data interpretation. For instance, using Ising model simulations, Veatch and colleagues showed that incorrect assumptions in svFCS experiments can result in erroneous interpretation of diffusion behavior as anomalous diffusion when in reality it could still be Brownian [58]. Thus, it is important to reconcile the results of FCS with other methods, especially SPT [59], to ensure that complementary techniques demonstrate consistent results.

Development of novel chemical tools to study lipids in membranes

As previously noted, a critical component for the use of fluorescent microscopy in studying lipid domains is the development of lipid-specific fluorophore probes. The development of specific sensors for different lipids has received considerable attention. Click chemistry has been used to generate novel ceramide probes which can be used in super-resolution studies [38]. Naturally derived proteins and peptides with binding propensity for specific lipids have been retooled as lipid sensors. Perfringolysin O (PFO) secreted by Clostridium perfringens is the most studied of these and has been used for sensing cholesterol [60]. In a recent study, PFO was modified to probe the transbilayer distribution of cholesterol on membrane bilayers [61]. Other proteins have been isolated from different organisms that bind either selectively or non-selectively to different lipids. Lysenin, a protein isolated from the earthworm Eisenia foetida, binds to sphingomyelin with high specificity [62]. On the other end of the spectrum, Nakanori, a protein recently isolated from a non-edible mushroom, was found to bind to both cholesterol and sphingomyelin [63]. Many of these proteins can be fluorescently conjugated and thereby are thus extremely useful in studying various membrane properties.

An important factor to be kept in mind in the use of lipid probes is that specificity does not imply uniqueness. Further, it is important to understand the effects of the probe on membrane rafts. A classic example is cholera toxin B (CTxB), a widely utilized lipid raft marker in cellular membranes. While CTxB is an indicator of ordered phase, it also induces phase separation and increases membrane order [64, 65]. Also, the phase partitioning of GM1 can influence CTxB binding to rafts [66]. Further, besides GM1, CTxB is capable of binding to glycoproteins [67]. Thus, it is important to understand the strengths and weaknesses of a particular lipid probe, especially when using single molecule and super-resolution methods. Also, no protein is completely raft associated or completely non-raft associated. Even model raft-associated proteins such as CTxB do not partition entirely into the ordered phase. Thus, a critical control for determining raft partitioning in cells is to use to two distinct control probes, one for the ordered phase and other for the disordered phase. A lipid raft probe should not only be raft-associated but also should be significantly depleted from the non-raft region. However, currently, there are not many good probes which partition into the ordered phase. Development of ordered phase preferring probes is an important area of future research [68].

Spectral Imaging

A prediction of raft hypothesis is the presence of nanoscale regions of altered membrane order in plasma membrane of cells. Microscopy can be used to detect the presence of increased membrane order in live cells the help of fluorescent probes whose emission spectra depend on the local environment (Figure 1). While most of the previously described optical techniques probe membrane heterogeneity through the lens of a few well-characterized membrane biomolecules, spectral microscopy has the ability to determine the properties of the membrane as a whole [69]. In spectral imaging, a lipid probe whose fluorescence spectrum is dependent on its local environment is added exogenously to the membrane system being studied so as to report the membrane order. The most well characterized of these dyes is laurdan, which absorbs in the UV range and has a blue-shifted spectral emission with increasing membrane order [70]. However, laurdan suffers from two limitations: a) the dye bleaches rapidly and b) the separation of two emission wavelengths corresponding to ordered and disordered phase (430–470 and 480–530) is not ideal for measurement in many confocal systems [71]. A derivative of laurdan known as C-laurdan with similar absorption/emission characteristics but less photobleaching has been developed [71, 72]. Another dye that is increasingly used is the commercially available ANEPPDHQ. This dye absorbs at 488nm and has emission spectra corresponding to 488 and Cy3 channels depending on membrane order. It thus can be used in most confocal microscopes with standard filters [73]. New probes are continuing to be developed with different spectral properties, enhanced photophysics and increased sensitivity to membrane order [74–76]. Currently, spectral imaging has been used to characterize gross membrane order changes [69]. It would be interesting to determine how membrane order changes when a specific protein binds. For these experiments it becomes imperative to use spectral probes along with fluorescent protein tags. C-Laurdan for instance can be used in conjunction with proteins conjugated with red tags such as mKate or mCherry to determine how binding of protein affects membrane order [25]. However, there is a need for developing environment sensitive red dyes which can be used in conjunction with GFP, for better simultaneous imaging of membrane order and protein binding.

Another bottleneck with spectral imaging had been the spatial resolution of the technique, which until recently was diffraction limited. More recently, spectral imaging has been combined with super-resolution methods to detect domains on the order of tens of nanometers in size [77, 78]. Using this technique, nanometer-scale spatially resolved domains in live cell membranes were observed. Spectral super-resolution imaging is still in its infancy and has the potential to directly visualize membrane rafts.

While we focus on the fluorescent experimental methods in this review, other non-fluorescent methods have also enabled substantial contributions in our understanding of membrane domain organization. Detergent Resistant Membrane (DRM), the traditional method of understanding of studying rafts, has largely been sidelined as the principal experimental method to determine whether a protein is targeted to ordered phase [7]. However, many biochemical experiments can now be combined with other experimental methods as microscopy or novel probes have been used to understand biological role for different lipids. For instance, biochemical methods were used to show that there are three pools of cholesterol that are important in regulating cholesterol levels in the plasma membrane [79]. This biochemical work has since been complemented using an emerging high resolution technique, NanoSIMS, where the authors have shown that accessible cholesterol is enriched in microvilli in Chinese hamster ovary cells [80]. Similarly, atomic force microscopy, a well-characterized experimental method to determine membrane properties has been combined with emerging methods such as STED-FCS, mass-spectrometry and X-ray techniques which has greatly expanded the applicability of each of the different methods [81, 82]. Other novel non-fluorescent methods such as high-resolution mass spectrometry, interferometric scattering microscopy, Raman microscopy have gained prominence in determining various membrane properties and detecting lipid clusters although given the specialized nature of these methods, wide use of these methods has currently been rather limited [83–88]. High resolution mass spectrometry methods have suggested for example that hemaggluthin, an influenza envelope protein which is known to cluster in plasma membrane, does not localize to cholesterol and sphingolipid enriched domains at a resolution of around 50 nm [85]. More recently, this technique has been expanded for three-dimensional imaging [89]. Other traditional methods such as EPR, NMR, x-ray and neutron scattering, though still not popular with biologists in the field are used extensively in determining various membrane properties including order in model membranes [90–92]. Classical approaches such as electron microscopy and its variants also remain useful to detect clustering of membrane-associated molecules [93–96]. Use of electron microscopy methods for studying lipid raft field has resurged in part due to its increasing use in studies of bacterial rafts [97–99].

Besides experimental methods, in-silico simulations have been used to study membrane properties inaccessible to experiments [100, 101]. For instance, molecular dynamics have been used look at chain lengths and how coupling between the bilayer affects membrane organization [102]. In recent years, several developments have emerged in computational approach for studying complex membranes, including development of better forcefields and increasing complexity of the lipids that are simulated [103–105]. Recently model membranes with more than 50 types of lipids have been simulated to look at domain formation [106, 107]. Finally, in recent years, in a growing number of studies computer simulations have been used to interpret and validate experimental results [58, 108]. Discussing all these are beyond the scope of this review and we would like to point an interested reader to several reviews that dwell more deeply into in silico methods for understanding membrane organization [109–113].

New insights into the role of rafts as spatial organizers in signaling and trafficking events

In addition to advances in technology that can be used to interrogate membrane architecture, our understanding of functional properties of raft-like domains has also continued to evolve over recent years. Lipid rafts have long been proposed to functionally segregate proteins and lipids within different local membrane compartments, thus regulating their interactions. Here we discuss recent advances in our understanding of rafts in compartmentalization of signaling and trafficking events.

Role of stabilized rafts in BCR signaling

A classic example of how clustering and crosslinking of proteins by rafts is linked to functional outcomes is the activation of immune receptors in cell membranes. Initial links between the activation of immune cell receptors and lipid rafts were established through the isolation of detergent-resistant membranes, cholesterol depletion, and biophysical approaches [114–116]. In recent years, single-molecule microscopy and super-resolution microscopy have shed detailed light on both the temporal and spatial activation of immune receptors in cells and how lipid domains play a role in these processes [29, 117, 118].

A recent example of the role of rafts in compartmentalizing immune cell signaling comes from studies of B-Cell Receptor (BCR) activation using super-resolution techniques [17]. To identify nanoscopic raft and non-raft domains in cells membranes, model peptides and proteins known to preferentially associate with either raft (ordered) or non-raft (disordered) domains in model membrane systems were used as markers. For instance, the well-established raft marker CTxB was found in distinct clusters which excluded a peptide with a polybasic sequence and geranylgeranyl modification that prefers the disordered phase. The size of the ordered phase clusters was found to be around 50nm, close to the resolution limit of the method. These findings suggest that phase-like domains exist in intact B-cell membranes.

Using these markers to differentiate raft and non-raft domains, the authors next asked whether the BCR was enriched in pre-existing ordered domains. In the absence of stimulation, BCR was only weakly associated with ordered domains. However, upon clustering with an antigen, BCR clusters become enriched in ordered phase probes while excluding disordered phase probes. This suggests that BCR clustering stabilizes an ordered membrane domain rather than moving into existing rafts in response to crosslinking. Interestingly, the ordered domains extended beyond the BCR clusters themselves, indicating additional signaling complexes may be assembled close to the BCR clusters. Lyn kinase, a regulator of early BCR signaling, was recruited into the ordered domains stabilized by BCR signaling, whereas the phosphatase CD45, another important regulator of BCR activation, was excluded (Figure 2A). Computational modeling demonstrated the stabilization of a raft phase by BCR crosslinking and corresponding exclusion of non-raft signaling partners can give rise to local phosphorylation of proteins, a result that was also observed experimentally. These experiments demonstrate raft stabilization induced by receptor crosslinking can functionally sequester receptors and their effectors in cell membranes [17]. It will be important to determine whether this represents a general mechanism utilized by other raft-dependent signaling pathways in future studies.

Figure 2. Role of lipid domains in signaling.

(A) Model highlighting the role of ordered phase domains in downstream signaling of BCR. Clustered BCR behaves similar to clustered CTxB in its preference for partitioning with ordered phase probes in cell membranes. These clusters also recruit kinases such as Lyn which enable phosphorylation of effector molecules and exclude CD45, a phosphatase. The curves on the bottom illustrate the level of enrichment or depletion of the indicated proteins from the clustered BCR and clustered CTxB domains. Figure is adapted from reference [17] (Creative Commons Attribution License). (B) Model for how mechanical force induces PLD2 signaling. In resting cells, PLD2 is localized in rafts, which sequester it away from its substrate PC and activator PIP₂. Mechanical disruption causes disruption of rafts, thereby increasing access of PLD2 to both PC and PIP₂. Figure is adapted from reference [119] (Creative Commons Attribution 4.0 International License).

Impact of controlled disruption of rafts on mechanotransduction

If rafts function to compartmentalize signaling, then disruption of rafts should enable molecules that are otherwise segregated from one another to interact. An interesting example of how disruption of rafts regulate cellular events comes from a recent study using fast super-resolution imaging to investigate the role of rafts in mechanosensation [119]. The focus of this study was phospholipase D2 (PLD2), an enzyme whose activation is important for the transduction of force in muscle and cell growth. PLD2 is dually palmitoylated [120] and is predicted to associate with rafts. Consistent with this, PLD2 was shown to partially localize to domains enriched in the raft marker CTxB in a cholesterol-dependent manner. These domains were distinct from those occupied by phosphatidylinositol 4,5-bisphosphate (PIP2), which is known to activate PLD2 [121]. This leads to the interesting question of how PLD2 can interact either with PIP2 or its substrate phosphatidylcholine, which also presumably resides primarily outside of rafts. One mechanism by which this could be accomplished is through the dissolution of PLD2-containing rafts. Consistent with this, cholesterol depletion led to the release of PLD2 from CTxB-containing rafts and increased its proximity to PIP2. To test whether mechanical force might have a similar effect, PLD2 activation was monitored using a fluorescent product release assay. Strikingly, either mechanical shear or cholesterol depletion was found to lead to activation of PLD2, and combining the two treatments had a synergistic effect.

These findings suggest an intriguing mechanism by which mechanical forces are transduced in cell membranes [119]. In this model, shear force disrupts rafts, thereby liberating PLD2 from a raft-confined localization and allowing it access to both its substrate and activator (Figure 2B). Thus, this model proposes force-induced mixing of membrane domains ultimately underlies mechanotransduction. However, it not yet clear how much mechanical force is required to cause raft disruption or how long this effect would be expected to last. It is also unclear if cholesterol depletion may have additional effects beyond raft disruption that contribute to these signaling events. Thus, more work needs to be done to understand how these interesting effects of force on membranes are mediated. Finally, it is possible that the coupling of mechanosensing and rafts is not restricted to PLD2 but is a general paradigm for many other membrane-associated proteins.

Raft targeting signals and their role in directing protein trafficking through the secretory pathway

Intertwined with the idea of rafts influencing biological functions is the question as to how certain proteins are targeted to lipid rafts. Several factors are thought to contribute to this process [122]. Lipid modifications such as palmitoylation and GPI anchors are well-known raft targeting motifs [123]. Further, certain proteins can bind to lipids which are enriched in the ordered phase. Crosslinking of membrane-associated biomolecules can also result in raft stabilization. Crosslinking of GM1 via CTxB binding to membranes close to a demixing point drives phase separation [65], and as discussed above, raft stabilization in response to receptor crosslinking is important in BCR signaling [17]. It should be noted however that this is not general scenario; immobilization of GPI-anchored proteins, for example, does not nucleate the formation of ordered domains [124].

Many transmembrane proteins are excluded from raft domains, and packing of transmembrane domains into an ordered raft phase has long been thought to be energetically unfavorable [125]. Nevertheless, a number of membrane proteins are associated with rafts, even independent of lipid modifications such as palmitoylation [126]. Until recently, the structural basis for raft targeting of these proteins was unclear. Recent work has now revealed that both the length and surface area of a transmembrane domain are critical determinants of raft partitioning [127, 128].

It is known from model membrane studies that lipid rafts are thicker than non-raft phase ordered phase [129]. Thus, a protein which has a longer transmembrane domain would theoretically partition into the ordered phase. This was recently experimentally tested using a panel of single-pass transmembrane domain proteins with varying lengths, and indeed, the length of the transmembrane domain was directly correlated with their partitioning into lipid rafts into GPMVs [127].

The solvent-accessible area of the transmembrane domain was also recently identified as an important determinant of raft partitioning [128]. In particular, proteins with a smaller solvent accessible area prefer to partition into raft phases, whereas the presence of bulky amino acid residues interferes with this process. These findings, together with information on whether a transmembrane protein contains a palmitoylation site, have been used to develop a model which predicts the affinity of single-pass transmembrane proteins for rafts [128]. Application of the model to the entire human proteome of single-pass transmembrane proteins revealed that compared to ER/Golgi associated proteins, plasma membrane proteins are more likely to be palmitoylated, have longer transmembrane domains, and a smaller solvent accessible area of their transmembrane domain. These findings strongly suggest raft partitioning is closely linked to the sorting and trafficking of single-pass transmembrane proteins to the plasma membrane. The mechanism by which this sorting is accomplished is unknown but appears to depend most importantly on the transmembrane domain. The mechanisms responsible for targeting multi-pass transmembrane proteins to ordered domains and the extent to which raft affinity governs their trafficking to the plasma membrane remain to be determined.

How and why pathogens target rafts

Pathogens interact with the plasma membrane of host cells during at least two stages in their life cycle, entry and egress. Rafts have long been thought to serve as a target that pathogens exploit to gain entry into cells as well as facilitate their spread [130, 131]. Here, we discuss three examples of how pathogens utilize rafts in different ways.

CTx and STx induce membrane reorganization to facilitate their uptake into cells

Association with lipid rafts is a common internalization strategy for many bacterial toxins and viruses [131, 132]. Perhaps the best-studied example of pathogens that target and exploit lipid rafts as a means to promote their internalization into host cells are the bacterial toxins cholera toxin and Shiga toxin, the causative agents of cholera and dysentery, respectively (reviewed in [133, 134]).

Cholera toxin and Shiga toxin are members of the AB5 class of bacterial toxins, consisting of an enzymatically active A subunit coupled to homopentameric membrane-binding B-subunits. Both toxins bind to multiple copies of their glycolipid receptors, ganglioside GM1 for the case of cholera toxin, and globotriaosylceramide Gb3 for the case of Shiga toxin. Membrane binding and trafficking of cholera toxin and Shiga toxin are driven primarily by their B-subunits (CTxB and STxB, respectively). CTxB and STxB are non-toxic and can be fluorescently labeled and added exogenously to cells. The endocytic mechanisms responsible for CTxB and STxB to enter cells have been extensively studied and are thought to consist of a combination of raft-dependent, clathrin-independent mechanisms and conventional clathrin-dependent endocytosis [131, 133, 134].

CTxB and STxB are not mere hijackers of rafts: they are instead thought to function as domain inducers that stabilize rafts. This activity has been especially well documented for the case of CTxB (reviewed in [135]) (Figure 3A). In vitro, CTxB binding induces the formation of a textured lipid phase [136]. The addition of CTxB to model membranes close to a phase boundary leads to the formation of large-scale liquid ordered (L_o) and liquid disordered (L_d) domains [65]. Similar effects have been observed in plasma membrane-derived vesicles membranes in response to CTxB binding, leading to the suggestion that cell membranes exist in a state where rafts can be readily coalesced [137]. CTxB binding is also correlated with an increase in the phase transition temperature of GPMVs, indicating rafts are stabilized in its presence [64, 138]. In cells, local ordering of the plasma membrane occurs in response to antibody-induced aggregation of CTxB which is correlated with an increase in actin filaments attached to the plasma membrane [139]. Additional evidence in support of the idea that CTxB binding induces membrane order comes from super-resolution spectral imaging, which revealed nanometer-scale domains in live cell plasma membranes form in response to CTxB binding [77].

Figure 3. Role of oligomerization in bacterial toxin-raft association.

(A) CTxB is a homopentamer that binds up to 5 GM1 molecules. Binding of CTxB to multiple GM1’s stabilizes rafts and enhances toxin partitioning into the raft phase. (B) VacA is a pore-forming bacterial toxin secreted by Helicobacter pylori. Binding of VacA to lipid rafts is uncoupled from oligomerization, suggesting clustering is not essential for VacA to associated with rafts.

One factor that contributes importantly to the ability of both STx and CTx to build stabilized rafts is their ability to crosslink multiple copies of their glycolipid receptors. This has important consequences on the ability of the toxins to intoxify host cells. The use of CTx mutants with reduced ability to bind multiple copies of GM1 clearly indicate multivalent binding enhances intoxification of cells by CTx [140, 141]. A single GM1 binding site is sufficient to complete the intoxification pathway but is much less efficient than the holotoxin [140]. This could potentially in part reflect an inability of CTxB to stabilize rafts and a decreased ability to associate with rafts when only a single GM1 binding site is functional [64]. Multivalent binding is also important for induction of membrane curvature by STxB [142]. The mechanism of curvature generation is linked to the structure of the toxin, which upon binding to multiple copies of Gb3 imposes local curvature on membranes [143]. This curvature generation capability turn is closely linked to the ability of the toxin to stimulate its own endocytosis [142]. Clustering of toxin-induced domains also appears to enhance their uptake into cells [144]. Clustering of STxB requires tight binding of the toxin to membrane surfaces and has been hypothesized to result from an effective attraction between toxin molecules generated by a thermal Casimir force [144].

Although there is clear evidence that lipid binding by these bacterial toxins increases local membrane order, stabilizes rafts, and bends membranes, how these processes are linked to cellular events required to facilitate toxin trafficking remains unknown. One open question is the identity of the cellular machinery involved in the initial uptake of rafts via clathrin-independent endocytosis. Some insights have recently begun to emerge [145, 146], but much of the machinery required for raft-dependent clathrin-independent endocytosis is still unknown. How rafts enable cells to sense and respond to toxin binding also remains uncertain. Given that multiple pathogens utilize similar strategies to gain entrance into cells, efforts to answer these fundamental questions are likely to have widespread implications in developing directed therapeutic targeting strategies.

VacA, an example of a pore-forming toxin that targets rafts

Even though a large number of bacterial toxins are thought to associate with rafts, in many instances it is not yet clear how they target rafts or how rafts impact their function. One interesting example is Vacuolating toxin A (VacA), the major secreted exotoxin of the bacterium Helicobacter pylori (reviewed in [147]). Intoxification of host cells by VacA is initiated by binding of the toxin to the plasma membrane, followed by toxin oligomerization, membrane insertion, and pore formation [148]. Current models suggest that one or more of these events occur in lipid rafts.

Early studies demonstrating VacA associates with lipid rafts relied on biochemical approaches to isolate raft-enriched fractions and/or depleting cells of cholesterol to interfere with raft integrity and function [5, 149–151]. More recent work has now confirmed VacA’s raft association by showing it preferentially associates with the raft phase in GPMVs [152]. How VacA is targeted to lipid rafts is currently not entirely clear and may involve multiple mechanisms. Some studies indicate that sphingomyelin, one of the receptors of VacA, acts to recruit VacA to rafts [5], while others have shown that initial binding of VacA is to receptors in non-lipid raft microdomains and the raft partitioning of VacA occurs subsequently as a result of clustering [151]. Interestingly, unlike other bacterial toxins such as CTx that depend at least in part on multivalent binding to their receptor to facilitate raft targeting, VacA need not form oligomers in order to partition into rafts [152] (Figure 3B). Furthermore, the ability of the toxin to form pores is not required for it to associate with rafts per se [152].

Why then does VacA associate with rafts? One potential answer is that this is linked to VacA’s internalization mechanism: VacA enters cells via clathrin-independent endocytic pathways, which are typically raft-dependent [153]. However, how rafts influence VacA’s pore-forming activity is not yet known. For example, it is currently unclear whether the structure of pores formed by VacA differs in raft versus non-raft environments. This is an especially important question because there are multiple examples of pore-forming toxins that associate with rafts [154]. Future studies using VacA should help to provide insights into this question, as well as to better delineate raft targeting mechanisms for this interesting class of toxins.

HIV selectively binds and fuses at raft/non-raft boundaries

Not just bacteria, but also viruses are known to target lipid rafts. One important example is the association of HIV, an enveloped RNA virus, with membrane domains [155, 156]. Rafts are thought to play a role in multiple steps in HIV assembly and release. For example, cholesterol is important for viral fusion and infection of cells by HIV [157]. Furthermore, the host cell receptor for HIV, receptor CD4, has been identified as a raft-associated protein [158]. However, until recently, the exact mechanisms by which the virus targets rafts for entry into cells has remained enigmatic.

In a series of interesting studies from both a membrane biology and virology standpoint, HIV has been shown to selectively bind and fuse to the interface between liquid ordered (L_o) and liquid disordered (L_d) domains [159–161]. Initial evidence in support of this idea came from studies showing that reconstitution of the fusion peptide (FP) of HIV gp41 into liposomes mimicking the composition of HIV viral membranes facilitates their fusion to supported bilayers consisting of mixtures of L_o and L_d domains [160]. Strikingly, liposomes containing HIV FP preferentially accumulated at the boundary between L_o and L_d domains. Further, both phase separation and cholesterol were found to be required to facilitate fusion. This behavior was specific to the HIV FP because liposomes containing the influenza FP showed no preference for the boundary [160]. HIV-1 psuedoviruses also preferentially bound to the domain boundary, demonstrating this behavior is not limited to the isolated FP [160].

An interesting question raised by these findings is why HIV virions prefer to fuse at domain boundaries. Both lipid-driven and protein-mediated factors have been shown to be important in this process. One contributing factor that promotes fusion is the hydrophobic mismatch and line tension at the L_o/L_d domain interface [161]. The phase preference of the HIV receptor CD4 and co-receptor CCR5 also contributes to this process [159]. Studies in GPMVs revealed CD4 partitions primarily into the raft phase, consistent with previous reports that it is a raft-associated protein. Intriguingly, CCR5 instead preferentially localizes to the raft/ non-raft domain boundaries [159]. The presence of CCR5 at the phase boundary appears to plays an important role in recruiting HIV to this site as evidenced by studies of the effects of the HIV entry inhibitors maraviroc and enfuvirtide on the recruitment of virions to the raft/ non-raft interface. Maraviroc is known to prevent binding of HIV to CCR5 [162], while enfuvirtide inhibits a conformational change in gp41 required for fusion [163]. Remarkably, maraviroc treatment prevents binding of virions to the phase boundary, but enfuvirtide does not, further supporting the notion that targeting of the virions to the L_o/L_d domain boundary requires binding to CCR5.

Taken together, these findings suggest a model in which HIV initially binds to CD4 in raft domains (Figure 4A) [159]. It subsequently binds to nearby CCR5 molecules located at domain boundaries. Exposure of the gp41 FP then occurs. The FP also prefers to bind to the raft/non-raft interface, resulting in fusion at this site.

Figure 4. Role of membrane domains in HIV-plasma membrane interaction.

(A) The host binding partners of HIV are the raft associated CD4 and raft/non-raft boundary associated co-receptor, CCR5. HIV binds initially to CD4 in rafts. It is subsequently associated with CCR5 and thus becomes enriched at the L_d/L_o interface, where the gp41 FP inserts to mediate fusion. (B) Larger domains have increased line tension, enabling more efficient fusion of HIV compared to the case of smaller domains with lower line tension. (C) T-cell activation, which is thought to occur prior to the entry of HIV into cells, is thought to cause clustering of raft domains. Linactants break up raft domains, making them a potential therapeutic option to inhibit HIV infection. In all panels, raft domains are shown in red and non-raft domains in blue. Figure is adapted from reference [159] (Creative Commons Attribution Noncommercial License). 2017, AAAS.

The finding that HIV preferentially binds and fuses at raft/non-raft domains not only provides insights into the mechanism of action of existing therapies but also suggests new approaches to drug development. One new line of investigation is the potential use of linactants to inhibit fusion (Figure 4B, C). These types of compounds are thought to partition preferentially at the domain boundary and decrease the line tension. In support of this idea, vitamin E (α-tocopherol), which has been reported to serve as a linactant [164], decreases HIV FP-mediated membrane fusion [161]. This finding also offers a potential explanation for why vitamin E suppresses latent HIV activation in vitro [165]. Another potential therapeutic approach is to shift CCR5 away from the domain interface. A related idea was tested earlier in a work that deliberately targeted CD4 to non-raft domains [166]. These types of interventions would be greatly aided by a better understanding of mechanisms responsible for targeting CCR5 to raft/ non-raft boundaries.

In addition to defining the role of rafts in HIV entry, these findings also have several broader implications. First, they identify CCR5 as one of the few known examples of membrane proteins that preferentially localize to raft/non-raft interfaces. To date, only a handful of proteins have been shown to target the raft/non-raft boundary, including Ras and Rac [167, 168]. Ras and Rac contain multiple lipid modifications that are thought to be important to target them to domain interfaces. How CCR5, a G-protein coupled receptor containing seven transmembrane domains, is targeted to the domain interface is currently unknown. Thus, CCR5 would be an ideal experimental model system to study how phase preference is determined for multi-pass transmembrane proteins. Second, they reveal that domain boundaries can function as a preferred site for viral fusion. Line tension and lipid packing defects appear to make this region of the membrane particularly vulnerable to attack. This does not appear to be a universal mechanism, as the FP of influenza [160] and VSV-G virions [160] do not prefer the phase boundary. Whether this mechanism is exploited by viruses other than HIV remains to be determined. Finally, they illustrate that the boundary between raft and non-raft domains is functionally relevant. This is important because, under physiological conditions where domains are predicted to be small, boundaries between domains would be predicted to be abundant.

Newly discovered sites of domain formation

Much of the work on membrane organization and spatial patterning mechanisms has focused on the plasma membrane of mammalian cells, a site where lipid rafts are thought to be especially prevalent. Given that membranes are ubiquitous in cells and the diversity of composition of plasma membrane between different organisms, the paucity of research in understanding membrane organization in non-mammalian systems is surprising. Here, we discuss two examples of newly described membrane domains found in very different places: large-scale domains in yeast vacuoles, and nanoscale domains in bacteria.

Large-scale domains in yeast

As discussed above, most recent studies suggest that raft-like domains are nanoscale in dimensions in live cell membranes. An interesting exception to the general rule of nanoscale domains comes from the observation that yeast vacuoles contain micron-sized membrane domains when the cells enter stationary phase [169]. Initial evidence for this type of phase separation came as early as the 1980’s [170]. These domains can be readily visualized by fluorescence microscopy and also have been detected by freeze fracture electron microscopy [171]. While their biological functions are not completely understood, initial reports suggest vacuolar membrane domains are important for lipophagy and microautophagy [171, 172].

The dramatic appearance of the vacuolar domains (Figure 5) is reminiscent of large-scale domains observed in model membrane systems exhibiting Lo-Ld phase separation often used to model rafts. Consistent with this possibility, several lines of evidence suggest the domains form in a manner consistent with a membrane demixing mechanism, including the observations that they segregate vacuolar membrane proteins and lipid probes [169, 171] and that depletion of the major yeast sterol ergosterol inhibits domain formation [169, 172]. However, until recently, whether they actually represent bona fide coexisting liquid-liquid phases has not been rigorously tested. Recent studies now show that these domains fulfill two major criteria of a membrane that separates into two liquid phases: i) the domains coalesce over time and ii) they undergo reversible mixing and demixing at a constant transition temperature [173]. Thus, these micron-scale domains in live yeast consist of co-existing fluid phases formed by a phase separation mechanism, similar to that observed in model membrane systems [173].

Figure 5. Scale of domains in cellular membranes.

Membrane domains span almost three orders in length. At the lower end of the scale, lipid domains found in bacteria are thought to be at the most tens of nanometers in length. The representation of bacterial domains was adapted from reference [177] (Creative Commons Attribution License). At the other end of spectrum, lipid domains reaching micron scale have been observed in yeast vacuoles. The yeast vacuole figure is reprinted from reference [173], Copyright (2017), with permission from Elsevier. For reference, typical immune receptor clusters are thought to be greater than tens of nanometers but smaller than the diffraction limit. The figure showing the BCR was taken from reference [17] and the figure showing a super-resolution image of LAT in T-cell was kindly provided by Dr. Dylan Owen and Dr. David Williamson.

These findings provide an important missing link between lipid-based domains in vitro and in vivo. It is not yet clear, however, why the domains in vacuolar membranes are so large while in other biological membranes domains tend to be much smaller in size. For instance, the role (or lack thereof) of actin in vacuolar membrane organization could be important in controlling the size of domains. Given that these domains occur in unperturbed cell membranes, in future studies, it would be interesting to use the yeast vacuole as a model to better understand the connection between membrane domain size, composition, and function.

Prokaryotic membrane domains

There is also now growing evidence that membrane domains, including raft-like domains, exist in prokaryotes [174, 175]. For example, the bacterium Borrelia burgdorferi, the causative agent of Lyme disease, contains raft-like membrane domains that contain cholesterol and cholesterol glycolipids [97, 99]. These domains can be isolated as detergent-resistant membrane fractions, and their formation requires sterols capable of forming ordered membrane domains. They can also be detected by FRET and transmission electron microscopy [97, 99]. There is also evidence that a bacterial homolog of a mammalian raft-associated membrane protein known as flotillin generates functional membrane domains in Staphylococcus aureus [176]. Disruption of these domains renders methicillin-resistant S. aureus (MRSA) susceptible to penicillin treatment [98]. These findings suggest bacterial domains are a potential therapeutic target.

Bacteria have also served as a testbed for the development of new methods to probe membrane organization in intact cells through a combination of metabolic labeling and neutron scattering approaches. Using the gram-positive bacterium Bacillus subtilis as a model, this approach was used to examine the overall structure of the bacterial membrane as well as to test for the presence of membrane domains [177]. A major advance put forward in this study was the use of a chemical biology-based approach to differentially label different cellular structures, thereby providing a contrast mechanism to separate out neutron scattering arising from the bacterial membranes. The bacterial membrane was shown to exhibit a characteristic lamellar structure expected for a membrane bilayer, with a hydrophobic acyl core thickness of ~24Å. Interestingly, this value is similar to that obtained from small-angle x-ray measurements of basolateral membranes purified from rat hepatocytes [178]. Even more remarkably, evidence of lateral demixing of lipids within the plane of the membrane over a length scale of 30–40 nm was observed (Figure 5). While the mechanisms underlying lipid demixing in this system are currently unknown, these results point to the presence of lipid domains in intact bacterial membranes.

Domains in bacterial systems are at the same time both complex (due to the presence of the cell wall) and simple (since there are no raft-targeting post-translational modifications such as palmitoylation). Further, most bacteria have neither cholesterol nor actin but instead possess analogs of both which are thought to be important in membrane organization [179, 180]. Thus, though rules governing bacterial rafts would have to be viewed differently compared to eukaryotic plasma membrane rafts, regulation by rafts could be evolutionarily ancient. Comparative analysis of the domains formed by these different systems would provide a comprehensive view of how lipid rafts are formed, how they function and how they have evolved.

Looking forward

The advent of newer, more sensitive experimental methods has made it possible to discern the structure and dynamics of membrane domains in biological membranes and probe the myriad physiological functions in which rafts play a role. Despite these advances, the field remains fraught with continued uncertainties about the nature of rafts and the exact mechanisms by which they control cellular functions [8]. In the final section of this review, we discuss several considerations and areas for future exploration that would advance our understanding of the role that rafts play in functionally patterning membranes.

With the continued development of many new experimental systems, it is important to bear in mind that different experimental methods report on distinct features of membranes. For instance, membrane order detected by spectral imaging is not necessarily linked to clustering of proteins in membranes reported by super resolution approaches. Further, different probes may also reflect different membrane properties. For the case of spectral measurements, the two commonly used dyes Laurdan and ANEPPDHQ do not actually measure packing of membranes the same way [181]. Thus, it is critical to probe membrane domains using a variety of experimental methods. By taking a multi-pronged approach, it is possible to develop a more nuanced understanding of the mosaic of data generated by different techniques.

It is also increasingly clear that cholesterol plays complex roles in biological membranes in addition to its proposed role in organizing rafts. A large number of proteins bind to cholesterol including PDZ domain-containing proteins [182, 183], and recent evidence suggests the transbilayer distribution of cholesterol is actively regulated [61]. It is thus not surprising that cholesterol depletion can have pleiotropic effects [184]. Despite this, much of the existing raft literature in cells relies on cholesterol depletion as a means to perturb raft-associated structures and events. In recent years, molecules that selectively tune membrane domains by either stabilizing or perturbing phase separation have been characterized [185–189]. However, to better differentiate between cholesterol-dependent and raft-dependent events, it would be helpful if new approaches to either perturb rafts or selectively move individual proteins in or out of domains were developed. The discovery of structural motifs that target single-pass transmembrane domain proteins to rafts [128] should enable the more selective testing of the role of raft association for specific proteins without disrupting rafts themselves.

The continued development of model membrane systems that recapitulate key features of biological membranes is also key to the advancement of the field. The properties of lipid domains have been extensively studied in well-controlled model systems such as the classic ternary lipid mixtures used to generate co-existing Lo and Ld domains [190, 191]. Over the past few years, efforts have been made to begin to incorporate increasing levels of complexity into these systems, such as the incorporation of actin cytoskeleton into model membranes to facilitate better understanding of the coupling between actin and domains [192–194]. This important goal merits further exploration given evidence that actin can influence the order of the outer plasma membrane leaflet, lipid domain formation, and the nanoclustering of GPI-anchored proteins in addition to its well-recognized role in compartmentalizing the plasma membrane via a picket fence mechanism (reviewed in [6, 195, 196]). Some efforts have also been made to develop asymmetric membrane models (reviewed in [195]). This also represents an area where further work is needed in order to enhance our understand how the lipid composition of each leaflet contributes to domain formation, as well as how coupling of domains between leaflets occurs [197].

More work also needs to be done to understand how the architecture of cell membranes regulates domain organization. One important goal in this direction is to continue to catalog the varied mechanisms that compartmentalize membranes into subdomains, generate protein nanoclusters, and make connections with other organelles [9, 198–200]. It is important to understand that in addition to lipids and membranous proteins, associated proteins also sculpt the membrane. For instance, both the physical basis and the consequence of coupling of actin and microtubules to plasma membrane order, while thought to be essential, are still not understood. Thus, while there is no single membrane clustering mechanism, a critical area of research is to understand the relationship between the different mechanisms of membrane organization. It is likely that the relative importance of each of these different processes is dependent on the probes and model systems that are being used for these studies.

Finally, in cells, besides compositional heterogeneity, cellular membranes have structural organization. Currently, the relationship between the complex topology and geometry of cellular membranes and lipid domains is not obvious. From a practical perspective, membrane topology impacts the interpretation of several microscopy-based approaches used to study the properties of domains [201]. There is considerable evidence in model systems to show that lipid phase separation is coupled to curvature in membranes [202], and geometrical constraints in membrane can alter diffusion and thereby protein sorting [203–208]. Thus, the interplay between the two could form the cornerstone of many membrane-associated biological processes [209].

Our understanding of organization of membranes arises from a patchwork of different experimental techniques performed on diverse model systems to probe a fragmented set of membrane properties using an incomplete set of molecular players. Despite this, we now have a framework in hand that has enabled us to apply principles of membrane organization to several biological systems. The next level of abstraction should start with the coordination between multiple systems and look at membrane organization holistically at different scales, both in space and time.