Twenty Questions

The participants of the Horizon Symposium on Twenty Questions
About the Participants
Download the pdf (1.5mb)

The discussions at this Horizon Symposium reflected the nature of the cell membrane — a dramatic and lively melting pot — and highlighted many of the recent successes and current challenges surrounding the field of membrane biology research. To capture the lively and informative discussions throughout the meeting, we asked the participants to respond to 20 of the most important questions that arose during the sessions, and here we present their replies.

SESSION 1: MEMBRANE COMPOSITION

SESSION 2: LIPID-PROTEIN INTERACTIONS

SESSION 3: HOW DO YOU BUILD A COMPARTMENT?

SESSION 4: TECHNOLOGIES — WHAT DO WE NEED TO KNOW AND CAN WE MEASURE IT?

SESSION 5: MEMBRANE SEGREGATION AT THE CELLULAR LEVEL

SESSION 6: SIGNAL TRANSDUCTION IN LIGHT OF MEMBRANE STRUCTURE

SESSION 7: MEMBRANE PATHOPHYSIOLOGY AND DISEASE

Session 1: Membrane composition

Back to list

Anthony G. Lee. It is all too easy to get bamboozled by the sheer complexity of the lipid composition of biological membranes. Two opposing conclusions could be drawn from this complexity. The first is that the composition of the membrane is finely balanced to produce a fully functional membrane, and that each species of lipid is present in the membrane for a distinct purpose. The other is that the complexity of lipid composition illustrates the relative unimportance of the lipid composition, so that many chemically distinct species of lipid can be used to produce a membrane that is functionally competent. The truth almost certainly lies between these two extremes. On the one hand, the costs involved in controlling the lipid composition of a membrane in which each and every species of lipid had its own distinct function would be prohibitive. On the other, the structures and functions of membrane proteins are certainly affected by the structures of the lipid molecules around them¹, and so nature has to get these structures reasonably correct. This is achieved, of course, by the normal processes of metabolic control in a cell; that is, by the selectivity shown by the enzymes involved in lipid synthesis and lipid turnover. That the exact lipid composition is not critical is illustrated by the fact that the fatty-acyl-chain composition of the membrane lipids can be changed significantly by changing diet, with no observable effects on function².

Ben de Kruijff. The lipid composition of membranes is most likely controlled by a combination of lipid-synthesis, -remodelling, -transport and -sorting processes. Each individual process involves regulation at the level of compositional control, as well as the control of gene expression. The major challenge is to identify the 'sensors' of lipid composition and to determine how they direct the activities of the proteins involved in determining composition³. However, it should be realized that composition is not very precisely regulated — considerable compositional differences can be brought about in a given membrane without producing a pronounced effect on function.

The question of why lipid composition is controlled relates to the specific functions of a given membrane. The most important function of a membrane is to act as a barrier. Lipids that can form thick, sturdy, impermeable bilayers are common to membranes that act as insulators (for instance, the myelin membrane around nerves⁴), or which have to withstand large mechanical forces for long periods, such as the membrane of the red blood cell⁴. Some lipid classes are required for specific functions and provide the proper conditions for membrane proteins to function optimally. For instance, the anionic membrane lipids are needed for the proper functioning of the bacterial protein translocase⁵. Others act as specific cofactors and are therefore essential for the activity of the proteins they bind to, such as the several mitochondrial proteins that are required by cardiolipin for proper activity. Specific cardiolipin-binding sites have now been identified in the high-resolution structure of membrane proteins⁶.

Michael A. Edidin. This question may have different answers depending on the membranes being considered. For example, poikilotherms (cold-blooded animals) and plants must regulate lipid composition to keep lipids above the gel-liquid-crystalline transition temperature. This is true for membrane lipids as well as storage lipids (triglycerides) and oils. Indeed, there are lots of data on changes in the composition of storage fats and oils as a function of growth temperature⁷. In general, lipids with lower melting points are found at colder growth temperatures, even in homeotherms, such as mammals, which tend to maintain a constant body temperature. In a classic experiment, for example, the melting point of back (storage) fat was compared between pigs kept outdoors in winter, pigs kept in a barn and pigs kept outdoors wrapped in sheep skin⁷. The results showed that membrane composition shifted to higher levels of unsaturated fatty acids, and so to lower melting points, in the colder animals.

More subtly, plants and animals also seem to regulate membrane lipid composition to maintain a near-constant membrane viscosity. Changing membrane lipid composition then reflects 'homeoviscous adaptation' (a phrase coined by Michael Sinensky in the 1970s)⁸. The idea of homeoviscous adaptation was developed on the basis of experiments on the lipid composition of bacterial membranes⁸, but has been observed in many other organisms⁹.

In principle, homeotherms maintain a constant body temperature, and so perhaps there is some other set-point for determining lipid composition. The physical and chemical properties of membrane lipids could be set to optimize the activity of membrane-spanning transporters (see REF. 10 for an example in yeast) or to regulate collisional coupling of membrane enzymes¹¹. There are certainly temperature differences between the body core and the surface and extremities of homeotherms, and one expects that, like storage lipids, cell-membrane lipid composition varies accordingly.

A range of mechanisms are available for modulating the lipid composition and physical properties of membranes. Levels of cholesterol synthesis affect membrane cholesterol content and physical properties, as does the balance between synthesis of saturated and unsaturated fatty acids. Changes in phospholipid composition can be affected by differential synthesis or by remodelling of previously synthesized molecules, such as exchanging fatty-acyl chains of one type for those of another¹². There are also mechanisms known (for protozoa) in which a fatty-acyl desaturase is activated by increasing membrane viscosity¹³. The activated enzyme then inserts double bonds into acyl chains of existing phospholipids until membrane viscosity is lowered.

Joshua Zimmerberg. It is abundantly clear that there are enzymes of lipid metabolism for which the activity of the enzyme is controlled by intracellular and extracellular activities (such as the control of phospholipase C by occupancy of G-protein-coupled receptors (GPCRs)^{14 , 15}). For these enzymes, it is then easy to see, for example, 'how' hormonal activation leads to a change in phosphatidylinositol lipid composition of plasma membranes. Because the new lipid products bind specific domains of proteins, and sometimes liberate cytosolic second messengers, it is also easy to see 'why' the composition is controlled in terms of their roles in cellular signalling pathways. For instance, in the activation of phospholipase C (PLC) at echinoderm egg fertilization, there is a massive synchronous production of diacylglycerol in the membrane¹⁶ at the same time that the cytosolic product inositol 1,4,5-trisphosphate (IP3; previously a lipid head group) binds to the IP3 receptor¹⁷ to open an intracellular calcium channel to release Ca²⁺. In mammalian egg fertilization, another PLC (PLCζ) moves from sperm to egg¹⁸ (FIG. 1; in addition, see Mark Terasaki's web page for video clips that depict the marked changes in membranes during fertilization).

		Figure 1 | Membrane compositional changes at fertilization. Figure 1 | The following is a composite of information from both mammalian and echinoderm species, although it is questionable whether all of these events occur in any one system. a | Upon the fusion of a sperm with an egg (see inset for large-scale view), there is a wave of lipid metabolism that leads to a wave of increased cytosolic free Ca2+, which has a number of effects: first, it initiates development of a new individual; second, it prevents polyspermy; and third, it protects the developing zygote from its environment. This is accomplished by phospholipase Cæ (PLCæ)moving from the sperm to the egg through a fusion pore. Next, this lipase converts phosphatidylinositol-4,5-bisphosphate (PIP2) to inositol-1,4,5-trisphosphate (IP3) and diacylglycerol (DAG). The IP3 diffuses to the neighbouring endoplasmic reticulum (ER), where it acts as a ligand for the IP3-sensitive receptor, which is a calcium channel. Upon binding, it opens up this channel (also known as the IP3 receptor (IP3R)), and releases intracellular stores of calcium. b | The ensuing increase in free Ca2+ concentration has two effects: it stimulates further endogenous phospholipase activity to propagate waves of calcium release, and it triggers the fusion of cortical vesicles (CV; membranes shown in green) that line the inner egg surface (bottom panel). The fusion of these vesicles increases the surface area of the egg about threefold, changes the membrane composition of the zygotic plasma membrane, and secretes (by exocytosis) components that raise, hydrate and crosslink the vitelline coat to form a strong barrier that provides the slow block to polyspermy and protects the developing embryo. Figure prepared by Joshua Zimmerberg.

Lipid composition is also controlled through relaxation of the gradients in lipids that normally exist between the inner and outer leaflets of the bilayer. For example, apoptotic and ageing cells express acidic phospholipids on their surfaces at a specific stage in their progression towards death. This exposure in turn allows scavenger cells to phagocytose apoptotic cells and the spleen to take up old red blood cells^{19 – 21}.

Another factor controlling lipid composition is the gradients of lipid length that are thought to occur throughout the secretory pathway, starting at the ER and ending at the plasma membrane. It is hypothesized by some that this gradient in membrane thickness facilitates the sorting of membrane proteins^{22 , 23}.

Yet another reason for cells to control membrane lipid composition is to control water and non-electrolyte permeability, which can vary by more than 100-fold due to lipid composition alone²⁴ (note that regulated changes in water permeability are dependent on the insertion of water channels, known as aquaporins, into the membrane)²⁵. For example, the water permeability of a fundulus egg is extremely low, which is required because the egg develops in waters of varying osmotic pressure²⁶.

Finally, we suggest that cellular membranes control and localize membrane lipid composition to regulate the assembly of specific macromolecular complexes²⁷.

Michael M. Kozlov. Lipid membranes in cells play a dual role. First, they serve as shells that physically separate the cytoplasm from the surrounding medium and intracellular organelles from the cytoplasm. Second, they represent a fluid matrix in which membrane proteins reside, interact and function. The lipid composition of a membrane bilayer in general, and of its domains in particular, controls the ability of the membrane to perform these two physiological tasks (BOX 1).

Membranes, in their role as a 'shell', mediate cell–cell interactions and determine the general mechanical stability of the cell boundary. Under specific circumstances, membrane lipid properties guarantee the local and transient membrane remodelling that results in membrane fusion (the merger of two membranes into one)^{28 , 29 , 30} and fission^{31 , 32} (the separation of one membrane into two), both of which are ubiquitous cellular processes (BOX 1). Membrane interactions are largely controlled by the electrical charge of the lipid bilayer surface³³. Enrichment of the outer membrane monolayer with acidic lipids that have negatively charged polar heads under physiological conditions, such as phosphatidylserine (PS) and phosphatidic acid (PA), provides a strong electrostatic repulsion between membranes and prevents their contact and aggregation^{34 , 35}. This repulsion should be especially important for membrane fragments depleted of membrane proteins with extracellular domains.

The balance between the mechanical stability of a membrane versus a tendency to undergo remodelling depends on the elastic properties of the membrane monolayers, and, specifically, the effective shapes of lipid molecules, which in turn are determined by the interplay between the molecular structures of lipids and the intramembrane lipid–lipid interactions. The presence in the outer membrane monolayer of lipids with a 'cone-like' effective molecular shape (so-called 'type II' lipids), such as phosphatidylethanolamines (PE) or diacylglycerols, facilitates the formation of non-bilayer intermediates of membrane fusion called fusion stalks³⁶. By contrast, lipids such as lysophosphatidylcholine (LPC) or lysophosphatidic acid (LPA), whose effective molecular shapes are inverted cones (so-called 'type I' lipids), inhibit membrane fusion if residing in the outer membrane monolayer³⁷. At the same time, lipids of the latter type have been predicted to facilitate membrane fission³⁸.

In their role as a matrix for hydrophobic and amphiphilic proteins, membranes can modulate protein conformational changes and determine protein motility in the plane of the membrane³⁹; this, in turn, is essential for the dynamics of forming protein domains and complexes that perform such functions as trans-membrane ion transport, and processes such as the formation of intracellular carriers, membrane fusion and others. The elasticity of membrane monolayers and the effective molecular shapes of lipids determine the energy cost of refolding membrane proteins by modulating the elastic energy of monolayer deformation, which is coupled to changes in protein configuration. This energy cost can be comparable to the energy of protein refolding, reaching tens of kcal M-¹. The motility of membrane proteins depends on membrane fluidity, which is modulated drastically by lipid monolayer transitions between the liquid-disordered, liquid ordered and crystalline phases. These transitions are largely determined by lipid composition, the most important factors being cholesterol content and the degree of saturation of the lipid chains^{40 , 41}.

Arlene Albert. Membrane lipids provide a matrix in which membrane proteins function. The variety of lipids found in biological membranes suggests that lipid composition must be important for the optimal functioning of the membrane. However, many proteins retain their native conformations and apparent functions in simple bilayers composed of a single phospholipid, or sometimes even in detergent micelles. Further examination is required to reconcile this apparent conflict, and more than one aspect of membrane function must be considered. On the most basic level, the membrane must provide a matrix in which the protein has sufficient freedom of motion to function. At the same time, the matrix must stabilize the protein against denaturation. These two as pects might not be optimized in a single-phospholipid environment. Furthermore, lipids can modulate the activity of membrane proteins. They accomplish this by interacting with membrane proteins in a complex manner, which includes direct interactions with the protein, as well as indirect interactions via the generalor 'bulk' properties ofthe bilayer. It is possible that different combinations of lipids can achieve similar effects. This is particularly true for indirect interactions. So, if particular fatty acids are not available to a cell, it can adapt its membranes, at least to some extent, according to the lipids available.

Lipids can be divided into two classes on the basis of their influence on membrane proteins: bulk lipids and annular lipids. First, consider bulk lipids, which do not interact directly with proteins, but provide a milieu in which the protein functions. The dynamic properties of these lipids are particularly important. Saturated fattyacyl chains can pack together efficiently and thereforere strict their dynamic properties. Conversely, unsaturated lipids can show considerable dynamics because of poor packing. The poor packing generates defects or 'free volume' within the hydrocarbon region. If a membrane protein undergoes a conformational change involving lateral expansion in the plane of the membrane bilayer, this 'free volume' can be recruited to accommodate the expansion. Membrane cholesterol reduces packing defects by becoming incorporated into the bilayer, essentially filling the partial free volume of the membrane.

Bulk lipid effects have been investigated for the GPCR rhodopsin. The influence of lipid dynamics on rhodopsin activation in reconstituted phosphatidyl-choline (PC)-rhodopsin vesicles demonstrates that systematic changes in cholesterol composition can influence the dynamics of the hydrocarbon region, which affects the formation of activated rhodopsin, R*(REFS 42–46). The activation of rhodopsin involves an expansion of the protein in the plane of the bilayer, which requires the recruitment of partial free-volume elements from the surrounding lipid bilayer. At high levels of cholesterol, this volume is essentially occupied by cholesterol and is unavailable to rhodopsin, soreceptor activation is inhibited. This illustrates that the formation of R* is exquisitely sensitive to the membrane lipid composition.

The activation of rhodopsin is also sensitive to the fatty-acid composition of the surrounding bilayer. Thecomposition of phospholipid hydrocarbon chains in reconstituted bilayers has been shown to regulate the formation of R*. In particular, docosahexaneoic acid (DHA) strongly promotes the formation of R* (REF. 47). This fattyacid is conserved in the retina, accounting for more than 50% of the phospholipid hydrocarbon chains in the disk membranes. Not only does this highly unsaturated lipid modulate rhodopsin function in the membrane, but its deprivation also degrades visual acuity as measured by electroretinograms^{48 , 49 , 50}. So the effects seen on the molecular level have consequences on retinal physiology.

Bulk lipids may also be important for stabilizing membrane proteins. It is likely that many membrane proteins exist in a kinetically stable, not a thermody-namically stable, conformation. This is in contrast to soluble proteins, which typically assume conformations that represent a thermodynamic minimum-energy state. If a membrane protein is kinetically stabilized, it will exhibit a half-life that would be influenced by the bilayer composition. For example, a rigid bilayer would inhibit denaturation of a protein that involves lateral expansion. This can be observed experimentally with rhodopsin, the melting temperature of which is higher in a saturated bilayer than in its native bilayer (A. Albert, unpublished data).

In addition to bulk effects, an annulus of lipid surrounds a membrane protein that directly interacts with it. The freedom of annular lipid motion is restricted compared with bulk lipids because the protein prevents full range of motion, much like a wall; bulk lipids, by contrast, can move in all directions within the plane of the bilayer. Little is known about the specific role of these lipids. However, it is tempting to speculate that they could provide a 'packing buffer' for the protein. Perhaps they have side chains that are capable of conforming to the irregular shape of the protein better than those of the bulk lipids. This would imply at least transient phase separations between bulk and annular lipids. Recent data suggest exactly this kind of lipid selection in the annulus next to rhodopsin, in which the unsaturated hydrocarbon chains of the lipid preferentially pack next to the protein⁵¹.

Particular lipids might form specific interactions with proteins. For example, ˜30 phospholipids for man annulus around rhodopsin. These annular lipids are in exchange with the remaining bulk lipids⁵², although ˜15 of them are in slower exchange⁵³. Again, little is understood about this particular category of lipid-protein interaction, although the lipids may play specific roles in protein function. For example, lipids that undergo slow exchange between the protein and the bulk bilayer may directly influence the light response of rhodopsin⁵⁴

It is clear that lipids can influence protein stability and function by modulating bulk lipid properties, through annular lipid effects or by specific protein-lipid interactions. Furthermore, different lipids impart different properties to the bilayer, and make different contributions to each of these regions. Although some alterations in lipid composition may be tolerated, the overall bilayer composition must be controlled to maintain optimal membrane protein stability and function.

Back to list

Sean Munro. At the level of a single cell, the overall lipid composition of membranes is controlled by regulating the amounts and activities of the enzymes that make and degrade the individual lipids. In addition, in multicellular organisms lipids are obtained from the diet or synthesized in particular tissues, and then transported through the organism for delivery to the endocytic compartments of other cell types.

Within a given cell type the constituent lipids are not distributed evenly between membranes. The mechanisms for establishing and maintaining this asymmetric distribution are beginning to berevealed⁵⁵. One contribution to this heterogeneous distribution may be that individual lipids are synthesized in distinct locations, such as the ER, mitochondria and Golgi complex. In addition, the vesicles and other membrane carriers that move between organelles may be selectively enriched or depleted for particular lipid species. Finally, lipid-transfer proteins mediate the non-vesicular movement of particular lipid species across the cytoplasm.

Ben de Kruijff. The asymmetrical distribution of lipids across some mammalian plasma membranes is a well-established fact and is generally considered to be highly important for membrane function⁴. It should first be noted that the lipid distribution across intracellular membranes in many eukaryotic organisms is largely unknown, and that most ideas on lipid asymmetry are based on mammalian plasma membranes such as the red blood cell⁴. Asymmetry is most likely the result of both active and passive mechanisms. ATP-dependent (energy requiring) lipid pumps play a role in determining asymmetry for the amino phospholipids, and might be involved in transmembrane transport of other lipids⁵⁵. Biogenic membranes — those in which phospholipid synthesis takes place — require special attention. The membrane of the ER, which is the main site of phospholipid synthesis in eukaryotic cells, is more-or-less symmetrical, despite the fact that all lipids are synthesized on the cystosolic leaflet of the membrane.

Transmembrane transport of phospholipids across artificial bilayers is usually very slow, and the consensus view is that the rapid transmembrane transport of lipids across biogenic membranes must be protein mediated. An attractive idea is that in biogenic membranes, the asymmetrical lipid synthesis and resulting imbalance in lipid packing across the membrane powers lipid translocation, which is mediated in a nonspecific manner by defects in the packing of the lipids around the helices of a subset of transmembrane proteins (the 'slip–pop' mechanism⁵⁶; FIG. 2).

		Figure 2 | The slippop mechanism for phospholipidtranslocation in biogenic membranes. Figure 2 | Schematic representation of a membrane, viewed from the side (a) and the top (b), with either a single membrane-spanning protein (green; left-hand side) or a multi-membrane-spanning protein (green; right-hand side) incorporated. The dynamic behaviour (part a; pale green silhouette sindicate motion) of single-membrane-spanning membrane proteins is postulated to cause transient defects in the lipid-helix interface more efficiently than a less dynamic multi-membrane-spanning protein. Part b shows that incorporation of proteins with a large cross-sectional area can offer a greater 'interaction surface' for phospholipids (grey; thickened line around the grey circle represents the interaction surface) than a single helix, resulting in more stable lipid–protein interactions. As a result of these two phenomena, small membrane-spanning proteins might be much more efficient in inducing phospholipid translocation than large membrane proteins. Figure prepared by Ben de Kruijff.

By contrast, lipid transport is much slower in the plasma membranes of eukaryotic cells, where no major lipid synthesis takes place. Moreover, lipid composition in the plasma membrane disfavours the slip–pop mechanism because of the presence of high concentrations of cholesterol.

Michael M. Kozlov. If a cell membrane consisted only of lipids and was in a state of thermodynamic equilibrium, lipids of different kinds would distribute symmetrically between the two monolayers. Any deviation from the symmetric distribution would relax by passive inter-monolayer lipid movement ('flip–flop') driven by the tendency of the system to reach a state of maximal entropy. The fact that in biological membranes the two lipid monolayers have different lipid compositions says that the membrane bilayers are not in an equilibrium state, and a persistent energy supply is therefore required to maintain the asymmetric lipid composition. An obvious source for the energy is membrane protein machines.

At least two mechanisms might contribute to the lipid asymmetry, both of which are based on membrane proteins that are orientated asymmetrically within the membrane (FIG. 3). The first is based on the action of special protein machines called flippases, which translocate lipids between the membrane monolayers⁵⁷. Putative flippases can be ATP independent or belong to various subfamilies of the P-type ATPases and ATP-binding cassette (ABC) lipid transporters. The asymmetric membrane composition produced by ATP-consuming flippases must represent a steady state in which the trans-monolayer lipid flux mediated by the flippases is compensated by the passive counter-flux guaranteed by the flip–flop mechanism.

		Figure 3 | Possible mechanisms of creation and maintenance of membrane lipid asymmetry. Figure 3 | a | Integral membrane proteins (transmembrane domain shown in dark blue) generate membrane bending, which results in positive curvature of the outer membrane monolayer and negative curvature of the inner membrane monolayer. Lipid molecules of different effective shapes re-partition between the two monolayers according to their spontaneous curvatures. The outer monolayer is enriched in inverted cone-like (type I) lipids (which have positive spontaneous curvature; shown in light blue), whereas the inner monolayer contains an increased concentration of the cone-like (type II) lipids (which have negative spontaneous curvature; shown in orange).b | An example of a nonspecific, long-range interaction between proteins and lipids is the electrostatic interaction. It has been proposed that clusters of positively charged residues on intracellular proteins attract acidic phospholipids (whose polar head groups are negatively charged at physiological pH) and concentrate them in the inner membrane monolayers. c | ATP-dependent flippases generate transmembrane transport of specific lipids and produce a steady-state asymmetry in membrane composition. Figure prepared by Michael M. Kozlov.

Another possible contributor to membrane lipid asymmetry may be a direct interaction of lipids with integral or membrane-bound proteins. Major membrane proteins are oriented in a certain way within the membrane. Direct interactions between lipids and different parts of proteins may result in lipid asymmetry. It is worth noting, however, that the energy of such interactions must be larger than ˜1 kBT˜0.6 kcal M^-1 to overcome the entropic effects, which favour symmetric trans-monolayer distribution of lipids.

Back to list

Anthony G. Lee. We need to understand which aspects of lipid composition are important and which are not. It is clear that variation in the head-group composition of lipids is more likely to be important than changes in lipid fatty-acyl-chain composition. We need to understand the range of fatty-acyl-chain compositions that are compatible with close-to-optimal functioning of the membrane and the processes that the cell uses to ensure that the fatty-acyl-chain composition remains within this range. An approach to the problem at the whole-cell level, which is particularly useful for bacteria and yeast, is to use genetics to knockout some particular species of lipid and then look for any associated changes in phenotype⁵⁸. An approach to the problem at the molecular level is to purify membrane proteins and reconstitute them into bilayers of defined composition, which allows a detailed study of the effect of lipid structure on membrane-protein function¹.

Tony Magee. Just accumulating comprehensive data on all the lipids (including cholesterol) in different kinds of cells, including levels of unsaturation, alkyl/acyl-chain variation and head-group variation (for glycol-ipids) will be tremendously valuable. Systematic mass spectrometry should be able to accomplish this. We also need to know the distribution of these species between different membranes in the cell (ER, Golgi, plasma membrane, ENDOSOMES, LYSOSOMES and so on) and also the trans-bilayer asymmetry. This will require good systematic methods for purifying subcellular membranes and analysing the two leaflets of the membrane. The former may be achievable with optimization of existing methods; the latter task is harder.

Kai Simons. The biggest barrier to understanding the lipidome is the lack of precise, quantitative and sensitive methodology to analyse lipids in cells. The big push forward is coming from the development of mass spectrometric methods for this purpose. Similarly to the proteome, the understanding of the lipidome will be driven by progress in mass spectrometric analysis. The National Institutes of Health awarded a grant of approximately US$30 million for a network called LIPID MAPS (LIPID Metabolites And Pathways Strategy; ). In Europe, we are pushing for funding in lipidomics as part of the 7th Framework Programme for European research funding. Only a concerted effort by networking will pave the way for the development of comprehensive methodologies for the analysis of lipids.

Back to list

Joshua Zimmerberg. What we need badly are more structures of the transmembrane domains of membrane proteins, preferably associated with lipids, so that we can begin to understand how the lipids sit with their neighbouring proteinaceous surfaces. In particular, we need more information about the nature of the forces that govern the microscopic configurational states of such lipids, and the identity of any actual lipid-binding sites within the transmembrane domain. It may also help to think of the interactions of lipids with proteins as a manifestation of the physics of 'wetting': the special physics that governs the wetting of a hydrophilic surface as water spreads over it may govern the spreading of a hydrophobic solvent over hydrophobic protein surfaces. Molecular dynamic simulations may prove extremely helpful in this regard.

Anthony G. Lee. The observation of lipid molecules in high-resolution structures of membrane proteins has already told us a lot about the ways in which lipid molecules can bind to membrane proteins (FIG. 4). But there are pitfalls with this approach. Often, only partial lipid molecules are identified in electron-density maps, which means that there is a real possibility of confusing lipid molecules and detergent molecules. There is also the problem that many of the lipid molecules reported in crystal structures have unusual conformations, raising the possibility of inadequate refinement⁵⁹.

		Figure 4 | Lipid-protein interactions in the membrane: insights from X-ray crystallography. Figure 4 | a | A crystal structure for the bacteriorhodopsin trimer shows a number of lipid molecules located around the hydrophobic circumference of the protein. The surfaces of the three protein molecules comprising the trimer are shown in yellow, green and blue. The lipid molecules resolved in the crystal structure are shown in space-fill format. In the native membrane the whole hydrophobic surface of the bacteriorhodopsin trimer would be covered by lipid molecules that form an annular or boundary shell around the protein (coordinates from Protein Data Bank file 1QHJ). The grey shaded region represents the width of the membrane. The lipid molecules shown here are likely to be more ordered than those in most membranes because bacteriorhodopsin forms a two-dimensional crystal with a low molar ratio of lipid to protein. b | The potassium channel KcsA opens only in the presence of anionic phospholipid. The crystal structure shows a lipid molecule bound at each protein–protein interface in the homotetrameric structure. This bound lipid molecule has been modelled as a simple diacylglycerol because the lipid head group is not resolved in the crystal structure. Two subunits (left and right) of the KcsA channel are shown here in a view in the plane of the membrane, with the lipid molecule bound between the subunits shown in space-fill format. The surfaces of the subunits have been coloured according to electrostatic potential. Figure prepared by Anthony G. Lee.

Tony Magee. Methods seem to be coming along for using novel detergents to obtain structures of membrane proteins (for an example, see REF. 60). This is an important advance (there are still relatively few solved structures of membrane proteins), but we are still a long way from obtaining structures in a real biological membrane. Membrane leaflet asymmetry could dramatically affect the conformation and activity of integral membrane proteins. For example, the inner leaflet of the plasma membrane may be more electronegative than the outer leaflet, which is highly enriched in oligosaccharide head groups of glycolipids, not to mention glycoproteins (although this varies between cells depending on the level of sialylation of these oligosaccharides). If it is too difficult to crystallize membrane proteins in asymmetric bilayers, then we should at least try to model the effects of asymmetric charge distribution and hydrogen bonding on protein structure. The lipid composition of the normal 'home' of the protein in question needs to be considered on a case-by-case basis. It will also be important to visualize how lipids that are covalently attached to proteins (such as fatty acids, prenyl groups, glycosyl phosphatidylinositol (GPI) ANCHORS and cholesterol) interact with membranes.

Anthony Watts. Some X-ray structures are now revealing lipids within and around the main protein back-bone^{61 , 62 , 63 , 64 , 65}. In some cases these are known to be derived from the originating membrane (for example, cardi-olipin with cytochrome c oxidase^{62 , 63} and diphytanyl lipids with bacteriorhodopsin⁶⁴). However, it is clear that in other cases, 'lipids' may actually be detergents, or have the incorrect structure ascribed to them in the modelling process to the electron density⁶⁶.

At one time it was suggested that all membrane proteins should be de-lipidated and all lipids totally removed before crystallization. Now, however, it seems that some lipids can stabilize protein structure and perhaps even facilitate crystallization (as, for example, in LHCII complexes^{67 , 68}, and the interaction of diphytonoelyl lipids with bacteriorhodopsin^{69 , 70 , 71 , 72 , 73}, bacterial reaction centres^{71 , 72} and photosystem I (REFS 74,75); these proteins are either less stable or do not crystallize in the absence of lipids). Other membrane proteins (especially transporters or receptors) are stabilized by ligands, some of which are lipophilic; for example, agonists or antagonists can stabilize receptors^{76 , 77 , 78}, and rhodopsin is stabilized by retinal^{75 , 79}. These stabilizing factors can also significantly facilitate protein handling.

From a functional perspective, it has long been recognized that lipids are required for the full or partial activity of many membrane proteins^{59 , 80 , 81 , 82 , 83}. These lipids need to be 'fluid', which implies some synergism between lipid and protein dynamics. X-ray structures provide rigid atom positions and therefore do not generate information on the relevant time scales (microsecond-picosecond) for understanding and describing such dynamics, although disorder is revealed in some cases. Energy minimization of lipid structures around protein crystal structures is now being attempted, which has given some insight into lipid-protein stoichiometry and the accommodation of lipids of different chemistry around the same protein interface⁵⁹.

Finally, the mere term 'bound' lipid may be misleading. In a crystal structure, a lipid is resident permanently within the complex. It was realized some time ago that spectroscopic methods with fast time windows, such as electron spin resonance⁸³ or fluorescence, may reveal 'bound' lipids, whereas slow spectroscopic methods, such as nuclear magnetic resonance (NMR), may detect quickly exchanging lipids⁸⁴. The crucial time-scale marker is the lipid-exchange rate, which is about 10⁷ sec-1 in a protein-free bilayer.

Without any doubt, lipids are essential for membrane function, and some specific and essential lipid-protein interactions (dependent on chain length, lipid type and so on) do exist, but a comprehensive description of lipid complexity in membranes still eludes us.

Akihiro Kusumi. Understanding the conformations of the lipids that are in direct contact with a protein is key to understanding the various subdomains in a membrane. In this sense, if the obtained X-ray structures of a membrane protein include the structures of lipids that are associated with the protein, it is a real bonus for us (for a review, see REF. 85). Such data would give us basic insights into how the lipids fit into the structure of the protein transmembrane domain; that is, how the amino-acid side chains that protrude from the α-helical transmembrane domain of the protein are accommodated by the neighbouring lipids. This would in turn provide a basis for understanding how the protein selects or deselects the lipids in proximity to it, which strongly influences its partitioning into different lipid domains. The surface of transmembrane α-helices are generally 'rough' due to protruding amino-acid side chains, and so the annular boundary region of lipids around the protein is not likely to accommodate cholesterol, which may be one of the reasons why many transmembrane proteins do not participate in the formation of stabilized LIPID RAFTS during signalling (signalling rafts are likely to contain high concentrations of cholesterol)⁸⁶. Nevertheless, some transmembrane proteins, notably T-cell receptors and the Fcη receptor, are thought to take advantage of membrane rafts for recruiting downstream effector molecules (kinases)^{86 , 87}. If the X-ray structures of these receptors could tell us whether, or how, they associate with cholesterol or reside in cholesterol-rich domains, it would be extremely interesting. Although we must be aware that interactions between proteins and lipids may be quite different in situ, X-ray structural data would provide a good starting point for a true understanding of the relationship between these two membrane components. Another important point to bear in mind is that the lipid binding seen in X-ray data would in fact be extremely dynamic in living cells, and these bound lipids may be exchanging with unbound lipids in the order of sub-microseconds⁸⁸.

Sean Munro. The structures of membrane proteins reveal the surfaces that are presented to the surrounding lipid bilayer (FIG. 5). The distribution of particular types of amino-acid residue on this surface reflects the types of interaction that occur between the protein and the surrounding lipids. For instance, hydrophobic residues are found in a 'belt' around the central section of the protein that faces the lipid acyl chains in the bilayer core. By contrast, aromatic and amphipathic residues are found where the protein is exposed to the 'interface' region that lies between the hydrophobic bilayer core and the surrounding water, and which contains the lipid backbones and head groups^{89 , 90}. The location of residues in the X-ray structure therefore provides a good idea of how the protein is sitting in the bilayer on average, although it does not reveal much about the dynamics of interactions with the lipids. In a small number of cases lipids are bound tightly to particular sites on proteins, and can even be seen in crystal structures⁵⁹. However, lipids around a protein are generally in rapid exchange with those in the rest of the bilayer. The dynamics of this exchange can be followed with spectroscopic methods, such as NMR or electron spin resonance (ESR), or investigated by computer modelling^{59 , 91}.

		Figure 5 | The structure of a membrane protein reflects the properties of the surrounding lipid bilayer. Figure 5 | The figure shows a cross-section of a membrane containing the bacterial ion channel KcsA. The X-ray structure of KcsA has been modelled in a bilayer of phospholipid palmitoleoyl phosphatidylcholine (POPC) by molecular dynamics simulation. The features typically found in the structures of membrane proteins are highlighted (left-hand column). The transmembrane domains (red) provide a girdle of hydrophobic amino-acid residues that face the lipid acyl chains in the core of the bilayer. On either side of this is a belt of aromatic and amphipathic amino-acid residues (typically tyrosine and tryptophan; green), which are well-suited to the 'interfacial' region of the bilayer, where the lipid back bones and head groups provide a transition from the hydrophobic core of the membrane to the aqueous solvent. Once fully exposed to the solvent, the protein surface mostly comprises polar and charged residues (blue). This figure is a static snapshot. In reality, biological membranes are fluid — lipids rapidly diffuse in the plane of the bilayer, and some of those adjacent to the protein are in continuous exchange with rest of the bilayer. In addition, the lipid acyl chains are flexible and dynamic, and can adapt to the surface of the protein and even stretch or compress to some extent to accommodate mismatch between the hydrophobic thickness of the protein and the thickness of the bilayer. In this example, the protein is slightly thinner than the lipid bilayer. However, there is a limit to the degree of mismatch that can be tolerated, and this might become important when proteins are moving between compartments or domains that have different lipid compositions and therefore thicknesses. Image created by Peter J. Bond, University of Oxford, using Visual Molecular Dynamics. Legend prepared by Sean Munro.

John Silvius. X-ray diffraction and other methods for high-resolution structure determination are best suited to revealing the structures of molecular complexes whose components all have well-defined, unique conformations. Diffraction methods have, for example, been very useful for revealing the means by which proteins recognize specific structural features of lipids, such as the binding of Pleckstrin homology (PH), Fab-1/ YGL023/Vps27/EEA1 (FYVE) or Phox homology (PX) domains to phosphatidylinositol phosphates^{92 , 93 , 94 , 95 , 96 , 97 , 98 , 99}. In a similar manner, high-resolution structures can help us to understand how some integral membrane proteins and membrane-associating proteins recognize lipids with specific hydrocarbon chains¹⁰⁰.

But what can high-resolution structures reveal about lipid-protein interactions of lower specificity? Lipids associated with the hydrophobic surfaces of integral membrane proteins are typically poorly imaged, which is consistent with previous conclusions from spectroscopy that most such lipids are not 'bound' in a unique conformation or with a high degree of hydrocarbon chain order. The latter conclusion is supported by data from the rare systems examined in which lipids associated with the bilayer-contacting surfaces of integral membrane proteins can be directly visualized (or modelled into the experimental density maps) with some confidence⁸⁵.

It is easy to see how this plasticity of interactions between membrane lipids and a protein's hydrophobic surface may favour dynamic aspects of membrane protein function — for example, by allowing at least modest rearrangements of protein tertiary structure within the membrane 'core'. A more intriguing question is how certain classes of lipids, such as sterols or the highly polyunsaturated phospholipids found in neural membranes, can modulate protein function in the absence of highly specific lipid-protein interactions. A related question is how integral membrane proteins discriminate between (or create?) distinct lipid environments within membranes that show lateral heterogeneity in their organization. High-resolution structures of membrane proteins, combined with molecular-dynamics simulations in realistic model bilayer environments, may provide important insights into such questions.

Lipid-protein interactions may compete with protein-protein interactions in cases in which membrane proteins form reversible protein-protein contacts, either intra- or intermolecularly, within the membrane interior. In such cases, the energetics of the competing interactions of lipids with these contact surfaces can affect how a membrane protein is poised physiologically between states of differing quaternary or tertiary structure. Determining the structures of the relevant protein contact surfaces may be useful for predicting whether, for example, different lipids can 'solvate' these surfaces with different affinities, and whether physiological changes in membrane lipid composition could therefore modulate protein activity by shifting the distribution of the protein between different conformational states.

Session 2: Lipid-protein interactions

Back to list

Joshua Zimmerberg. We have published extensively on experiments in which changing lipid composition alters membrane protein function in biological membranes¹⁰¹, on the fusion of cells expressing viral fusion proteins¹⁰² and on the formation of an apoptotic pore in membranes by B-cell lymphoma (BCL) protein family members¹⁰³. In all systems studied the curvature stress of membrane monolayers has been linearly additive in relation to composition, and so we alter the stress by changing membrane composition in a definable way. Moreover, one can estimate the spontaneous curvature of a monolayer using several methods, including X-ray diffraction. By adding lysolipids to membrane leaflets, we can generate leaflets with a more positive spontaneous curvature¹⁰⁴. Alternatively, we can neutralize this increased curvature stress by adding PE or long-chain fatty acids, as these induce negative spontaneous mono-layer curvature¹⁰⁵.

If the function of a particular protein involves bending membrane monolayers in a particular direction, then changing the spontaneous curvature of the mono-layer by changing lipid composition will make it easier or harder for the protein to function. Then, if we model the geometry and the elastic moduli of the system, we can estimate the change in free energies of transition for any particular change in lipid composition and compare models to experimental results to query mechanistic hypotheses. In our experiments, the initial intermediate of membrane fusion linking the two contacting mono-layers has overall negative spontaneous curvature (the stalk), whereas the pore in bilayer membranes that allows the transit of cytochrome c has overall positive spontaneous curvature (the apoptotic pore)¹⁰⁶.

Michael M. Kozlov. A lipid monolayer is characterized by two sets of bulk properties: those of an elastic film and those of a two-dimensional fluid, both of which influence different aspects of membrane protein function.

The determinants of the elastic properties of the membrane include the effective molecular shapes of the lipids, quantified by their spontaneous curvature^{107 , 108} and their molecular area¹⁰⁹; and the elasticity of the lipid monolayer, which is characterized by its resistance to the three major types of deformation — area stretching-compression¹¹⁰, bending¹⁰⁷, and tilt of the hydrocarbon chains with respect to the membrane plane^{111 , 112}. Conformational changes of integral membrane proteins whose permeability depends on the elastic stress, such as mechano-sensing ion channels, generate deformation of the lipid monolayer that surrounds the protein, and result in the accumulation or relaxation of elastic membrane energy. This variable energy contributes to the overall energy of the protein refolding, changing it significantly (see my response to question 1, page 4). As a result, the elastic properties of the lipid monolayer, and, specifically, the average spontaneous curvature related to variations of lipid composition in the vicinity of the protein, can modulate the energy of protein refolding considerably. Furthermore, the elastic properties can control membrane-mediated interactions between proteins that are inserted into the bilayer matrix and generate membrane deformations (see REF. 106 and references therein).

The main property of the membrane that characterizes it as a fluid is its two-dimensional fluidity. Lipid bilayer fluidity determines the effective protein motility in the membrane plane and must influence the dynam-ics of protein organization in the membrane plane. As mentioned in my response to question 1, the fluidity very much depends on the phase state of the monolayer lipids, which in turn depends on the lipid composition. At physiological temperatures, lipids with unsaturated hydrocarbon chains form liquid-disordered phases characterized by relatively fluid monolayer regions, whereas sphingolipids in combination with cholesterol form liquid-ordered phases, whose fluidity is several times lower than that of the liquid-disordered phases. By contrast, lipids with saturated chains form viscous gel phases in the physiological temperature range¹¹³.

Ben de Kruijff. This question requires consideration of both polar head groups of lipids and hydrophobic membrane interactions. Surface charge, hydrogen bonding and hydration are measurable bulk properties of lipids in bilayers that can be expected to affect protein function. For example, surface charge can control electrostatic interactions involved in anchoring proteins to membranes¹¹⁴, and head-group hydration creates a local micro-environment that could play a role in protein-mediated fusion of membranes. Furthermore, hydrogen bonding could be required for the formation of functional lipid domains, such as those formed by sphingomyelin and cholesterol.

When considering hydrophobic interactions, the situation is less clear. It is known that lipid packing is actively controlled in biological membranes and is required for protein function⁴. Fluidity, chain/packing order, membrane thickness and curvature stress are all measurable, but they are interrelated, which makes it impossible to separate their possible individual contributions to protein function. A new emerging concept is that of the lateral pressure profile of biomembranes, which could unify the different ways of describing lipid organization¹¹⁵. It describes the way in which pressure is dependent on the position in the bilayer. The pressure profile across the bilayer depends on both the lipid composition and the presence of other molecules, such as membrane-active alcohols that act as anaesthetics. It can explain the 'bilayer paradox' of biological membranes— namely, that lipids in biomembranes are present as a bilayer, yet every membrane contains substantial and regulated amounts of lipids that on their own prefer to organize into inverted, non-bilayer structures. The presence of such lipids in biomembranes will decrease the lateral pressure at the membrane-water interface, but will increase the pressure at the centre of the membranes. This could stabilize contacts between trans-membrane helices of proteins, thereby contributing to membrane protein stability, and could also facilitate interfacial conformational changes that are required for functions such as the gating of channels. Evidence that the lateral pressure profile across the membrane plays an important role in stabilizing the oligomeric functional structure of a bacterial potassium channel protein has been reported¹¹⁶.

Anthony G. Lee. There is no doubt that changing the chemical composition of the lipid bilayer surrounding an integral membrane protein affects its activity^{1 , 59}. The changes in lipid structure could be directly responsible for any observed changes in protein function; changing lipid head-group structure will lead to changes in hydrogen bonding and charge interactions between the lipid head groups and amino-acid residues in the protein, and changing the lipid fatty-acyl chain lengths will lead to conformational changes in the protein because of the requirement of minimizing hydrophobic mismatch between the protein and the lipid bilayer. Changes in lipid structure will also lead to changes in bulk properties of the lipid bilayer, such as fluidity, elastic energy stored in membrane curvature and the pressure profile within the membrane. It has been suggested that it is the changes in these bulk properties of the membrane, rather than the actual changes in chemical structure, that are responsible for changes in protein function^{117 , 118 , 119 , 120}.

It is very difficult to distinguish between these two possibilities, and often the two explanations are simply different ways of saying the same thing. However, there is a point of fundamental difference. If it is the structural properties of the lipids that are important, then the activity of a membrane protein will depend only on the structures of the lipid molecules in its immediate environment, and this will enable the proteins in the membrane to act as independent entities, which is a very desirable property in a system as complex as a biological membrane. On the other hand, if bulk properties of the membrane are important for function, then these bulk properties will serve to couple together the functions of all the proteins in the membrane, which is undesirable. For example, it has been suggested that stored elastic energy, a thermodynamic property of the whole membrane, can be used to increase the binding constant for extrinsic membrane proteins and can also be used to change the equilibrium between different conformational states of intrinsic membrane proteins⁵⁹; this means, for example, that binding of an extrinsic membrane protein will lead to a change in conformational state for proteins such as receptors in the membrane, which is a recipe for disaster. Perhaps intrinsic membrane proteins have evolved so that they are not sensitive to the kinds of change in membrane bulk properties to which they are likely to be exposed.

Arlene Albert. The measurable properties of bulk lipids have been studied extensively in defined lipid systems, and can be predicted from the properties of the constituent lipids. The effects of cholesterol serve as an example of this. Simple phospholipid bilayer systems have been used to show that cholesterol alters the properties of the bulk bilayer phase by interacting with phospholipids^{121 , 122}. Cholesterol also increases the bilayer thickness and decreases membrane permeability. Many years ago, it was also observed that cholesterol broadens the gel-liquid-crystal-phase transition temperature of phospholipid bilayers¹²³. It does this by disordering the packing of the phospholipid hydrocarbon acyl chains below the transition temperature, while increasing the ordered packing of these hydrocarbons above the transition. It has since been proposed that cholesterol is a primary modulator of the 'fluidity' of the hydrocarbon region of the bilayer, by affecting the motional freedom and packing of the hydrocarbon side chains.

The effect of changing lipid composition on a measurable bulk property such as partial free volume or bilayer thickness can be predicted from the known properties of the lipids. However, predicting the effect that this change would have on membrane proteins in general is problematic at best. Characteristics of the protein would have to be well understood, and functions of a diverse group of proteins will be sensitive to different lipid properties. For example, in the case of bilayer thickness as a bulk membrane property, a protein requires a match between the hydrophobic thickness of the bilayer and the corresponding hydrophobic surface of the transmembrane region. The protein can accommodate some changes in bilayer thickness. However, if the conformation of the protein becomes distorted, function will be compromised. The degree of tolerance will vary among proteins. Cholesterol is able to modulate the function of some membrane proteins through its impact on lipid packing. If increasing the saturation or the cholesterol content of the bilayer reduces the partial free volume of the membrane, protein conformational changes that require lateral expansion will be hindered. This may be used to modulate the activity of some proteins. For example, rhodopsin can be modulated in this way. In disk membranes with low cholesterol, rhodopsin is active; in high-cholesterol plasma membranes, rhodopsin is maintained in an inactive state¹²⁴. So, although the bulk properties clearly influence protein function, it is unlikely that a single measurable property will affect a diverse group of proteins in the same manner.

Sean Munro. One interesting bulk property is the thickness of the lipid bilayer. Integral membrane proteins present a hydrophobic 'belt' to the surrounding bilayer, and mismatch between the hydrophobic thickness of the protein and the lipid bilayer could perturb the properties of a protein or cause it to aggregate. Studies in model systems suggest that multi-spanning transporters or receptors can tolerate some degree of mismatch, presumably due to the stretching or compressing of lipids, but the activity of the proteins can be affected when the mismatch is too large^{81 , 85}. Correlating this with the situation in vivo is hampered by the difficulty of determining bilayer thickness in living cells. However, it is conceivable that if bilayer thickness increases through the secretory pathway¹²⁵, then this could be used to ensure that particular proteins are active only once they arrive at the plasma membrane.

Michael A. Edidin. If membranes are organized into domains (whether by lipid-lipid or by lipid-protein interactions), then the answer is that we cannot. However, we ought to step back and ask about the scale of the membrane properties that have been probed by different techniques. For example, the infamous phenomenon of membrane 'microviscosity' is sensed differently by electron paramagnetic resonance (EPR) probes, NMR probes, fluorescence probes and mechanical probes (such as membrane tethers)^{126 , 127}. We do not know the relevant scale for connecting protein function with these bulk properties. Indeed, the scale may depend on the spatial changes involved in a protein's function. Conformational changes could be accommodated by one level of lipid viscosity, whereas translation coupling of enzymes by lateral diffusion could require another. If domains with different properties coexist, then changing bulk properties may change the composition of some domains and not others. Hence, the functional consequences of changing membrane lipids may be sensed differently by different proteins.

Akihiro Kusumi. I do not believe that 'bulk' averaged properties of fundamental importance to measure actually exist. It is vital to realize that the membrane is a mosaic of various domains, with many layers of hierarchy. It is my opinion that the membrane allows variations in the composition of lipids and proteins (in time and space) by varying the relative amounts of different subdomains in which specific lipids and proteins carry out specific functions in an environment that they 'enjoy' or can endure. Such subdomains may be best characterized by the alkyl chain order (which is determined by the gauche-trans isomerization of alkyl chains), the conformational compatibility of the constituent molecules and/or the rate of gauche-trans conformational fluctuations. When the lipid composition changes, readjustment of the relative amounts of such subdomains would occur naturally as a result of the dynamic molecular interactions that continuously occur. Therefore, it might not be appropriate to say that the membrane allows variations of the composition of lipids and proteins (see above) — these dynamic interactions as a whole in fact form the membrane and keep it functioning. Therefore, if one wants to measure properties of the membrane environment, they must actually be local, rather than bulk, properties. The closest measurable properties to 'bulk' properties are, paradoxically, the relative amounts of each subdomain. This is somewhat analogous to determining the fractions of coexisting phases and the composition of each phase. However, this simplistic analogy stops here. As the number of membrane components is so great, I do not think the concept of phase is useful for biological membranes. Better concepts and words would be 'clusters' and 'microdomains' (perhaps nanodomains, depending on which hierarchy one is addressing), or words of similar meaning.

However, we have to remember that the residency time of each constituent molecule in a cluster may be limited, and molecules may be exchanging with those in other domains very rapidly, perhaps in the order of 1 μs. Assuming a relatively slow diffusion rate within a microenvironment of 1 μm² s^-1, a membrane molecule could cover an area of 20 lipids in 1 μs, or 20,000 lipids in 1 second (in a single leaflet of the bilayer). Such area sizes suggest that a membrane molecule could leave the microdomain (or the cluster) very quickly, although the microdomain size may vary substantially, depending on which level of the hierarchy it belongs to. Therefore, the second and third useful properties I would propose to measure in addition to the relative amounts of each subdomain are the size of the domain, and the residency time of each constituent molecule within the domain, or the lifetime of the domain.

To substantiate my statement above — that is, that the membrane is a mosaic of various domains, with many layers of hierarchy — I would like to give two examples.

First, by mapping the membrane based on its fluidity at an optical resolution of about 200 nm, Enrico Gratton's group showed that the plasma membrane consists of micrometre-sized domains with different fluidities¹²⁸. Second, we have already shown that the whole plasma membrane is partitioned into many small compartments of 30–200 nm in diameter (this size depends on the cell type), and that membrane molecules undergo short-term (1–10 ms) confined diffusion within these compartments and long-term hop diffusion between these compartments^{129 , 130} (BOX 2).

In addition, we need to bear in mind that about 15% of membrane proteins are bound to the membrane skeleton (the part of the cytoskeleton that is associated with the membrane, which mainly consists of actin filaments) at any given time, although this binding is transient and molecules continuously bind and dissociate. (Here, the term 'membrane skeleton' indicates the part of the cytoskeleton that is associated with or located very close to the cell membrane. It is an interfacial structure between the cell membrane and the cytoskeleton, which belongs to the cell membrane because it functions as an indispensable element for membrane function, but is also continuous with (and belongs to) the cytoskeleton (although the membrane skeleton has unique constituent molecules and a unique structure that enables it to interact with molecules integrated into the bilayer part of the cell membrane).) These transmembrane proteins are anchored to and lined up along the membrane skeleton (like pickets along a fence), which creates thin strips that form boundaries between the compartments, regions in which the diffusion rate is substantially reduced. Because such 'boundary domains' between compartments exist every 30–200 nm, they are at odds with the concept of a 'bulk' membrane domain¹³¹. As transmembrane proteins are generally conformationally incompatible with cholesterol, these rows of pickets may act as 'raft breakers'. Therefore, the existence of large raft domains (which are enriched in cholesterol) that encompass the compartment boundaries is unlikely in steady-state cells.

Although many membrane biologists are not aware of such partitioning of the cell membrane into compartments, dismissing the plasma membrane partitioning will profoundly skew one's view of the membrane. It is my view that if we acknowledge the existence of microdomains, clusters and membrane-skeleton-induced compartments, we must conclude that no fundamentally important measurable 'bulk' properties exist.

Back to list

Anthony Watts. On purely stoichiometric grounds, it seems unlikely that any one protein could significantly deplete any membrane of a particular lipid entirely, but limited preferential associations are well established¹³².

For example, rhodopsin resides in an unusually protein-dense membrane (70/1 lipid/protein molar ratio) and has an annulus of about 23 lipids. The protein could select out 23 of one type of lipid from the membrane, which is typically 30-50% C22:6n–3 chain composition, and ˜40% PE, 40% PC and ˜10% other lipid types (mole percentage of total phospholipid chains or types, respectively)^{133 , 134 , 135 , 136 , 137}. As selection of lipids to any location requires diffusion within the membrane, at any one time the interface of lipid with the receptor could statistically consist entirely of one (or a limited number) type of lipid (PE, for example) or chain length (such as C22:6n–3). But this would leave a small pool of different lipids with which to exchange. A snapshot would necessarily generate an average picture, possibly with a small number (maybe <5) of specific sites at a protein interface with which lipids have a higher affinity, sites which may of course be crucial to activity^{132 , 138}.

A more likely situation, but one that is still open to question, is the segregation of lipids with specific physical properties into regions of a membrane in which their solubility (and residency time) is higher (or lower)¹³⁹. Single proteins may be responsible for this phenomenon, but it is more likely that complexes or other interactions might result in a heterogenous lipid and protein distribution in the membrane.

Anthony G. Lee. In general, intrinsic membrane proteins have little capacity to select specific membrane lipids from their environment. Most lipid-binding sites on the hydrophobic surfaces of membrane proteins (known as boundary or annular sites) show very little structural selectivity⁸⁵. However, some intrinsic membrane proteins also bind lipid molecules at specific sites, often between transmembrane α-helices at protein-protein interfaces in multi-subunit membrane proteins. A good example is provided by the potassium channel KcsA, which requires anionic lipid to function and whose crystal structure reveals an anionic lipid molecule bound at each of the four protein-protein interfaces in the tetrameric structure¹⁴⁰. A consequence of the lack of structural specificity of the annular sites is that the lipid molecules immediately surrounding an intrinsic membrane protein will have a composition very similar to the bulk composition of the membrane, and therefore the bulk lipid composition of the membrane has to be suitable to support close-to-optimal function for all the proteins in the membrane.

Erwin London. We now know that some membrane proteins bind a very small number of lipids at specific sites within clefts, much like they would bind any other ligand. However, most of the lipid around the hydrophobic surface of a protein still seems to have only a limited degree of contact with the protein, and lipid binding to these sites is not likely to be highly specific¹. In terms of the hydrophobic regions of lipids, the ability of lipids with different degrees of acyl-chain unsaturation to make close contact with the protein surface and generate strong van der Waals interactions is something that should be examined in the future. The strength of cholesterol interactions with the protein surface relative to those of other lipids should also be examined more carefully. With regard to the specificity of integral membrane proteins for the polar head groups of lipids, electrostatic (Coulombic) interactions have been studied in the past, and often seem to have a significant, if not huge, effect¹. By contrast, for peripheral proteins electrostatic interactions can be the main driving force for binding to lipids¹⁴¹. The ability of certain integral membrane proteins to recognize lipids with specific complex head groups, such as gangliosides, may constitute another major source of selectivity. Again, this may involve a binding site for a single lipid molecule. In this case, the binding site would be formed by parts of the protein outside the hydrophobic part of the bilayer.

Akihiro Kusumi. Many membrane proteins and lipids, notably those thought to be involved in raft domains, may mutually select each other to form small membrane domains or clusters that may be necessary for their functions. However, many membrane proteins are unlikely to be very fussy about their neighbours, and these proteins may be partitioned into both raft and non-raft domains. Overall, the capacity of membrane proteins to select or deplete specific membrane lipids is widespread, but is carried out with varying degrees of effectiveness. If one looks at the published data carefully, many proteins that are thought to be raft-associating in the sense that they are recovered in detergent-resistant membrane (DRM) fractions (which biochemically define lipid-raft association) often have comparable amounts of the protein partitioned into both DRM and non-DRM fractions.

For some membrane proteins, such as GPI-anchored receptor molecules, many researchers have found that the capacity to associate with cholesterol and lipids with saturated alkyl chains was crucial for their functions. Upon ligand binding, GPI-anchored receptor molecules tend to form clusters that lead to the formation of 'receptor cluster' rafts in which cholesterol and saturated lipid molecules are concentrated. This eventually induces the recruitment (albeit transiently) of downstream signalling molecules that are able to associate with such stabilized receptor-cluster rafts.

However, I am a bit worried about the way this question was posed. It sounds as if it is being assumed that the interaction of membrane proteins with surrounding lipids is long lasting. Even in the case of 'raft preferring' protein molecules, the lipids that are in direct contact with the protein may be rapidly exchanging with bulk lipids or those in other domains. This could occur on a timescale much shorter than a millisecond, as was shown by Kawasaki and colleagues for hemagglutinin in the influenza viral envelope membrane¹⁴².

Stuart McLaughlin. Richard Anderson and Ken Jacobson have argued that proteins might nucleate lipid-raft formation¹⁴³. Proteins may be particularly important in forming rafts on the inner leaflet of the plasma membrane — work by John Silvius and others has shown that rafts do not form spontaneously in bilayers formed from a lipid mixture that corresponds to the inner leaflet. So, how might proteins attract a 'shell' of lipids enriched in cholesterol? One clue is that phosphatidylinositol-4,5-bisphosphate (PIP2) is often associated with non-caveolar rafts¹⁴⁴. PIP2 is an unlikely raft component: it has a polyunsaturated arachidonate chain, which causes it to be excluded from a conventional sphingomyelin/cholesterol-enriched raft¹⁴⁵. But peripheral and integral membrane proteins thought to be associated with rafts often contain a cluster of basic/hydrophobic residues that can laterally sequester PIP2. This sequestration occurs because the cluster of basic residues produces a local positive electrostatic potential that attracts polyvalent acidic lipids such as PIP2 (REF. 146). The basic/ hydrophobic cluster of the MARCKS protein contains 13 basic and 5 phenylalanine residues. Recent structural studies from the laboratories of Steven Smith and David Cafiso show that these five hydrophobic Phe residues penetrate to the level of the hydrocarbon fatty-acid tails in the interior of the membrane (see figure 1 of REF. 145 and references therein). This implies that several adjacent residues are pulled into the polar head-group region of the bilayer, which produces a local increase in pressure^{115 , 116} that attracts lipids with a small, polar head group, such as cholesterol. Pico Caroni and colleagues have provided evidence that MARCKS in biological membranes is associated with rafts in some manner¹⁴⁶.

The possible role of basic/hydrophobic clusters in contributing to the formation of rafts is not established. By contrast, there is good evidence from a number of physical measurements (such as fluorescence resonance energy transfer (FRET)) that the basic/hydrophobic clusters can laterally sequester PIP2 by a non-specific electrostatic mechanism, even when the membrane contains physiological levels of the monovalent acidic lipid phosphatidylserine¹⁴⁵.

Either high affinity (K d ˜ 10 nM) binding of calcium/calmodulin (Ca/CaM) to, or protein kinase C (PKC) phosphorylation of, three serine residues within the MARCKS basic effector domain can reverse binding of this domain to the membrane; this releases the laterally sequestered PIP2 (REF. 146). Indeed, cell biology experiments from the laboratories of Pico Caroni and Michael Sheetz, and biophysical experiments on simpler systems, suggest that one function of the natively unfolded MARCKS protein may be to reversibly sequester the important lipid PIP2.

Back to list

Erwin London. So far, the biophysical experiments on the sequence-dependence of both helix position in membranes and helix-helix interactions have generated no serious conflicts with biological data.

Akihiro Kusumi. Polar residues are likely to have key roles in determining the correct folding and maintenance of membrane protein structures. Just as the folding of water-soluble proteins is driven by the strong tendency of hydrophobic amino-acid residues to be excluded from water, so polar residues drive the folding of membrane proteins. They do this by forming ionic and hydrogen-bonding interactions in the core of the protein structure, thereby excluding themselves from the hydrophobic environment that surrounds the surface of the protein (see REF. 147 for an example).

Gunnar von Heijne. The relevance of biophysical studies of peptide-lipid interactions for investigating the in vivo process of membrane protein assembly has long been unclear. In a general sense, many of the conserved characteristics seen in the transmembrane helices in high-resolution structures of membrane proteins are mirrored in the biophysical studies¹⁴⁸. However, it is only very recently that detailed, quantitative studies of the insertion of transmembrane helices in vivo have been undertaken; the first results suggest that membrane insertion depends on protein-lipid interactions very similar to those addressed by the biophysical studies¹⁴⁹.

As far as is known today, associations between trans-membrane helices — both intra- and intermolecular — can be driven by different kinds of interactions: van der Waals interactions between close-packed surfaces, weak Cα-H hydrogen bonds, and strong hydrogen bonds between polar amino acids¹⁵⁰.

Ben de Kruijff. The insertion of proteins into membranes is usually studied by cell biologists using in vivo or sophisticated in vitro systems and biological materials. Biophysicists often study this process using model membranes that contain synthetic transmembrane pep-tides. One way to bridge this gap is to use specifically designed, plasmid-encoded transmembrane peptides that are co-translationally inserted in model membranes¹⁵¹. For investigating protein-mediated insertion processes, the model system in general should be more complex and would involve reconstitution of the protein components essential for insertion, such as the YidC protein. In some cases, single-lipid vesicles can still be used to faithfully mimic the assembly process; this is true even for more complex processes, such as the assembly of the oligomeric protein KcsA, for example¹⁵². The challenge will be to select proteins that can be considered to be good models for specific assembly processes, such as the potassium channel KcsA for tetramer assembly¹⁵² and glycophorin for dimer formation¹⁵³. Testing such proteins both in the biological system and in simple model systems should provide the desired insights into the assembly processes.

Session 3: How do you build a compartment?

Back to list

Michael M. Kozlov. Taking into account that lipid composition may considerably influence the functions of membrane proteins (see my response to question 5, page 12), the role of membrane compartmentalization may be to produce the lipid environment necessary for effective functioning of specific proteins.

Michael J. Caplan. The division of membranes into sub-compartments can be appreciated over several size scales, and each of these scales is associated with a distinct set of biological processes within the cell. The partitioning of the plasma membranes of polarized epithelial cells into distinct apical and basolateral domains lies at the more macroscopic end of this spectrum, whereas the segregation of membrane components into rafts can probably be considered to be among the most microscopic manifestations of this phenomenon (FIG. 6). From a physiological perspective, the necessity for epithelial polarity is perhaps best exemplified by the need to carry out vectorial transport of solutes and fluid, often against steep concentration gradients. In order to perform net unidirectional transport, epithelial cells must endow their apical and basolateral surfaces with distinct cohorts of transport proteins that can work in series to drive transepithelial fluxes¹⁵⁴. The smaller and less morphologically distinct variety of membrane subdomains, which share the designation 'raft', probably exist to create regions in which membrane components can be concentrated in two-dimensional space in order to facilitate cell biological processes¹⁵⁵. Such membrane subdomains seem to play a prominent role, for example, in the generation of biochemically and spatially defined signals. The involvement of membrane subdomains in the production of local signals is apparent at the immunological synapse, a specialized membrane region that is crucial in the process of antigen presentation to T lymphocytes. The recruitment of distinct protein components into this subdomain correlates strongly with the nature of the resulting immunological response¹⁵⁶. Similarly, processes such as membrane trafficking clearly depend on the formation of membrane subdomains that group cargo proteins with a common destination together with the cellular machinery that will mediate their sorting and targeting¹⁵⁷.

		Figure 6 | Macroscopic and microscopic compartmentalization of the plasma membranes of polarized epithelial cells. Figure 6 | The distinct physiological functions of the apical and basolateral domains of an epithelial cell's plasma membrane require that they express completely different complements of membrane proteins (apical proteins: green and red; basolateral proteins: blue, purple and pink). In order to generate this macroscopic anisotropy, epithelial cells must possess sorting pathways and machinery that ensure that newly synthesized membrane proteins arrive at and are retained within the appropriate cell-surface domain. Proteins internalized from the plasma membrane by endocytosis must also be subject to similar sorting processes that coordinate their recycling to the proper surface domain, or diversion to a degradative pathway. This sorting can occur within several subcellular structures and can involve both direct and indirect pathways (arrows). Within the plane of the apical and basolateral membranes, it has been proposed that many proteins are further segregated into microscopic sub-domains (boxes) that modulate their potential to participate in signalling and trafficking processes. ARE, apical recycling endosome; ASE, apical sorting endosome; BSE, basolateral sorting endosome; CRE, common recycling endosome; ER, endoplasmic reticulum; LE, late endosome; LYS, lysosome; TGN, trans-Golgi network. Figure prepared by Michael J. Caplan.

Lewis C. Cantley. In theory, compartmentalization of proteins to specific sub-compartments of the plasma membrane can accelerate signal transduction events that require protein–protein or protein–lipid interactions. In addition, compartmentalization could enhance polarizing events in the cell, such as directional motility, by facilitating local positive-feedback loops. The most extreme and well-defined compartmentalization is the division of epithelial cells into lumenal and basolateral compartments, for which it is very clear that proteins can be delivered to and retained in specific regions¹⁵⁸. It is more difficult to visualize microdomains in nonpolarized live cells, and better techniques are needed for this purpose before firm conclusions can be made about the importance of these domains in cell function.

Judith Klumperman. The 'classical' role of membrane compartmentalization is to sequester specific cellular functions within defined environments. This creates optimal conditions to carry out specialized tasks and protects the cell from harmful substances, such as lysosomal enzymes. In addition, compartmentalization may also be an important tool in organizing cytosolic signalling events. Association of signalling molecules with specific membrane compartments or vesicular carriers might be a necessary factor to provide kinetics and directionality to the propagation of signals into the cell. This is well illustrated in neurons, in which signals from axon terminals to the cell body must travel over long distances^{159 , 160 , 161}; such transport may involve the use of specialized endosomes. A second feature of compartmentalized signalling is that by creating a specialized environment, one molecule may initiate different signalling pathways in different compartments, such as plasma membrane versus endosomal signalling^{159 , 160 , 161 , 162}. For example, upon activation with epidermal growth factor (EGF), the downstream signalling molecule growth-factor-receptor-bound protein 2 (GRB2) preferentially associates with the plasma membrane pool of EGF-receptor (EGFR), whereas epidermal growth-factor receptor pathway substrate 8 (EPS8) associates with the endosomal EGFR pool¹⁶³.

Boris N. Kholodenko. The localization of signalling proteins to various internal cell membranes and different membrane micro-environments, including lipid rafts, modulates their signalling outputs^{164 , 165 , 166}. clathrin -coated pits, caveolae , focal adhesions and other membrane compartments all contain specific sets of receptors and proteins that regulate signal transduction. Endocytic pathways, which involve clathrin-coated pits and caveolae, have different dynamics and involve different molecular factors. The functional implications of membrane compartmentalization for intracellular signalling are much more intricate than previously thought.

For example, in protein phosphorylation cascades, such as mitogen-activated protein kinase (MAPK) cascades, the plasma membrane confinement of the input kinase activity and the cytosolic distribution of phosphatase activities could result in unfavourable phospho–protein gradients¹⁶⁸. At distances greater than several micrometres from the plasma membrane, the phosphorylation signal would decrease to sub-threshold levels, unless there were additional mechanisms other than slow protein diffusion to spread a signal through these cascades. In the past, receptor endocytosis was generally believed to be a mechanism to attenuate signalling. However, recent compelling evidence suggests that traf-ficking of endocytic vesicles directly contributes to signal propagation from cell-surface receptors to the nucleus^{169 , 170 , 171}. Endocytosis emerges as a robust, immediate signal transducer on a short time scale, and as a downregulator of receptor signalling on a longer time scale. Distinct endocytic compartments, including Rab domains, rafts and caveolar domains, can deliver differential sets of signalling proteins to diverse cellular targets, thereby generating specific signalling outputs.

Vivek Malhotra. Membranes provide a convenient surface for a number of biochemical reactions. Take, for example, the ER, which has two recognizable domains: the rough and the smooth ER. The rough ER is studded with ribosomes and concentrated with components of the protein-translocation machinery. It is therefore involved in the synthesis and translocation of signal-sequence-containing proteins¹⁷². In the smooth ER, a separate set of protein components regulates the exit of the newly synthesized, signal-sequence-containing proteins from the ER^{172 , 173}. How these components are concentrated and segregated in these domains is not known. However, this segregation is likely to be essential for the efficient translocation of newly synthesized proteins into the ER and their eventual packaging and transport out of the ER from the smooth region.

Compartmentalization of a similar kind is also likely to be important at the trans-Golgi network (TGN). The TGN has at least four major exit routes. Proteins are transported out of the TGN to the endosomes, the plasma membrane and (probably) back to the ER, and in certain cells are also packaged into secretory storage granules¹⁷⁴. Protein kinase D (PKD), which is involved in the formation of transport carriers destined for the cell surface, is localized to one end of the TGN. Clathrin-coated vesicles and transport vesicles with different destinations are also probably formed from highly localized and distinct sites of the TGN.

In addition to the examples of compartmentalization described above for the ER and TGN, a clear segregation in the binding of Rab5 and Rab to the contiguous membrane of the endosome is observed on exogenous expression of these proteins¹⁷⁵. These two Rabs recruit different effectors, which in turn are involved in the transport of specific cargo from endosomes to their respective cellular destinations. This compartmentalization is likely to be necessary for the efficient sorting and transport of cargo, but how it is established and maintained (similar to ER compartmentalization) remains unknown.

Jennifer Lippincott-Schwartz. Membrane compartments are a fundamental feature of eukaryotic cells, without which there would be no energy-transduction flow, no secretory or uptake activities, and no nucleus. There are several reasons why membrane compartments play such diverse and crucial roles inside cells. First, they have surfaces that can provide platforms for organizing chemical reactions that are highly evolvable. These surfaces have embedded lipids and proteins with exposed hydrophobic and hydrophilic domains that are either shielded in the bilayer or extend from the bilayer. This organization imparts an electrostatic potential to the membrane that can be used to drive chemical reactions (for example, oxidative phosphorylation and photosynthesis, which couple the transport of H+ to the synthesis of ATP). Second, because compartments create enclosed areas that are separate from the cytosol they can readily become functionally specialized. An example is the lumen of the ER, which provides an oxidizing environment that is suitable for disulphide-bond formation — a key characteristic of the tertiary structure of cell-surface receptors. Third, membrane compartments have surfaces with which cytosolic proteins can associate and polymerize into coats that impart shape changes, such as the budding of the membrane into a vesicle. This provides a way to transfer material between compartments within the cell.

These properties are exploited in numerous ways by cells. One important way is through the generation of an elaborate endomembrane system, comprising the ER, Golgi complex, endosomes and lysosomes; the advantages of this include increased capacity, quality control and regulation of biosynthetic processes. The substantially increased surface area of the membrane generated by compartmentalization (for example, the surface area of the ER in some cell types is 38 times that of the plasma membrane) increases the capacity for biosynthetic processes, such as protein translocation, assembly and processing. By translocating the proteins into a compartment (the ER) rather than the plasma membrane, these proteins are not immediately lost by the cell but can be retained and modified. This provides the opportunity for quality control. A protein can be modified through glycosylation, remodelled through the formation of disulphide bonds, or re-routed for degradation if it is not assembled properly. This has allowed the evolution of very intricate secretory and multi-component membrane complexes (such as the T-cell antigen receptor) whose abundance can be modulated in response to need. Compartmentalization also allows secretory and membrane proteins to be stored until they are needed, at which point they can be rapidly delivered to selected regions of the cell by exocytosis . This allows the secretion and surface expression of proteins to be temporally and spatially regulated.

Finally, by allowing cells to have precise control of secretory and membrane protein biogenesis, compartmentalization has played a crucial role in the evolution of multicellular organisms. This is because the fitness of these organisms fundamentally depends on the ways that groups of cells interact, communicate and function as units. Such communication and interactions are crucially dependent on secreted and cell-surface proteins, the expression of which at the right time and place must be carefully regulated. Because multicellular organisms place the highest premium on precise control of the biogenesis of secretory and membrane proteins, their evolution has been linked to that of the endomembrane system.

Suzanne Pfeffer. In the secretory pathway, secretory and membrane protein biosynthesis is segregated from later steps to ensure that misfolded or partially assembled protein complexes are not exported from the cell, and can be recognized instead for degradation. So only properly folded and assembled proteins are transported to the Golgi complex for oligosaccharide processing and subsequent packaging into distinct transport vesicles¹⁷⁶.

Proteins destined for different cellular destinations (for example, apical versus basolateral plasma membrane, endosomes and lysosomes) are segregated and sorted into different vesicles. Domains within the plasma membrane can be considered to be different compartments because they contain distinct sets of proteins and are physically segregated; their content depends on their specific functionality. For example, in the gut, membrane transport proteins on the apical surface of intestinal cells import disaccharides from the intestine, whereas baso-lateral receptors interface with the blood^{177 , 178}.

Membrane compartmentalization therefore segregates different functions. In the endocytic pathway, compartments become more acidic and as a result have greater degradative activity^{179 , 180}. Lysosomal enzymes are most active and most concentrated in lysosomes, which are the most acidic organelles and are specialized for degrading cellular components. Obviously, segregation of degradative enzymes is essential for cell viability. Earlier endocytic compartments are organized into sub-domains that sort proteins for either recycling or degradation, or into domains that exist to maintain the compartment; they act to receive cargo or fuse with another compartment. The proteins that mediate distinct functions such as organelle fusion or cargo receipt are organized into subdomains of a given compartment¹⁸¹.

Sandra L. Schmid. Membrane compartmentalization is required to segregate different and sometimes conflicting biochemical activities/functions within the cell, such as proteolysis in lysosomes and oxidation reactions in peroxisomes and mitochondria. These diverse biochemical activities might require specialized chemical environments (with distinct pH, divalenthation concentration, membrane potential and so on) that are at odds with other biochemical reactions. Membrane-bound compartments therefore allow the generation and maintenance (through integral membrane pumps and transporters) of a distinct ionic milieu within the lumen of the organelle. Lateral segregation along a membrane allows membrane domains with distinct functions to be generated in which proteins involved in a common function (such as vesicle formation and sorting, signal transduction or cell adhesion) are grouped together for increased efficiency.

Kai Simons. Membrane compartmentalization is a dynamic device to create membrane microdomains for functions in signalling, membrane trafficking and cell polarization. The beauty of the lipid-raft mechanism is that rafts can be clustered by specific protein–protein interactions or other means to form specific platforms. There are significant advantages in terms of enhancing or reducing bimolecular reactions through the dynamic compartmentalization of the lipid bilayer by raft clustering. In addition, activation or inactivation of membrane proteins could result from changed lipid environments. The propensity of cell membranes to compartmentalize by raft clustering introduces a new organizational principle into cell architecture beyond organellar specialization; this feature of the bilayer allows dynamic segregation of activities in and out of liquid-ordered matrices⁴¹. A corollary of raft formation is that membrane proteins have probably evolved to adapt to functions within or outside rafts. How the lipid diversity in cells and the membrane proteins have been structured to serve these purposes is an area of research still in its infancy. Obviously much work remains to be done before we will understand how proteins interact with lipids in different ways.

Ritva Tikkanen. I would suggest posing an additional question: how do we define a compartment? We tend to use the presence or absence of marker molecules, proteins or lipids to define a compartment (one that 'includes this, excludes that'). However, these classifications are mostly based on captured frozen images, snap-shots that do not reflect the dynamic nature of the cell. Therefore, we tend to make oversimplifications just to make it all nicely fit our models.

Back to list

Jennifer Lippincott-Schwartz. This is a central question in cell biology and has generated great interest and much controversy among scientists over the years. As there are many types of compartment within cells — including the ER, Golgi, endosomes, lysosomes, plasma membrane, secretory granules, lipid droplets, nuclear envelope, peroxisomes, mitochondria and chloroplasts and — there are many ways in which compartments are generated and regulated.

It is generally recognized that mitochondria, chloro-plasts, the ER and the plasma membrane are generated from pre-existing copies and that the cell cannot synthesize these membranes without a template membrane to work from. The generation of these compartments is therefore intimately tied to the mechanisms that control the protein translocation and import, protein binding/dissociation and lipid biogenesis that occurs on their surfaces. All other organelles in cells are generated by the outgrowth and differentiation of membranes derived from the ER and plasma membrane.

The Golgi complex is generated by outgrowth from the ER, whereas lysosomes and endosomes are derived by internalization and differentiation of the plasma membrane. As a consequence, the location, size and fate of these compartments is intimately tied to the proper functioning of secretory and endocytic pathways. For example, when secretory transport out of the ER is inhibited the lipid and transmembrane resident components of the Golgi are absorbed back into the ER and the Golgi disappears from cells¹⁸². The nuclear envelope, which is an extension of the ER that wraps around chromosomes, is studded with nuclear pores. Lipid droplets bud out of the outer leaflet of ER membranes. Finally, peroxisomes are now thought to differentiate initially out from the ER¹⁸³. Although biogenesis of these organelles is dependent on the ER or plasma membrane, it also involves the recruitment of specialized sets of peripheral proteins from the cytosol. The peripheral proteins (including kinases, lipases and small GTPases) play significant roles in further differentiating the compartments. These proteins can alter the lipid composition of the compartment or recruit other molecules such as motor proteins that can cause the compartment to move and change its shape.

Sean Munro. Some compartments, such as the plasma membrane and ER, essentially persist indefinitely, although their constituent components are turned over. Other compartments, such as early endosomes and the cis-cisternae of the Golgi complex, are apparently continuously formed by the fusion of specific vesicles from other compartments^{184 , 185}. Once formed, they mature into later endosomal or Golgi compartments, while new early compartments are formed to replace them. These processes are almost certainly subject to control, but the basis of this regulation is in most cases unclear. What has emerged in recent years is that most compartments express an 'identity' in the form of a unique set of small GTPases of the Rab and Arf families, and specific lipid species, often phosphatidylinositol phosphates (PIPs)¹⁸⁶. These serve as landmarks for the recruitment from the cytosol of specific peripheral membrane proteins, such as vesicle coats, motor proteins and tethering factors, which control compartment organization. The enzymes that generate the active forms of the GTPases and the specific PIPs are themselves distributed in a compartment-specific manner; understanding their localization and regulation may be a route to answering the question of how compartmentalization is generated and regulated.

Scott D. Emr. During the past 5 years it has become clear that phosphatidylinositol (PI) lipids form part of a complex system for defining compartmental identity by the selective expression of specific PI lipids on the cytoplasmic surface of certain organelles in eukaryotic cells^{187 , 188} (FIG. 7). The PI lipids function as spatial and temporal regulators of membrane trafficking and organelle function in the secretory and endocytic pathways. Differential phosphorylation of the PI lipids at the 3, 4 or 5 position of the inositol ring in the lipid head group generates a family of distinct stereoisomers. The basic compartmental code that has been characterized so far is composed of pools of specific PIs that are enriched in compartments of the secretory and endocytic pathways. At present, this code is defined by phophatidylinos-itol-3-phosphate (PI3P) on membranes of the endosomal system, phosphatidylinositol-3,5-bisphosphate (PI3,5P2) on the lysosome, phosphatidylinositol-4-phosphate (PI4P) on the Golgi complex and phosphatidyli-nositol-4,5-bisphosphate (PI4,5P2) on the plasma membrane^{189 , 190}. Each of these PI lipids is generated by the action of specific lipid kinases. Compartment-specific localization of PI kinases leads to restricted synthesis/localization of PIPs. The organelle-restricted PIPs then determine the transport activity of the membrane by recruiting and activating specific effector proteins and because they contain domains (such as PH, FYVE, PX or epsin N-terminal homology (ENTH) domains) that directly bind the head group of the PI lipids in a stereoisomer-specific manner^{191 , 192}. A set of PI-specific phosphatases then terminate PIP signalling and inactivate PIPs at inappropriate membrane sites. Together, the selective localization and regulation of the PI lipid kinases and phosphatases seems to restrict the PIPs to their appropriate compartments.

		Figure 7 | Synthesis of phosphatidylinositol phosphate isoforms on membrane compartments programs organelle identity. Figure 7 | a | Phosphatidylinositol (PI) kinases differentially phosphorylate PI to generate seven phosphatidylinositol phosphate isoforms (PIPs) that recruit or regulate downstream PIP-binding effector proteins. PI phosphatases turnover PIPs and limit PIP signalling events on membranes or at inappropriate membranes. b | PIP isoforms are spatially compartmentalized on membranes and contribute to organelle identity. Phophatidylinositol-4-phosphate (PI4P) is generated at the Golgi membrane and is essential for secretion. PI4P is converted to phosphatidylinositol-4,5-bisphosphate (PI4,5P2) at the plasma membrane, where it functions in the control of the actin cytoskeleton and endocytosis. The majority of phophatidylinositol-3-phosphate (PI3P) is generated in the endosomal system. PI3P is converted to phosphatidylinositol-3,5-bisphosphate (PI3,5P2), which functions in the regulation of lysosomal membrane homeostasis. c | PIPs, along with small GTPases such as Rabs, regulate downstream effector proteins that direct membrane trafficking, such as transport factors and fusion machinery (coats, adaptors, SNAREs and tethers). PI kinases and Rabs participate in a positive regulatory feedback loop that reinforces and amplifies these organelle identity tags. The PIPs seem to serve a more global role as organelle identity tags that are distributed over an entire organelle system (for example, PI3P is present on both early and late endosomes), whereas the Rabs function as more specific tags (for example, Rab5 associates with early endosomes and Rab7 associates with late endosomes), allowing the cell to precisely discriminate between the different compartments. Figure prepared by Scott D. Emr.

The Rab GTPases (and other small GTPases such as Arfs) define a second component of this organelle identity code: the combination of a specific PI lipid and a specific Rab on the cytoplasmic surface of an organelle defines the array of effector proteins (such as components of the vesicle docking/tethering machinery, as well as vesicle coat proteins) that will bind and regulate the transport activity of each organelle^{186 , 193}. The Rabs, PI lipids and their effectors, together with the membrane-anchored, compartment-specific SNARE proteins, define the basis of a complex combinatorial organelle-identity code. These molecules permit cells to form and maintain the highly compartmentalized organization of the secretory and endocytic membrane systems.

Suzanne Pfeffer. Compartmentalization is initially generated by the interaction of particular components that act together in a specific functional pathway. So proteins may bind other proteins and lipids, and a coordinated set of protein–protein and protein–lipid interactions leads to the generation of membrane specialization or a microdomain. These microdomains have distinct functional capacities because they contain distinct sets of proteins and lipids. Microdomains need to communicate with each other — cargo entering a microdomain may need to be handed on to the next domain, which carries the cargo to another site. For example, in early endosomes cargo may be sorted into a tubule that will recycle the protein back to the cell surface. Organelles need to maintain the capacity to communicate with one another, and proteins will therefore exist to link microdomains to ensure that both types of microdomain exist within a single membrane-bound compartment.

Little is known at present about domain regulation; however, early data using small interfering RNA (siRNA) strongly suggest that a cycle of interplay between effectors and Rab GTPases is important for the stability of specific microdomains¹⁹⁴.

Sandra L. Schmid. Membrane-bound compartments are generated by the selective sorting, targeting and retention of the membrane proteins/enzymes that functionally define each compartment. Although some intracellular compartments (such as mitochondria and the ER) are fairly autonomous, others (such as endosomes and the Golgi complex) reflect a steady-state flux of proteins into and out of a compartment along a path-way of interconnected intracellular organelles. There is growing evidence that membrane compartments are defined, at least in part, by small Ras-related GTPases — the Rab proteins — that regulate the sorting and targeting machinery, and which function by generating positive-feedback loops that bring compartment-defining molecules, lipids and proteins together to define a specific organelle composition^{181 , 195 , 196}. This is best established for Rab4 and Rab5 in the case of the early endosomal compartments^{197 , 198}. Activated Rab⁵ recruits phosphatidylinositol-3-kinase (PI3K), which generates PI3P, which recruits early endosome antigen 1 (EEA1), which recruits more Rab5. The relevant effectors of other organelle-specific Rab proteins and their function in establishing organelle identity are less well defined. The idea that positive- and negative-feedback loops create a unique balance of protein–protein and protein–lipid interactions, and define composition at steady-state, is likely to be generally applicable to the generation and regulation of organelle identity.

Vivek Malhotra. It is clear that cellular membranes are compartmentalized, but how this is achieved is not known; for example, do Rabs and PKD (see my response to question 8, page 21) bind to pre-determined and segregated domains, or does binding of one prevent recruitment of the other to the same domain? Are these domains permanent, or is their formation regulated depending on cellular needs? It is possible that the arrival of cargo into endosomes acts as a signal for the recruitment of all the other downstream components that are required for the efficient packaging and transport of the respective cargo. This hypothesis is testable. For example, would overexpression of cargo that requires Rab5, in the absence of cargo for Rab9, prevent the recruitment of Rab9 and its effectors to the endosomes? Such an investigation could lead to an understanding of whether compartmentalization is a permanent feature or is regulated by cargo-dependent cellular needs.

Kai Simons. Membrane compartmentalization into lipid rafts is based on their propensity to cluster. This clustering can be generated by protein–protein crosslinking or by the generation of clustering lipids, such as ceramide¹⁹⁹. It is important to note that raft clustering is a specific process that generally includes only a small fraction of all potential lipid-raft components present in a membrane. A cell membrane can therefore simultaneously have different raft clusters that function in separate cellular processes²⁰⁰. The association and dissociation of raft proteins is controlled by the normal regulatory machinery in cells, such as reversible phosphorylation or other post-translational modifications, such as palmitoylation²⁰¹.

Back to list

Vivek Malhotra. This question is closely related to question 9 (page 25). We know very little in general about the significance of cargo in initiating membrane traffic. It is therefore difficult to assess the significance of cargo in the generation of a compartment. One could argue that cells that secrete more proteins (such as algae) contain a highly developed Golgi complex. The Golgi stacks in these cells contain as many as 30–40 cisternae, compared with 4–6 in mammalian cells^{202 , 203}. Algae secrete proteins with extensive carbohydrate modifications, and in order to carry out these cargo-specific modifications efficiently the Golgi stack is extended.

Judith Klumperman. Cargo can travel through pre-existing compartments yet also induce the generation of a compartment. For example, overexpression of a lysosomal protein can induce the formation of a larger number of lysosomes (J. Klumperman, unpublished observations). Recent studies indicate that cargo can also induce the formation of specialized endosomes. For example, EGF travels to an endosomal compartment bearing Rab5 and APPL1 and -2 — two Rab5 effectors. In response, APPL1 translocates from endosomes to the nucleus where it interacts with an established regulator of chromatin structure and gene expression²⁰⁴. Together this suggests a picture in which cargo is much more than just a passive traveller.

Jennifer Lippincott-Schwartz. I believe that cargo plays an active role in the biogenesis of compartments generated by outgrowth and differentiation from the ER or plasma membrane — that is, the Golgi complex, endosomes and lysosomes. After all, cargo processing and transport are two of the major functions of these compartments (and presumably underlie their evolution), so it makes sense that these functions are linked to how the compartment is generated and maintained within a cell. What evidence exists to support this? With regard to the Golgi, time-lapse sequences following a bolus of secretory cargo released from the ER upon a temperature shift have revealed that as cargo passes through the Golgi complex, the complex distends and expands its shape²⁰⁵. This is also observed when there is a block or slowing of cargo efflux out of the Golgi induced by a temperature reduction or by expression of mutants that block Golgi export^{206 , 207}. Cargo can therefore dramatically affect the size of the Golgi.

More recent evidence suggests that cargo can modulate the activity of coat proteins, whose binding to the ER or Golgi membranes underlies sorting and transport between these compartments²⁰⁸. This has been proposed to occur by the binding of cargo proteins (such as SNAREs) to GTPase-activating proteins (GAPs)^{209 , 210}. This would then stimulate the hydrolysis of SAR1 or ARF1, the major GTPases regulating coat activity in the ER and Golgi, respectively. Consistent with this possibility, we have recently found in live-cell-imaging experiments that increased amounts of ARF-GAP1 are recruited onto membranes when a bolus of VSVG cargo moves through the Golgi²¹¹.

Session 4: Technologies — what do we need to know and can we measure it?

Back to list

Jennifer Lippincott-Schwartz. Light-level imaging techniques are one of the most promising approaches for investigating membrane dynamics and organization because they can reveal when and where genetically or biochemically modified proteins, signals or processes are formed, transformed and then consumed in space and time (see Table 1 for a summary of imaging tools and technologies available for investigating membrane biology). This was initially made possible by the discovery of genetically encoded fluorescent proteins such as green fluorescent protein (GFP), and then by the development of commercially available instruments for performing kinetic techniques, such as time-lapse imaging, fluorescence recovery after photobleaching (FRAP), FRET, photoactivation and fluorescence correlation spectroscopy (FCS). Using GFP chimeras and these kinetic imaging techniques, we can now address numerous topics related to membrane organization, including determining the size and dynamics of rafts, protein–lipid and protein–protein interactions within membranes, and binding/dissociation reactions. This is possible because these techniques allow specific populations of molecules within a cell to be highlighted. This permits the kinetic properties of a protein to be determined (for example, whether the protein is freely diffusing, bound to an immobile scaffold or undergoing binding and dissociation with other molecules), which cannot be obtained by viewing cells alone.

One highlighting technique that has recently become available is photoactivation, which works by converting molecules to a fluorescent state by using a brief pulse of high-intensity irradiation. After fluorescently highlighting specific populations of molecules using this method, the fluorescent proteins can be followed as they re-equilibrate in the cell. The extent and rate of re-equilibration can be quantified and used to describe the kinetic parameters of a protein, such as its diffusion rate, compartmental residency time and speed of degradation. Currently, three such photoactivatable fluorescent proteins are available: photoactivatable GFP: the Kaede protein; and kindling fluorescent protein 1 (KFP1)²¹². To further expand the available applications of photoactivatable proteins, newer versions of these proteins need to be developed, including ones with different spectra that allow multiple protein species to be highlighted and followed simultaneously within cells, or which can be used with FRET-based techniques to study protein–protein interactions in greater detail.

An advantage of fluorescence-based imaging approaches is that they can be used in different combinations to produce new imaging strategies. This includes FRAP in combination with either two-photon microscopy or total internal reflection microscopy (TIRM) for studying events at specific sites in a cell, and photoactivation combined with FRET techniques for studying protein–protein interactions among subsets of proteins²¹³.

Although kinetic imaging techniques such as photoactivation and FRAP are extremely powerful, they nevertheless have drawbacks. Foremost is their spatial resolutions of >200 nm. This means that they cannot resolve biological machines (for example, ribosomes or proteasomes) or signalling domains on membranes, which have dimensions of ˜5–200 nm.

A new optical imaging technique that is pushing the limits of temporal and spatial detection is stimulated emission depletion (STED) microscopy, which breaks the diffraction limit by suppressing spontaneous emission at the periphery of a diffraction-limited fluorescent spot by stimulated emission. With this microscope system, it is possible to resolve cellular structures of ˜100 nm in size (ref. 214).

To obtain finer spatial resolutions, researchers will need to rely on other microscopic techniques, such as conventional electron microscopy, atomic force microscopy (AFM) and near-field scanning optical microscopy (NSOM), which have 30–100-nm resolution. Like fluorescence microscopy, AFM and NSOM are capable of non-destructive continuous imaging of living specimens, but at a much higher resolution. Their usefulness has been limited, however, by their inability to obtain molecular specificity and resolution over short time-frames. Conventional electron microscopy has the desired molecular specificity and spatial resolution, but requires fixed material. Molecular specificity by electron microscopy has been achieved using antibody labelling or genetically encoded labels, such as horseradish perox-idase (HRP) attached to a protein of interest (for example, the tetracystein-ReAsH system²¹⁵), or immuno-gold particle labelling. Immuno-gold labelling can distinguish two or more distinct molecular species simultaneously by using differentsized gold particles, but the low percentage of antigens that are labelled by this technique makes detailed analysis of protein binding sites and interactions difficult to measure.

Nicoletta Kahya. Optical imaging techniques allow for direct visualization of biological processes with a fairly high temporal resolution (from milliseconds up to seconds). In many instances, non-invasive molecular tracking in real time unravels crucial details in the live cell, tissue or even entire organism, provided that a suitable labelling strategy ensures chemical specificity. However, the high degree of heterogeneity in membranes also calls for high spatial resolution, frequently between the micrometre and nanometre scales — that is, beyond the diffraction limit of visible light. It is therefore imperative to direct our efforts towards improving the spatial resolution while maintaining high temporal resolution. Within the context of direct visualization, single-particle tracking (SPT) provides both high spatial (tens of nanometres) and temporal (down to fractions of milliseconds) resolution²¹⁶.

A promising alternative is NSOM, which uses the contrast of light microscopy (mainly fluorescence) to achieve tenfold better spatial resolution than conventional optical imaging²¹⁷. Optical fibres with a small aperture (a few tens of nanometres) are brought close to the sample surface and illuminate it. Key to the success of this technique is the great illumination intensity and accurate control of the vertical distance of the fibre from the sample surface. However, we need further advances to overcome the technical limitations associated with imaging live cells, such as the poor temporal resolution (in the order of seconds), sensitivity and the difficulty of capturing the details of the rough cell membrane surface in physiological liquids with nanometre spatial resolution. Furthermore, a strong limitation to both the above-mentioned techniques is the labelling strategy. The fluorophore has to emit enough photons over the acquisition time to ensure a good signal-to-noise ratio. For SPT, the highest temporal and spatial resolutions have only been achieved with 40-nm beads, which are prone to artefactual aggregation; no single dyes are efficient enough to offer a satisfying performance¹²⁹. The same problem potentially holds for NSOM.

The development of new labelling strategies, such as the use of quantum dots^{218 , 219}, will greatly contribute to progress in this area. Alternatively, some techniques, such as steady-state and time-resolved FRET microscopy²²⁰, and fluorescence cross-correlation spectroscopy (FCCS)^{221 , 222}, intrinsically achieve spatial resolutions beyond the diffraction limit, in addition to being non-invasive and capable of high temporal resolution. While waiting for improvements in materials technology, the enormous potential of such techniques should be promoted and exploited.

Patricia Bassereau. Fluorescence techniques (one-or multi-photon) have been particularly successful for imaging living cells, provided that a specific protein or lipid is labelled beforehand²²³. Such techniques have been essential for revealing the dynamics of the cell. The limited spatial resolution and the difficulty of deriving direct quantitative concentration measurements from fluorescence intensity are the main limitations of these techniques. In some cases, the bleaching of fluorophores can also be an issue, as undesirable oxidative processes can be generated. For example, we have shown recently that bleaching of fluorophores can induce phase separation in model membranes due to oxidation of a substantial fraction of cholesterol²²⁴.

FCS and FRAP have better spatial resolutions, and quantitative analysis of diffusion coefficients can be derived; however, the interpretation of the data they generate may depend on the geometry of the membranes, which is difficult to determine. The diffusion coefficient can be derived exactly from FCS or from fluorescence recovery in the case of standard two-dimensional conditions²²⁵, but cell membranes often have a more complex geometry, which can result in an incorrect derivation. In some cases, the contribution of the shape of the membrane to the measurement has been theoretically calculated²²⁶, but this does not often apply to practical situations²²⁷. Some other potential artefacts have now been identified (such as photo-bleaching and probe aggregation)²²⁸, and specific simu-lations need to be developed (see the example of the brush border membrane²²⁹).

I believe that recent developments in X-ray microscopy, or X-ray spectro-microscopy (in particular the reduction of focused beams to sub-micrometre sizes), are very encouraging, and that by using these techniques it will be possible to obtain better spatial resolution in the absence of labelling. At present, spatial resolution about 100 times better than optical microscopy can be achieved²³⁰. Internal structures of cells can be observed (see, for example, ref. 231). X-ray micro-spectroscopy in the multi-keV range offers unique possibilities for studying the biological roles of metals ions and their distribution in bacteria and cells^{232 , 233}.

Judith Klumperman. These days a wealth of new techniques have become available for the imaging of cells and individual molecules. The more classical microscopy techniques involve light- and electron-based imaging. Important progress in this field has been the development of live-cell imaging to visualize protein transport in living cells^{213 , 223 , 234}. However, these light-microscopy-based techniques provide no structural information below the 200–500-nm scale — the level of organization at which many sorting events take place^{235 , 236}. Electron microscopy is required to relate protein localization to cellular ultrastructure and to define protein distributions over subdomains of continuous membranes^{236 , 237 , 238}. However, electron microscopy does not provide dynamic information. To fill the gap between live-cell imaging and electron microscopy, and to combine the strongest qualities of each approach, correlative-microscopy methods are currently being developed. Here, a fluorescent spot of proteins can be imaged in living cells and at any time be prepared for immuno-electron microscopy to visualize the very same spot at high resolution^{239 , 240 , 241}. These correlative microscopy techniques literally bridge the gap between micro- and nanometre scales.

Daniel Axelrod. The advantages and limitations of imaging techniques are as closely paired as the two sides of a coin, and several such 'coins' can be considered. First, the ability to image intrinsically clear and colourless structures (such as cells cultured on a coverslip) by use of sophisticated optical and/or spectroscopic approaches is clearly an advantage in making the 'invisible' become 'visible'. But such approaches might not be as specific as desired, because they are often based on differences of refractive index between an object and its surroundings, rather than on any biochemical properties. Such approaches might also perturb the sample (for example, through the introduction of extrinsic probes or the induction of photodamage).

Another advantage of imaging techniques is the ability to detect small objects, even those as small as a molecule. But the corresponding limitation is the finite optical resolution arising from the finite wavelength of light: objects smaller than about one-half the wave-length of visible light will appear only as a featureless circular blur. An advantage of modern imaging is the ability to digitally process images to enhance both resolution and contrast in all three spatial dimensions^{242 , 243}. But the corresponding limitation is that the processing might introduce artefactual features, as well as non-linearities in reported intensities.

Many imaging techniques are based on fluorescence, which in principle can be exceedingly sensitive down to the single-molecule level²⁴⁴, because (again, in principle) fluorescence intensity is measured up from a zero-intensity background. However, in practice the background might not be zero due to cellular autofluorescence or instrumental contributions. In addition, the foreground might photobleach away during observation, thereby providing only a limited number of photons with which to form an image. Because these photons arrive randomly at each camera pixel, any particular camera exposure of a finite duration may record more or less photons than might be expected from the long-term average. This effect, called 'shot noise', generates effective graininess in the image. The shot-noise graininess makes both intensities and positions somewhat uncertain and it is most severe at the low light intensities common in single-molecule detection.

In the lateral direction (that is, transverse to the optical axis of the microscope), the optical resolution 'limit' is generally in the order of 200 nm for wide-field optical detection. For NSOM, the lateral resolution limit is bet-ter, at around 50 nm (ref. 245). But NSOM is, unfortunately, difficult to use on bumpy surfaces such as cells, and works better on artificial planar membranes. In the axial direction, the optical resolution limit ranges from 1,000 nm for conventional epifluorescence microscopy²⁴⁶, to about 500 nm for confocal²⁴⁷ and two-photon²⁴⁸ microscopy, to about 50–100 nm for TIRM²⁴⁹. Quantitative image-analysis techniques can increase effective resolution by reducing all these numbers by a factor of about five.

Nonetheless, there still remains a gap of approximately one order of magnitude between these resolutions and the single (or sub-) nanometre scales that are characteristic of biochemical interactions in living cells. This gap can be partially filled by taking advantage of spectroscopic effects, the most commonly used of which is FRET²⁵⁰. FRET can detect the proximity of two fluorophores to within about 5 nm. Evidence of FRET can take many observable forms, including acceptor sensitization, donor quenching, reduction of donor photobleaching and donor fluorescence lifetime, and changes in acceptor polarization^{251 , 252}. Other forms of energy transfer, such as quenching and surface plasmon effects at metallized surfaces, can also be used to detect short (1–5 nm) distances.

Distances smaller than the optical-resolution limit can also be inferred indirectly from certain spectroscopic properties of fluorescence. Fluorescence polarization by itself does not have a characteristic distance because it arises only from fluorophore orientation and tumbling. However, in biological systems the orientation and rotation rate of cellular components are affected most strongly by the immediate molecular environment (within a distance of about 1 nm), so polarization effectively 'reads' information on these small-distance scales. Likewise, fluorescence lifetimes and spectral features (such as colour) are affected mainly by the immediate environment.

Finally, an intrinsic feature of a single visible photon emitted by an excited fluorophore — that its energy (about 3 eV) is well above the quantal unit of thermal background energy (kT, about 0.03 eV) — opens up the possibility of 'seeing' individual molecules. The now-successful detection of single molecules with fluorescence microscopy essentially provides a probe into distance scales smaller than the optical-resolution limit. The orientation and conformational state of those single molecules can be detected spectroscopically, effectively 'bridging the gap' between the optical-resolution limits and molecular scales.

Akihiro Kusumi. To address these questions, I would like to advance the argument that single-molecule techniques for studying events occurring in living cells are likely to open up a new frontier in biology. Conventional live-cell imaging is interesting and very useful — it allows local changes and spatial variations within a live cell to be monitored, rather than averaging over the whole cell or a large population of cells, and it is particularly useful for studies of compartmentalization in the cell. However, to understand exactly how molecular events take place, and how and why cellular systems, such as signalling systems, operate in living cells, we need to observe single molecules rather than collections of molecules. Otherwise, information on the molecular interactions that make the molecular systems work will be lost²⁵³.

Consider, for example, a raft-like domain 20 nm in diameter, and assume that molecule X enters and exits from this domain with two characteristic residency times of 50 ms and 500 ms, depending on which molecule (Y or Z, respectively) it interacts with in the domain, and also leaves with them, either into the cytoplasm or into another region of the membrane, respectively. Because the domain is small, the residency time is short (only a small fraction of all available molecules can be in the domain at one time) and the recruitment events take place randomly. A technique such as FRAP would therefore fail to detect the transient incorporation of molecule X into the raft domain, and we would fail to realize that such a raft works as a scaffold for inducing crucial interactions between X and molecules Y and Z.

Another simple example is the use of FRET for cell studies, which has recently become popular. Weak FRET signals are often observed, but there is no way of knowing, for example, whether only 5% of the molecules being monitored are engaged in FRET (and form a stable complex throughout the experiment), or whether all of the molecules are engaged in the formation of complexes that have a very short lifetime (say in the order of 100 ms). Distinguishing between these two possibilities is crucial for understanding how the cellular process involving the complex works and is regulated — that is, for understanding the underlying principles that cause, for example, signal transduction involving the complex to occur. Such an issue could be resolved satisfactorily by carrying out single-molecule FRET observations at high time resolutions (the term 'high time resolution' can mean, for modern scientific cameras, either short frame-exposure times or high frame-repetition rates, depending on the exact context; here the emphasis is on the high frame-repetition rates).

My argument here is that because cellular systems such as signalling networks consist of biological macro-molecules and molecular complexes, we need to start quantitatively defining the principles that make such systems work in the restless universe of thermal agitation and fluctuation. For this purpose, we first need to know the bulk on- and off-rates and effective concentrations of the molecules involved (although we cannot forget the cooperative formation of many molecules in the cellular signalling system, particularly the molecular complexes that involve scaffolding proteins or scaffolding membrane microdomains, such as rafts).

The first useful basic parameters to measure are the bimolecular collision rate and the off-rate, although these are difficult to determine in living cells. However, the development of single-molecule techniques could help — because molecular events in the cell are probably governed by a small fraction of molecules that are transiently brought together in a small molecular complex or a subdomain, single-molecule approaches might be one of the best techniques for studying such complexes in living cells. In addition, as single-molecule observations necessitate molecular-level resolutions (that is, at the nanometre scale), single-molecule technology can bridge the gap between micrometre and nanometre scales, addressing the second question above. I think more and more people are becoming interested in doing single-molecule observations in living cells.

Tony Magee. The length scale of imaging techniques is a problem. We do need techniques that can bridge the gap between light methods (˜200-nm resolution) and electron microscopy (1-10 nm). Preferably, these should be applicable to living cells. It was hoped that FRET could do this, but studies on lipid-raft systems reveal diametrically opposed interpretations, with some finding evidence for clustering of proteins and others not, depending on exactly how the technique is used^{220 , 254}, so this is still some way from being a universally and simply applied method. There is nothing quite as powerful as being able to 'see' a structure, rather than inferring its existence from indirect measurements. New high-resolution techniques that lower the resolution limit of light microscopy — such as stimulated emission depletion (STED), developed by Stephan Hell²⁵⁵ and 4Pi (a reference to 4πr² (the surface of a sphere) and the attempt to visualize a sample from all angles), developed by Mats Gustafsson²⁵⁶ — are promising, but are still a long way from being broadly useful in living cells. FCS could have a role, but it is an indirect method. NMR on biological membranes, or even living cells, has the potential to answer many questions if enough sensitivity can be achieved and probes can be incorporated without themselves perturbing the system.

Ernest M. Wright. Popular techniques now in use are largely those based on light microscopes, with an attendant limiting resolution of ˜250 nm, which is well above the resolution required to view membranes and membrane proteins. To increase the resolution of subcellular organelles by a factor of ten or more it is necessary to turn to electron tomography. Much progress has been achieved due to advances in technology, both in hardware such as goniometers and charge-coupled device (CCD) cameras, and in software that allows three-dimensional reconstructions and visualization of the maps²⁵⁷. Unfortunately, the resulting maps have limited resolution owing to difficulties associated with sample preparation and radiation damage. However, one approach that might have promise in imaging membranes and organelles in their natural environment is freeze-fracture electron microscopy²⁵⁸. Its validity was established by studies on a membrane protein of known structure (AQP0), for which freeze fracture was combined with multi-axis shadowing, random conical-tilt electron microscopy and Kohonen self-organizing maps to align images^{259 , 260}. The quality of the three-dimensional model of the aquaporin envelope is evident by docking it with the X-ray structure. Technical advances with goniometers have more recently enabled Zampighi and colleagues to use conical tomography to reconstruct an entire replica and obtain three-dimensional models of membrane proteins at 2–3-nm resolution²⁶¹. This method can also be extended to thin sections of plastic-embedded tissues. Conical electron microscopy could therefore prove to be a general method for reconstructing biological organelles and macromolecules in their cellular environments at high resolution. Nevertheless, such snapshots will have to be combined into a temporal series to obtain the dynamic information that we would all like to have.

Michael A. Edidin. This is too broad a question to be answered easily. The big trade-off for all imaging techniques is between perturbation of the native state and strength of signal. At one extreme we have electron microscopy, which gives the highest possible image resolution, but requires fixation, sectioning and the addition of contrast agents At the other, we have phase-contrast and differential interference contrast (DIC) imaging, which are applicable to live cells and do not require contrast agents, but at the price of relatively low resolution.

Newer imaging techniques, such as near-field microscopy (NFM) or TIRM, mainly use nonlinear optical effects to overcome the wavelength limits on spatial resolution. A good summary of these techniques is provided in a recent series of review articles^{249 , 262 , 263 , 264 , 265 , 266 , 267}. However, it remains to be seen how broadly applicable the techniques will be.

Whatever the technique, it cannot be used in isolation. The lessons learned early in cell biology, from the work of Palade and Siekevitz and their followers (see ref. 268 for a review), is that a single technique, whether electron microscopy or immunohistochemistry, must be complemented by others. It took a combination of electron microscopy and biochemistry to define and characterize cell organelles.

Back to list

Nicoletta Kahya. Model membranes, in particular liposomes, represent an essential tool for providing insight into lipid properties and mechanisms of function of membrane proteins^{269 , 270}. It is only in a simplified context that we can unambiguously identify and characterize the interplay of lipids and proteins. In addition, only if we are able to reconstruct a biological machine starting from its building blocks can we be sure to understand its mechanisms of function, even if the in vitro efficiency in terms of kinetic rates might be different from that in the cell. Although liposomes tell us something about the single (or a few) component(s) acting at the membrane, they lack information on important supra-molecular properties of biological membranes. Therefore, results obtained using model membranes can be used only in conjunction with a number of assumptions. First, model membranes lack the complexity of biological membranes, and possibly lose cooperative interactions and molecular crowding effects. In addition, model membranes are most likely systems at equilibrium, whereas biological membranes are most probably described by out-of-equilibrium thermodynamics. Finally, artificial membranes lack the ability to control transmembrane asymmetry²⁷¹, and there is also an absence of the underlying cytoskeleton, whose components have been shown to reorganize lipids and proteins at the membrane (see, for example, refs 272,273). In order to avoid oversimplifications, it is important to take a systematic approach, starting from well-defined models and gradually increasing complexity so as to mimic natural membranes as much as possible. The challenge for the future is to introduce these missing features into liposomes, either one by one to evaluate their effects on membrane organization, or all at once to bridge the gap between artificial and biological membranes.

Joshua Zimmerberg. Membrane models are extremely useful at many levels. First, they serve as 'test tubes' for membrane proteins, because the function of many membrane proteins cannot be studied in detergent solution. Planar phospholipid bilayer membranes are particularly suited to the study of transport properties, because the experimentalist has independent access to the aqueous solutions bathing both sides of the membrane. The permeability of the bare bilayer can be significantly altered by a single molecule of transporter, so model membranes can serve as exquisitely sensitive assay systems and sometimes reveal the conformational changes of single protein molecules. Here, they reveal the dynamics of protein activity²⁷⁴.

Second, the transformations of membrane shape and topology via budding, fusion, fission and poration can be studied by adding proteins to lipid bilayers and devising assays to measure the resulting changes in curvature, coalescence, fragmentation or leakiness. Third, it is essential to know the contribution of the lipid bilayer to the free energy of any postulated intermediates in biological membrane remodelling. Model membranes allow us to measure the elastic moduli, radii of spontaneous monolayer curvature and other physical parameters of membrane energy as a function of membrane composition. Such models also allow us to test hypotheses based on theoretical models of model membranes (!) and increase our confidence in our ability to calculate the free energy of an intermediate as a function of its composition. Fourth, the lateral de-mixing of lateral lipid microdomains in bilayers was first discovered in model membranes, and is currently an extremely hot topic in physics, chemistry, chemical engineering and biophysical circles²⁷⁵. We are only now beginning to understand the role that cholesterol plays in determining features of membrane domains, such as their size and structure. Investigating the behaviour of membrane domains in model membranes will facilitate our understanding of how the organizational potential of such lipid segregation is exploited by biological systems.

Anthony G. Lee. Whatever their limitations, membrane models are essential. In a sense, a crystal structure of a membrane protein is a 'model'; the environment of the protein in the crystal is very different to that in a real biological membrane, but few would argue that the crystal structure of a membrane protein is not an excellent model for the structure of that protein in a membrane. A major limitation of reconstituted membrane systems is that they lack the lipid asymmetry between the two halves of the lipid bilayer that is found, for example, in cytoplasmic membranes, but there is currently no evidence to suggest that this has any large effect on protein function.

Michael M. Kozlov. Experimental work with model lipid membranes is one of the main sources of quantitative information about the properties of lipid bilayers ('bulk' properties) that determine the behaviour of cell membranes containing proteins. Experiments with model lipid membranes allow the spontaneous curvatures and elastic moduli of the major biologically important lipids and their mixtures to be measured (see ref. 276 and references therein). Studies of structural rearrangements of model membranes driven by temperature, hydration and lipid composition have allowed us to understand intermediate stages of membrane-fusion reactions³⁶. Tubulation of fluid membranes upon application of local pulling forces revealed the formation of tubules imitating those observed in cells, and demonstrated that the values of the forces applied are close to those expected to be generated by molecular motors operating in the intracellular transport system²⁷⁷. Observations of shape transformations of model membranes caused by asymmetric lipid compositions of lipid monolayers, and the interplay between the membrane area and the enclosed volume, have enabled a partial understanding of the processes of membrane budding and fission²⁷⁸. Investigation of the formation of membrane domains by saturated and unsaturated lipids, as well as cholesterol- and sphin-gomyelin-containing mixtures, is indispensable for understanding membrane compartmentalization and, possibly, the formation of membrane rafts.

An important issue that needs to be taken into account when applying the results obtained in model systems to cell membranes is the constancy of lipid composition. In most set-ups, model membranes have constant, well-controlled lipid compositions. In addition, the lipid bilayers of model membranes tend to be in a state of thermodynamic equilibrium or are relaxing towards this state. By contrast, biological membranes never seem to be in an equilibrium state: their general and local lipid compositions change steadily because of metabolic processes and lipid transport to and from membranes. Therefore, a direct analysis of the properties of biological membranes based on those of model membranes can be performed on time scales shorter than the characteristic times over which lipid composition varies. Otherwise, the time variations of the local and global bulk properties of cell membranes have to be taken into account.

Patricia Bassereau. By definition, model membranes are composed of a very limited number of components compared with cellular membranes. Such membranes can be used to identify the minimal system required for a given cellular function to occur, such as intracellular trafficking, cell motility, cell adhesion and so on²⁷⁹. The effect of adding a specific protein or any other additional components to the system can then be directly monitored and understood. This progressive 'complexification' method is an interesting complement to classical molecular biology approaches, such as gene knockouts or mutations, for which the role of a protein is investigated by suppressing its expression (knockout) or modifying its structure (mutation).

Using model membranes, the lipid composition and concentration of proteins or ligands can be controlled and monitored. Furthermore, in the case of vesicles or lipid nanotubes, size and shape can also be controlled. This enables important physical parameters of model membranes, such as rigidity, tension, curvature and size, to be measured and tuned (see, for example, refs 280-282). Such parameters are much more difficult to measure directly in cells. Because quantitative measurements are possible with model systems, and because the effect of each parameter can be de-coupled from all others, a direct comparison with theoretical models or numerical simulations is possible and general predictions can be made.

For example, it has been possible, using a general phase diagram, to determine the number of molecular motors pulling tubular transport intermediates when the membrane tension and the motor density on the membrane are known, in addition to the thresholds for both the membrane tension and the motor density required to observe tube formation²⁷⁷. The role of each parameter can therefore be understood and subsequently analysed in the cellular context. As an example, the existence of a tension gradient between the ER and Golgi has recently been measured in vitro, which explains lipid transfer from the Golgi to the ER²⁸³. Furthermore, the effects on lipid sorting of bending rigidity in different lipid phases have been investigated²²⁴. Ultimately, when a general model for a cell function is proposed, it can be directly tested on the cell, provided that essential parameters can be tuned. In this respect, protein concentration or membrane composition can be specifically modified, although membrane tension is certainly more difficult to control.

It is clear that model membranes are much less complex than real membranes; for example, lipid composition is generally symmetrical in model membranes²⁷¹, and phenomena such as membrane coupling to the cytoskeleton are difficult to reproduce in vitro ²⁸⁴. When several constituents have to be assembled together to reproduce a particular cell function, preparing such a model system might be too great a challenge, and it is probably more appropriate to study the cell directly. A good example could be the achievement of a complete mimetic system for cell adhesion. Cell-adhesion molecules (CAMs), such as integrins and cadherins, can in principle be incorporated into membranes (although it is not an easy task), and the resulting adhesion of the decorated membrane can be studied. An artificial cytoskeleton can then be added and its contribution measured, while remaining decoupled from CAMs. However, the full signalling pathway from adhesion molecules to actin polymerization is probably impossible to reproduce in this system. Finally, the cell is never at thermodynamic equilibrium, and although some of the non-equilibrium processes of cellular membranes can be reproduced in vitro and incorporated into theoretical models^{285 , 286}, it is often not possible to do so. For example, lipid domains can be formed in giant vesicles using realistic lipid mixtures, but no conclusions about raft size can be made from these experiments at thermodynamic equilibrium, as the domain size in vivo certainly results from many non-equilibrium processes, such as membrane fusion and fission²⁸⁷.

In conclusion, although model membranes do have limitations, they can contribute to a better global understanding of cellular functions.

John Silvius. Membrane models composed purely of lipids have become highly sophisticated, ranging from diverse types of supported mono- and bilayers to lipid vesicles with diameters ranging from tens of nanometres to tens of micrometres. Such systems can be very useful for investigating properties of the membrane bilayer that reflect intrinsic properties of the membrane lipids, as well as various aspects of the function of reconstituted proteins, such as the effects of lipid composition on protein activity. Nonetheless, as membrane models often fail to incorporate certain central properties of biological membranes, their reliability to model some key aspects of membrane behaviour can be limited. Below are three important cases in point.

Model membranes seldom replicate the asymmetric distributions of different lipids that characterize many biological membranes. Bilayers with asymmetric lipid compositions might be needed to model faithfully some fundamental physical characteristics of biomembranes (such as curvature properties or 'domain' formation), and potentially also to reconstitute physiologically relevant behaviour of proteins found in membranes that normally exhibit lipid asymmetry. Although methods have been described to prepare both supported and free-standing bilayers with asymmetrical lipid compositions, with the exception of asymmetric 'black' lipid membranes these systems have not enjoyed widespread use, largely due to the technical challenges of preparing them^{288 , 289}.

Even when reconstituted with membrane proteins, model membranes often contain considerably less protein per unit mass of lipid than do biological membranes. As a result, studies of model systems to date have contributed little in illuminating questions such as the effects of integral membrane proteins on 'domain' formation in biomembranes. This limitation of model systems is, of course, surmountable, although controllable reconstitution of many membrane proteins remains an experimental challenge.

Finally, model membranes are normally considered to exist 'at equilibrium' (a statement which is not always correct), whereas biological membranes are typically subject to ongoing synthesis, degradation, and influx and efflux of their components. The importance of this difference can be overestimated; in many respects a membrane in steady-state can behave like a membrane at equilibrium, for example. Nonetheless, if the rate of the introduction or removal of biological membrane components is faster than the rate of lateral (or transverse?) equilibration of these components through diffusion, a membrane model at equilibrium is obviously a poor model for the analogous biomembrane under physiological conditions. This issue needs to be assessed with care when considering a given application of a model system to understanding biomembrane properties. Fortunately, the availability of increasingly detailed and quantitative information about the trafficking, synthesis/degradation and intramembrane diffusion of various membrane components promises to facilitate this task^{213 , 290}.

Tony Magee. Membrane models could be very useful, but they need to be better informed by knowledge of lipid composition and asymmetry of biological membranes, as described above (see question 3, page 9). They are too simplistic at present (although one has to start somewhere!). Asymmetric membranes could be made as described by Sophie Pautot and colleagues²⁸⁹, or using a double-dipping Langmuir-Blodgett trough. The binding of basic peptides/proteins to acidic membrane surfaces could also be very important in modulating the local environment of membrane proteins, and this could be studied in model systems.

Akihiro Kusumi. Conceptual models of the cell membrane are extremely useful for studying any processes and events that occur in/on the membrane. For example, I would not think that any fundamental understanding of molecular events occurring in the plasma membrane is possible without considering the Singer-Nicolson model²⁹¹. Membrane models provide a structural and dynamic framework for determining the underlying principles that make a particular molecular event happen in the membrane. However, it is important to always keep in mind a healthy doubt about the validity of a given membrane model. What is observed in practice might be inconsistent with the structure proposed by existing models, and one must always be ready to create new models that are consistent with observations.

Many artificial experimental model membranes, such as reconstituted protein–lipid membranes and giant liposomes, have been used. These experimental model membranes are useful for understanding the basic molecular interactions and dynamics that might be occurring in the cell membrane, and are therefore useful for conceptualizing the cell membrane. The conceptual models are basically cartoons that emphasize only a small number of the essential aspects of membrane structure, dynamics and molecular interactions, and experimental models should be designed to test the simplified ideas depicted in these cartoons, but hopefully with fully quantitative analysis. Therefore, my ideal paper using experimental model membranes should end with a cartoon with numbers placed on it.

Richard A. Bond. This is a really difficult question for me to answer. First, because I am not a membrane biologist; and second, because I do not like to sound negative about any attempts at discovery and understanding. However, the reductionist approach is something I am familiar with in drug discovery and perhaps I can make some analogies with those results. The drive towards simpler models is derived from the fact that more parameters can be controlled. Although this makes sense, it often comes at a very high price in terms of the physiological relevance of the results. Using membrane models seems appropriate, to me, for studying the physico-chemical properties of certain lipids, but it does not seem appropriate for understanding a biological membrane. In terms of properties as simple as composition, not to mention the milieu of the cellular environment, the models are unrealistic. To assume that the model reflects biology is naive at best. It is perhaps analogous to developments in drug screening in recent years that have resulted in higher-throughput assays, although fewer new drugs have been discovered²⁹². Research using membrane models will yield new knowledge, some of which might even be exploitable in financial terms, but I believe these models will turn out to be a bad investment for gaining knowledge about biological membranes and pathophysiology.

Session 5: Membrane segregation at the cellular level

Back to list

Kai Simons. We assume that there are two major mechanisms for sorting proteins to the cell surface. One mechanism is driven by receptor-mediated cargo capture as a consequence of specific protein–protein interactions. Such sorting proteins include adaptor complexes that recognize determinants in the cytosolic domains of the cargo proteins. The other mechanism is lipid-raft-driven, as described in my response to question 14 (see page 40). The delivery of cargo carriers to different surface domains is organized by microtubular and/or actin tracks under the control of specific motor protein complexes and by localizing the receiving Rab- SNARE machinery at the surface sites at which delivery should occur^{293 , 294}.

Jennifer Lippincott-Schwartz. There are three principal ways that protein sorting at the cell surface is achieved: post-Golgi targeting to specific surface domains (as occurs in polarized cells when apical and basolateral proteins are directly targeted from the trans-Golgi network to apical and basolateral domains, respectively); endocytosis and recycling to a specific surface domain (also called transcytosis); and retention of a protein at a particular surface domain through interactions with the extracellular matrix, other membrane proteins or the cytoskeleton^{295 , 296 , 297}. Different cell types rely on these mechanisms to different extents for differentiating their cell surfaces.

Michael J. Caplan. Even closely related epithelial cell types derived from a single tissue can sort the same protein to distinct subcellular localizations. Although in most cases the mechanisms responsible for this variable sorting behaviour are not well understood, it is clear that this phenomenon must be attributable, at least in part, to the differential expression or function of proteins involved in sorting. The subcellular destination of proteins carrying tyrosine-based sorting motifs, for example, depends on whether these proteins are expressed in epithelia that express the μ1b adaptor subunit protein (in which case they generally accumulate at the basolateral surface) or in epithelial cells that are devoid of μ1b expression (in which case they are apically or randomly targeted)²⁹⁸. By carefully controlling the types of sorting machinery that it expresses, an epithelial cell can therefore individually select the destinations specified by each class of sorting signal.

According to this model, epithelial cells can exploit the differential expression or activation of sorting machinery to ensure that the distribution of plasma membrane proteins among their apical and basolateral domains is in concert with their specific physiological requirements. This interpretation further suggests that sorting signals do not actually specify a single destination, such as 'apical' or 'basolateral'. Instead, they serve to mark those proteins that bear them as belonging to a unique cohort. All of the members of this cohort will be sorted together in every epithelial cell type in which they are expressed, and their shared destination will be determined by the constellation of sorting machinery present in the epithelial cell types that express them.

Back to list

Kai Simons. Our idea of raft-driven protein sorting is the following. Raft clustering provides a mechanism for the generation of transport carriers. If phases with different properties coexist in the same membrane, as in the case of liquid-ordered microdomains in a membrane mainly in the liquid-disordered phase, there is line tension at the phase boundaries. The energetic cost of line tension can be reduced by decreasing the contact between phases; that is, by decreasing the boundary length of domains that constitute the minority phase. This is achieved most effectively when domains that are part of the minority phase bud out of the majority phase and eventually detach from the parent membrane. The reduction in line tension is sufficient to pay for the cost of bending membranes. This mechanism, termed domain-induced budding, was originally postulated on theoretical grounds but has recently received experimental support from studies of model membranes²⁷⁵. In cells, the clustering could be driven by glycan-protein, glycan-glycan, lipid-protein or protein–protein interactions.

Erwin London. There are some good hypotheses that have been proposed about the origin of sorting. However, this question cannot be answered from the biophysical perspective until the relative affinities of different proteins for sphingolipid-cholesterol microdomains, especially of proteins with different length transmembrane segments, have been more accurately measured.

Sean Munro. The short answer to this is that we do not yet know for certain, and indeed even the existence of such microdomains in living cells is not universally accepted. What is generally agreed on is that cholesterol and sphingolipids can form microdomains in protein-free model membranes composed of appropriate lipids^{299 , 300 , 301}. These microdomains have different physical properties from the rest of the bilayer, being thicker and more ordered, and so if they formed could preferentially include or exclude specific membrane proteins. It has been proposed that such partitioning could be responsible for a number of protein-sorting processes in living cells, including sorting between the apical and basolateral surfaces of polarized cells, recruitment of proteins into 'signalling platforms' at the plasma membrane, uptake of receptors by clathrin-independent endocytosis and retention of proteins in the Golgi apparatus^{299 , 300}.

It is difficult to demonstrate that these models apply to living cells because the removal of cholesterol and sphingolipids affects many properties of bilayers in addition to domain formation, such as bilayer permeability and fluidity^{301 , 302 , 303}. Moreover, the properties of biological membranes might inhibit the domain behaviour that is seen in vitro with model membranes. In particular, biological membranes are highly asymmetric, with sphingolipids being located principally in the outer leaflet of the plasma membrane, which might prevent the coupling between leaflets required for domains to mediate the transduction of signals across bilayers. In addition, the levels of sphingolipids and cholesterol in the plasma membrane are sufficiently high that it raises the possibility that much, if not all, of the outer leaflet of the plasma membrane is a uniform ordered domain. However, it should also be stressed that none of these models of microdomain-based sorting have been definitively disproved. Nonetheless, even if some of these models prove to be correct it seems unlikely that they are all correct, as some seem to be antagonistic. For instance, if microdomains mediate sorting to the apical surface^{299 , 300}, then proteins with a preference for such domains should be excluded from the basolateral domain, and therefore microdomains could not be responsible for signalling events or endocytosis at the basolateral surface.

Akihiro Kusumi. I have not seen any convincing evidence of the existence of sphingolipid-cholesterol microdomains that drive membrane protein sorting in the absence of inductive cues, such as extracellular signals, crosslinking (intentional or unintentional as a result of using fixation reagents) or lowering the temperature. I therefore believe that it is premature to ask whether these domains drive the sorting of membrane proteins. However, if a mechanism were discovered that induces the clustering of (raft-related) cargo molecules (such as binding and crosslinking by a hypothetical protein that recognizes only GPI-anchored proteins in the trans-Golgi region), it would be a different story (for interesting initial data, see ref. 304). Such a mechanism would create stable raft-like domains in which, in addition to cargo molecules, sphingolipids and cholesterol could be concentrated, and would help to sort these clustered proteins, sending them to the right places in the plasma membrane.

It is my opinion that clusters of membrane protein cargo recruit sphingolipids and cholesterol or induce very small and unstable sphingolipid-cholesterol domains to form relatively stable protein–sphin–golipid–cholesterol complexes (stabilized rafts). That is to say, I do not believe that membrane sphin-golipid-cholesterol microdomains drive membrane protein sorting. Rather, the sorting process is likely to be a cooperative process that involves the formation of protein complexes and the stabilization of the clusters of the protein, sphingolipid and cholesterol. The crucial issues now are therefore to identify the mechanism (or proteins) that induce clustering of membrane protein cargo in the Golgi complex, and to find out exactly how the formation of such stabilized rafts helps sort the protein cargo and transport proteins to their correct destinations.

Back to list

Vivek Malhotra. Surely the question is not whether the Golgi complex can control mitotic progression, but how and why it achieves this control. It is remarkable that only the pericentriolar Golgi membranes are found in a fragmented form during mitosis. In plants, fly embryos and yeast, Golgi membranes are not found in the pericentriolar region, and they do not fragment during budding/cell division³⁰⁵. By contrast, Golgi membranes in mammalian cells, which are confined to the pericentriolar region, do fragment during mitosis^{305 , 306}. So, is there a connection between centriolar localization of the Golgi apparatus and its fragmentation during the cell cycle?

The obvious possibility is that a defect in Golgi fragmentation acts as a signal that one of the newly formed daughter cells will be devoid of this compartment, so the cell aborts entry into mitosis. But how a defect in fragmentation is monitored by the cell and then used to prevent entry into mitosis is not known. It is also quite possible that these (fragmentation) dynamics serve other purposes. Interestingly, fusion of FIG, a gene coding for a Golgi-complex-associated protein, to the kinase domain of the proto-oncogene c-Ros results in the generation of a potent oncogene whose transforming potential requires association with the Golgi complex³⁰⁷. In other words, the Golgi complex functions as a signalling surface by acting as a host to components with the potential to transform cells. Fragmentation releases these components and allows cells to progress into and through mitosis. An understanding of pericentriolar, Golgi-specific check-point/signalling will therefore not only reveal how cells co-ordinate cell division, but also provide novel therapeutic agents for inhibiting uncontrolled cell division.

Jennifer Lippincott-Schwartz. This possibility was first proposed on the basis of the finding that injection of an antibody to GRASP65 (a peripheral Golgi protein) in G1 of the cell cycle seemed to inhibit/retard entry into mitosis some 9 hours later, unless the Golgi had been fragmented previously by treatment with nocodazole³⁰⁸. More recent work expressing a wild-type GRASP⁶⁵ carboxy terminus to perturb GRASP65 function revealed a delay but not a block in mitotic entry when GRASP65 activity was altered in interphase³⁰⁹. There therefore seems to be a link between Golgi structure and cell-cycle control, but perhaps not a cell-cycle checkpoint in the conventional sense.

In addition to potentially regulating mitotic entry, the status of Golgi structure is also important for regulating processes occurring during mitosis itself. This has been demonstrated in studies examining the effect on mitotic cells of expressing a constitutively active form of ARF1, a Golgi-associated, small GTPase that plays a central role in recruiting and retaining numerous signalling and regulatory molecules on the Golgi. ARF1 normally becomes inactive in mitosis, causing numerous ARF1 effectors to redistribute off Golgi membranes and into the cytosol³¹⁰. If ARF1 remains in an active state during mitosis, however, these effector molecules (including Snk Akin kinase 1 (SAK1), myosin IIA, coatomer and cullin 2) remain Golgi-associated and the Golgi does not disassemble³¹⁰. The failure to release these and other molecules from the Golgi during mitosis leads to impairment of chromo some segregation and inhibition of cytokinetic fur row ingression³¹⁰. The status of Golgi membranes in mitosis is therefore important for the regulation of various aspects of cell division, including mitotic entry, chromosome behaviour during metaphase and cytokinesis.

Session 6: Signal transduction in light of membrane structure

Back to list

Sandra L. Schmid. Cell biology emerged following the ability to visualize intracellular organelles and particles through light and electron microscopy, originally in fixed cells³¹¹. We need now to emerge from static thinking into a steady-state, dynamic view in which organelles are in constant flux and are defined by their content and function over time. Are there different types of early endosome? I don't believe so. But early endosomes can attain a distinct steady-state composition, and therefore distinct function, that is driven by the distinct cargo molecules that they carry. Cargo molecules do not seem to be passive passengers, like those that might ride city buses. Rather, some cargo molecules can actively control their vehicles, like taxi passengers, and through their protein interactions and signalling capacity control the content of the vehicle (that is, the early endosome), its destination and even its rate of travel. We have seen this clearly when the cargo is a patho–genic bacterium that can alter the composition of an early endosome so that it no longer interacts with the endocytic pathway, thereby avoiding acidification and so on³¹². Is it not likely that endogenous molecules, such as GPCRs or receptor-tyrosine kinases, can also modify the organelle in which they are transported for physiological rather than pathological purposes? We need not invoke distinct organelles, but rather different functional variations along a continuum. The variations could be kinetic and quantitative rather than qualitative.

Ritva Tikkanen. I think the endosomal system is a good example of a highly complex compartment, and I am sure that we still haven't discovered all the possible subclasses of endosomes, including the early endosomes. Different cell types have different types of endosome, so we have a long way to go before we can conclude that we more-or-less understand exactly how these compartments work!

Judith Klumperman. The classical function of early endosomes is to select proteins either for recycling to the plasma membrane or for degradation in the lysosome. With the emerging role of endosomes in other cellular functions, such as signalling and temporary protein storage^{169 , 170 , 204 , 313}, the possibility is raised that these specialized functions might require a distinct population of endosomes.

Suzanne Pfeffer. Yes — Marino Zerial has strong evidence for multiple classes of Rab5-containing structures, some with and some without the protein EEA1 (ref. 198). Recycling endosomes are a class of early endosome that contain different Rabs and different cargos to EEA1 compartments¹⁹⁷. Early endosomes will continue to be characterized as we identify more of the molecules that occupy and define these structures.

There are also different types of late endosome — some Rab7-positive, some Rab9-positive^{194 , 194}. As for early endosomes, this sub-categorization will continue.

Frederick R. Maxfield. There are several types of early endosome¹⁸⁴. Early endosomes might be considered to be all endosomes that contain cargo within 5 minutes after it has been cleared from the cell surface. The best-characterized early endosomes arise from internalization through clathrin-coated pits. These are tubulovesicular compartments that are also called sorting endosomes. The small GTPases Rab4 and Rab5 associate with these organelles. They are very dynamic compartments, and many proteins, such as transferrin receptors, as well as many lipids, exit these sorting endosomes with half-times of 2 minutes or less. At the same time, ligands such as low-density lipoprotein (LDL) are released from their receptors due to the low pH in the endosomes, and these ligands accumulate in the lumen of the sorting endosomes.

With a half-time of 8-10 minutes, the organelles and their accumulated contents start to move towards the cell centre, and the organelles, which are maturing into late endosomes, no longer accept recently internalized molecules. Most of the membrane components that exit the sorting endosomes are in tubules that bud out from the organelles. These tubules can return directly to the cell surface, or they can follow an indirect route through an endocytic recycling compartment (ERC). Rab11 is associated with the ERC. Internalized molecules can enter the ERC within 2 minutes after internalization and so the ERC, which is part of a recycling itinerary, is often described as part of the early endosomes.

There are other endocytic pathways that do not start with clathrin-coated pits, and these account for about half of the membrane internalization in many cell types^{315 , 316}. Caveolae can form endocytic compartments under some circumstances, but they are probably a minor component of non-clathrin endocytosis under most conditions. The initial endocytic compartments formed by these alternate mechanisms are generally not well characterized. In some cases, material that is internalized by these pathways enters the same endosomes that contain cargo brought in by clathrin-coated pits.

Back to list

Tony Magee. It is clear that covalently attached lipids (such as fatty acids, prenyl groups, GPI anchors and cholesterol) dramatically influence protein function by specifying subcellular and perhaps extracellular (in the cases of Hedgehog and Wnt proteins) localization³¹⁷. Reversible S-acylation (palmitoylation) can control this dynamically, and Lewis Cantley referred to the work of Philippe Bastiaens³¹⁸ showing that this is the case for signalling Ras, which cycles between the plasma membrane and the ER/Golgi depending on its S-acylation status. So, understanding the enzymology and regulation of the mechanism of S-acylation will be crucial.

Recent progress has been made in this area with the identification of genes encoding putative S-acyl transferases in yeast^{319 , 320} and humans (B. Deschenes & M. Linder, unpublished observations) — the DHHC-CRD (Asp-His-His-Cys cysteine-rich domain) genes. S-acyl-transfer-promoting activities (which are not necessarily strictly 'enzymatic', although the jury is still out on this) have also been identified for secreted proteins such as Hedgehog/WNT and for the SNARE VAC8 (ref. 321). For hepta-spanning receptors, S-acylation of the cytoplasmic tail occurs in most members of the family and has been shown to regulate the reversible association of the tail with the membrane and therefore modulate coupling to G-proteins and desensitization in different contexts. I will leave it to others to comment on the roles of non-covalently attached lipids in regulating receptors.

Session 7: Membrane pathophysiology and disease

Back to list

Joshua Zimmerberg. I believe that a treasure trove of therapeutic targets awaits us. When we considered the lipid bilayer backbone of the membrane to be a uniform matrix in which proteins functioned independently, not much in the way of therapeutics were envisioned (save general anaesthetics) due to the lack of specificity and the importance of preserving the barrier function of the lipid bilayer to ion permeability. However, our view of the membrane is changing and we increasingly appreciate its heterogeneous composition. I propose that there are a myriad of specific local membrane compositions for each membrane function, particularly in the assembly of macromolecular protein–lipid complexes that act as machines. Any specific macromolecular machine that requires specific lipids for its assembly will be a candidate target for a drug, such as a specific lipid analogue that might interfere with or stabilize the assembly of that machine, or an inhibitor of the synthetic pathway for a crucial lipid required for assembly. Or we could add compounds that take advantage of specific membrane composition. For example, the polyene antibiotics nystatin and amphotericin B bind ergosterol, the major yeast sterol, 300-fold more strongly than cholesterol, and so we treat yeast infections with these lipoidal antibiotics so that they enter the plane of the plasma membrane of the yeast cell, bind ergosterol and form channels that puncture the yeast cell's barrier to ions^{322 , 323}.

Akihiro Kusumi. Increasing our understanding of membrane complexity will generate a tremendous array of possibilities for developing new drugs. Complexity is coupled to specificity, and drugs that can affect very specific events, subdomains or pathways in the cell membrane could be developed. One could think of a huge variety of therapeutic opportunities related to the complexity of membrane composition. For example, drugs that modulate the lipids associated with lipid-anchored proteins could change their associative properties with specific microdomains in the membrane. Existing drugs could be chemically modulated so that they target specific or different microdomains in the membrane, to enhance their effects or reduce their side effects.

Tony Magee. The enzymes that attach lipids covalently to proteins could be good drug targets. Farnesyl transferase was originally thought to be a good target due to the farnesylation of the Ras oncoprotein³²⁴, but has turned out to be disappointing. This might in part be because there is a single farnesyl transferase that modifies all farnesylated proteins, which makes it difficult to achieve specificity. S-acylation might be better from this point of view, because there are more than 20 DHHC–CRD, genes in humans, many of which could promote S-acylation of a subset of targets (and studies in yeast do indicate protein-substrate specificity)^{319 , 325}. Similarly, there are different gene products that promote S-acylation of the secreted signalling molecules Hedgehog and WNT, both of which are key factors in many cancers^{317 , 325}.

Ernest M. Wright. It is clear that the function of membranes as gatekeepers of the flow of molecules and information in and out of cells is determined by membrane composition. It is also becoming clear that the composition of membranes is altered in many diseases. For example, a number of genetic diseases, such as cystic fibrosis, diabetes insipidus and glucose galactose malabsorption, are caused by disruptions in the trafficking of membrane proteins to and from the plasma membrane. In order to devise therapeutic interventions in such diseases, it is necessary to understand what controls the density of transporters, channels and receptors in plasma membranes.

Back to list

Tony Magee. Compartmentalization is crucial for normal cell function and for regulating many processes, such as signalling. Lipid rafts are a good example — in T cells they seem to both concentrate molecules that need to interact for efficient transduction of T-cell receptor (TCR) signals (LAT, LCK, CD4/8 and PIP²) and exclude the negative regulatory tyrosine phosphatase CD45 (ref. 87), as well as controlling the activation of Src kinases by regulated binding of C-terminal Src kinase (CSK) to the raft-associated docking molecule Cbp/PAG (ref. 326).

Boris N. Kholodenko. External information received by receptors on the plasma membrane is transmitted to, and processed in, different cellular compartments, including the plasma membrane, the cytoplasm and intracellular membrane compartments, such as endocytic vesicles and the Golgi complex. The same protein cascades operate in surprisingly dissimilar ways when localized to different cellular compartments³²⁷. Sometimes, this difference is attributed to distinct enzyme isoforms, as in the case of PKC or Ras signalling from different locations, but often it is purely a localization effect^{328 , 329 , 330}.

Following receptor activation, many cytoplasmic signalling proteins are targeted to cellular membranes. In contrast to previous suggestions that the role of this recruitment is to increase diffusion-limited (first-encounter) rates, the function of membrane localization has recently been shown to amplify the number of complexes formed between signalling partners, resulting in up to a 1,000-fold increase in the steady-state activity of a pathway³³¹. MAPK cascades control cellular decisions to proliferate, differentiate or undergo apoptosis, and display remarkably different input-output behaviour when assembled on scaffolding proteins. Whereas in the cytoplasm, MAPK cascades behave as ultrasensitive or bistable switches, their assembly on a scaffold drastically decreases thresholds for input activities (by thousands of fold) and reduces ultrasensitivity of responses^{332 , 333}. As intracellular protein components of pathways initiated by different plasma membrane receptors greatly over-lap, distinct signalling outputs from different compartments are crucial in ensuring signalling specificity and, therefore, normal cell function.

Ritva Tikkanen. Cellular signalling pathways are highly complex, and are based on multi-protein assemblies and extensive protein modifications, such as phosphorylation. Ever more new molecular factors are discovered in the 'old' pathways, such as the EGF-receptor or insulin signalling pathways, which reveals how complex cellular signalling really is. Regulation of these signalling pathways is extremely important, and defects in regulation can have fatal consequences, such as the development of cancer.

Many regulatory factors compete for the same binding partners, and the resulting interactions are crucial for directing the signals in various directions (proliferation, cytoskeleton and so on). So, compartmentalization, such as the formation of microdomains or endocytosis of signalling complexes, can favour certain interactions by increasing the relative abundance of respective binding partners in these compartments. This might help to direct signals in certain directions according to the needs of the cell, and may enable molecules to choose the 'right' binding partner from all the available candidates. This would make regulation much easier, because some competitors would not be accessible anymore. For example, it was shown for the insulin receptor that an additional, PI3K-independent, signalling pathway exists that leads to recruitment of the GLUT4 glucose transporter to the plasma membrane even in the absence of PI3K activity³³⁴. This signalling proceeds through membrane rafts and requires a trimeric complex consisting of the ubiquitin ligase c-Cbl, its associated adaptor protein CAP/ponsin and reggie 2/flotillin 1. The discovery of this novel pathway leads to the conclusion that such alternative routes might be necessary to tightly regulate the direction or specificity of receptor signalling.

Michael J. Caplan. Cells make use of a fairly limited repertoire of signalling molecules to communicate a multitude of different messages. The specific meaning of a given signal therefore often depends not only on its biochemical composition but also on its subcellular localization. By restricting the production and interpretation of molecular signals, cells can imbue these messages with spatial contexts that help to prevent them from becoming degenerate. Processes such as cellular locomotion, cytoskeletal assembly and regulated vesicular fusion, for example, are all susceptible to regulation by changes in the concentration of cytosolic calcium³³⁵. Cells are able to select among these and many other responses to variations in cytosolic calcium, in part because the mechanisms through which calcium concentrations are induced to change can be governed by the generation of compartmentalized messages that are tightly restricted in space and time.

It is important to note that in addition to constraining a message's information content, compart-mentalization of signalling machinery within defined subdomains of the membrane can also facilitate the message's generation. This is especially true of signalling molecules produced through the actions of enzymes on specific species of membrane phospholipids. Membrane subdomains permit these lipid precursors to be concentrated in extremely close proximity to the proteins that actuate their signalling potential. The activities of numerous ion channels and transport proteins, for example, are tightly controlled by multiply phosphorylated species of phosphatidylinostiol that are generated and degraded in the immediate neighbourhood of the channels and transporters themselves³³⁶. Signals can therefore be generated without the sort of temporal lag or losses of spatial resolution that might be expected to characterize the generation of a messenger whose production requires substrates and enzymes to meet through diffusion in three dimensions.

Lewis C. Cantley. As indicated in my response to question 8 (see page 20), the most well-studied compartmentalization is that of lumenal versus basolateral membranes of epithelial cells, which are joined by tight junctions¹⁵⁸. In these cells, it is very clear that certain receptors are confined to each of these compartments and that this is important for cellular responses. For example, EGFR is specifically localized to the basolat-eral side of epithelial cells to ensure specific activation of receptors in this compartment³³⁷. A second well-studied example is neuronal cells, in which certain signalling proteins are confined either to dendrites or to axons³³⁸. In addition, sub-compartments of dendrites called 'spines' are enriched in receptors and channels that respond to the axons of adjacent neurons. Lymphocytes have an analogous structure (sometimes called the immune synapse) that concentrates signalling proteins at the site at which antigen-presenting cells make contact. There is clear evidence that these local clusters are crucial for immune signalling³³⁹. There is also evidence, although less convincing, that micro-compartments exist in the plasma membrane that are crucial in concentrating signalling proteins and lipids³⁴⁰. However, it is difficult to visualize these compartments in real time in live cells due to signal-to-noise problems in fluorescence microscopy. And it is an even greater challenge to show that they are important in signalling. New techniques are needed to address this question. For example, improved instrumentation and improved molecular probes for measuring FRET, FCS, fluorescence photobleaching, fluorescence polarization and fluorescence lifetimes in live cells will allow a better understanding of micro-domains and their role in signalling.

Back to list

Joshua Zimmerberg. Simply put, the function of cholesterol homeostasis is to keep membrane cholesterol levels high enough for the activity of important membrane proteins and low enough to prevent rheo-logical changes in the red-cell membrane. In the context of this membrane-focused discussion, there are at least two dimensions to this question. One has to do with the levels of cholesterol in different cellular membranes (as opposed to levels of plasma cholesterol in humans, which are clearly risk factors for atherosclerosis leading to coronary artery disease and stroke, and for which there is a huge literature not considered here); the other has to do with the chemical structure of cholesterol. There is a strong degree of homeostatic control of membrane cholesterol within fibroblasts, which maintains the level of cholesterol in the cell membrane near its set point³⁴¹. Because most of the current in vitro studies on lowering cellular cholesterol measure whole-cell cholesterol and not membrane cholesterol, we do not really know the dose-response curve for cholesterol concentration and membrane function; however, from both published work and our unpublished results, it would seem that relatively small changes in cellular cholesterol affect important cellular functions (see ref. 342 for an example).

This implies to us that cooperative weak forces govern the interaction of cholesterol with membrane proteins, and therefore homeostasis and a set point for a particular mole fraction of cholesterol in the membrane would be very important. The exact chemical structure of cholesterol might also be very important. In the genetic disease Smith-Lemli-Opitz syndrome, merely changing one double bond in cholesterol to 7-dehydrocholesterol can lead to multiple congenital malformations, mental retardation and death in the first weeks of life³⁴³. It is not yet clear whether this results from markedly diminished serum cholesterol levels or the build-up of 7-dehydroc-holesterol and its inability to replace cholesterol, despite being nearly chemically identical.

Sean Munro. A key property of cholesterol is to increase the order of phospholipid bilayers (sometimes called the 'condensing' effect) and thereby reduce their permeability to small molecules³⁰². This permeability-reducing property is shared by sphingolipids with saturated acyl chains, which might account for the high levels of these two classes of lipids in the plasma membrane^{301 , 303}. Indeed, cholesterol-like molecules called hopanoids are even found in certain bacteria. Hopanoids also reduce bilayer permeability in vitro, and are induced in vivo in response to low pH or desiccating conditions, and in some nitrogen-fixing bacteria they are major components of an internal membrane that acts to prevent oxygen from reaching nitrogenase^{344 , 345}.

Another key property of cholesterol is to promote lipid mixing. In particular, sphingolipids with long saturated acyl chains are abundant in many biological membranes, but in isolation form a solid 'gel' phase at physiological temperatures. Even when mixed with fluid phospholipids they still segregate into gel-phase domains. However, the presence of cholesterol promotes the mixing of the two other lipid species into a uniform fluid bilayer at physiological temperatures.

This 'fluidizing' capacity of cholesterol ensures that the biological membranes can accommodate these high-melting-temperature sphingolipids, and this might be crucial for reducing membrane permeability. Indeed, sphingolipids with very long acyl chains, and therefore very high melting temperatures, are most abundant in particularly impermeant bilayers, such as myelin and the barrier epithelia of the kidney and intestine^{346 , 347}. A role for cholesterol and sphingolipids in increasing the impermeability of the plasma membrane might preclude these lipids being clustered in microdomains (for a discussion of microdomains, see my response to question 14, page 40). However, permeability issues might be less crucial in some internal membranes, such as endosomes and the Golgi complex, and so if cholesterol also contributes to domain organization it might be more likely that this occurs in internal membranes.

Apart from its role as a bulk constituent of bilayers, trace amounts of cholesterol have other roles — for example, cholesterol is the precursor for the biosynthesis of steroid hormones, and is attached as a lipid modification to the signalling protein Hedgehog³⁴⁸. Nonetheless, it seems possible that cholesterol homeostasis principally keeps plasma membrane cholesterol levels at sufficiently high levels to ensure that our membranes do not leak (and hence we can think!), and yet are uniformly fluid to allow bilayer components to diffuse rapidly and thereby associate and dissociate as required by the dynamic processes of life.

Kai Simons. A mammalian cell that is capable of synthesizing cholesterol uses close to 30 enzymes for this purpose. The cholesterol levels are controlled by an intricate network that regulates the homeostasis of endogenous production, exogenous uptake, storage in fat droplets and cellular efflux of cholesterol.

Why do we need such an elaborate regulatory system? Because cholesterol is toxic at high levels. On the other hand, the function of cholesterol in cell membranes is essential for raft-driven membrane sub-compartmentalization. The employment of raft clustering enables the cell to create sub-compartments when needed — an incredibly dynamic device for organizing membrane function. So cholesterol levels have to be regulated within a narrow window to avoid toxicity and to keep rafts functional.

Boxes

		Box 1 | The structure of membrane lipids and their influence on membrane properties
		A lipid molecule within a monolayer can be characterized as having an 'effective' molecular shape, which is determined by its molecular structure and its interactions with other molecules in the monolayer. The effective molecular shape of a lipid is determined by the shape of the monolayer in whatever phase the lipid spontaneously forms in aqueous solution349. The curvature of a spontaneously formed monolayer consisting of a single lipid is commonly considered to be the effective molecular spontaneous curvature for that lipid. The spontaneous curvatures of lipid molecules influence the elastic behaviour of the membrane, and determine the energy cost of the drastic remodelling of membrane structure that is required for membrane fusion and fission. Panel A of the figure shows some typical examples of spontaneously formed lipid phases, and the corresponding effective molecular shapes and spontaneous curvatures of the constitutent lipid molecules. Panel B depicts the stalkpore pathway of lipid bilayer fusion, starting from two separate membranes (part Ba). Point-like merger of the proximal monolayers of the two membranes occurs during the first stage of the pathway, while the distal monolayers remain separate. The resulting structure is called the fusion stalk2 (part Bb). The stalk then expands into a hemifusion diaphragm (part Bc), followed by pore formation in the diaphragm, which completes the fusion process (part Bd). The first steps of fusion — formation of the fusion stalk and the hemifusion diaphragm — are promoted by the cone-like (type II) lipids, whereas the completion of fusion through formation of the fusion pore is promoted by the inverted cone-like (type I) lipids 350 , 351 , 352 , 353 , 354 , 355. Panel coutlines the proposed pathway of membrane fission via hemifission356. Fission starts with a constricted membrane neck (part Ca). Hemifission results from 'self-fusion' of the inner monolayer of the neck (part Cb), with subsequent decay of the hemifission structure leading to the separation of the two membranes and the completion of fission (part Cc). According to a theoretical analysis356, the two stages of the process — hemifission and complete membrane separation — are promoted by the inverted cone-like (type I) lipids. DAG, diacylglycerol; DOPE, dioleoylphosphatidylethanolamine; LPA, lysophosphatidic acid; LPC, lysophosphatidylcholine; PC, phosphatidylcholine. Part B reproduced, with permission, from REF.354© Annual Reviews (2003). Part C reproduced, with permission, from REF.356© The Biophysical Society (2003). Michael M. Kozlov

		Box 2 | Membrane pickets and fences
		The entire plasma membrane is partitioned into compartments of 40–23 nm in diameter (size is dependent on cell type); virtually all of the molecules in the membrane seem to undergo short-term confined diffusion within these compartments and long-term 'hop' diffusion between these compartments (hop events are shown by arrows in part a). Molecules within a membrane compartment move as fast as those incorporated into artificial membranes, such as liposomes, but frequently bounce off the compartment boundaries. The frequency of hop between the compartments is on average once every 1–100 ms (depending on the cell type and type of molecule). Therefore, the fluid-mosaic model of Singer-Nicolson is perfectly adequate to describe diffusion events in a membrane region of around 10 nm2 (part a). However, when considering membrane events that involve larger regions, a shift from this model might be required. An important consequence of the membrane partitioning is that the rate of diffusion cannot be described by a single diffusion coefficient, so all of the diffusion coefficients obtained by methods such as fluorescence recovery after photobleaching (FRAP), or by using single-molecule techniques at slow rates, must be thought of as 'effective' diffusion coefficients, which might be useful only when the time window involved is specified. The bottom line here is that no membrane molecules undergo simple Brownian diffusion in the plasma membrane. For around 30 years, membrane biophysicists have been puzzled by observations that the diffusion rates of membrane molecules in the plasma membrane are reduced by a factor of 5-50 compared with those found in artificial membranes, such as giant liposomes. However, this can be easily explained by the partitioned fluid model: molecular diffusion rates obtained using FRAP depend on the compartment size and the hop rate between compartments. The 'fences' that separate the membrane compartments are likely to be formed from the actin-based membrane skeleton (see part b and main text). Transmembrane proteins anchor to and line up along this membrane skeleton mesh, like pickets along a fence (part c), creating thin strips along the membrane skeleton where the diffusion rate of membrane molecules is substantially reduced. The fences only act on transmembrane proteins (part b), whereas the pickets can affect the diffusion of all the molecules incorporated in the plasma membrane. This partitioning is likely to be important for maintaining spatial information during signal transduction in the plasma membrane. Akihiro Kusumi

Biography Boxes

		Anthony G. Lee
		Professor of Biochemistry, School of Biological Sciencés, University of Southampton, Bassett Crescent East, Southampton SO16 7PX, UK e-mail: agl@soton.ac.uk

		Ben de Kruijff
		Professor, Biochemistry of Membranes, Utrecht University, Kruytgebouw, Padualaan 8, 3584 CH Utrecht, The Netherlands e-mail: b.dekruijff@chem.uu.nl

		Michael A. Edidin
		Professor of Biology, Medicine and Pathology, BiologyDepartment, Johns Hopkins University, 3400 North Charles Street, Baltimore, Maryland 21218, USA e-mail: edidin@jhu.edu

		Joshua Zimmerberg
		Chief, Laboratory of Cellular and Molecular Biophysics, National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland 20892-1855, USA e-mail: joshz@helix.nih.gov

		Michael M. Kozlov
		Professor, Department of Physiology and Pharmacology, Sackler Faculty of Medicine, Tel Aviv University, Ramat Aviv, 69978 Tel Aviv, Israel mail: michk@post.tau.ac.il

		Arlene Albert
		Professor, Department of Molecular and Cell Biology, North Eagleville Road, University of Connecticut, Storrs, Connecticut 06269, USA e-mail: arlene.albert@uconn.edu

		Sean Munro
		Principal Investigator, Medical Research Council Laboratory of Molecular Biology, Hills Road, Cambridge CB2 2QH, UK e-mail: sean@mrc-lmb.cam.ac.uk

		Tony Magee
		Professor of Membrane Biology, Section of Cell and Molecular Biology, Division of Biomedical Sciences, Imperial College London, Sir Alexander Fleming Building, Room 110, Exhibition Road, London SW7 2AZ, UK e-mail: t.magee@imperial.ac.uk

		Kai Simons
		Professor and Executive Director, Max-Planck Institute for Molecular Cell Biology and Genetics, Pfotenhauerstraβe 108, 01307 Dresden, Germany e-mail: simons@mpi-cbg.de

		Anthony Watts
		Professor of Biochemistry, Oxford Biomembrane Structure Unit, Biochemistry Department, University of Oxford, South Parks Road, Oxford OX1 3QU, UK e-mail: anthony.watts@bioch.ox.ac.uk

		Akihiro Kusumi
		Professor of Biophysics, The Research Centre for Nano Medical Engineering, The Institute for Frontier Medical Sciences, Kyoto University, Shougoin Kawahara-cho 53, Sakyo-ku, Kyoto 606-8507, Japan e-mail: akusumi@frontier.kyoto-u.ac.jp

		John Silvius
		Professor of Biochemistry and Coordinator of the Chemical Biology Graduate Program, Department of Biochemistry, McIntyre Medical Building, Room 820, 3655 Promenade Sir William Osler, Montreal, Quebec, Canada H3G 1Y6 e-mail: john.silvus@mcgill.ca

		Erwin London
		Professor of Biochemistry and Cell Biology and Professor of Chemistry, 420 Life Sciences Building, Department of Biochemistry and Cell Biology, Stony Brook University, Stony Brook, New York 11794-5215, USA e-mail: erwin.london@stonybrook.edu

		Stuart McLaughlin
		Professor of Physiology and Biophysics, Department of Physiology & Biophysics, HSC, Stony Brook University, Stony Brook, New York 11794-8661, USA e-mail: SMCL@epo.som.sunysb.edu

		Gunnar von Heijne
		Principal Investigator, Department of Biochemistry and Biophysics, The Arrhenius Laboratories for Natural Sciences, Stockholm University, SE-106 91 Stockholm, Sweden e-mail: gunnar@dbb.su.se

		Michael J. Caplan
		Professor of Cellular and Molecular Physiology and Cell Biology, Department of Cellular and Molecular Physiology, Yale University School of Medicine, 333 Cedar Street, PO Box 208026 , New Haven, Connecticut 06520-8026, USA e-mail: michael.caplan@yale.edu

		Lewis C. Cantley
		Professor of Systems Biology, Harvard Medical School, 250 Longwood Avenue, SGMB 221, Boston, Massachusetts 02115, USA e-mail: lewis_cantley@hms.harvard.edu

		Judith Klumperman
		Professor of Cell Biology and Head of Cell Microscopy Center, UMC Utrecht, Cell Microscopy Center, G02.525, PO Box 85500, NL-3508 GA Utrecht, The Netherlands e-mail: j.klumperman@lab.azu.nl

		Boris N. Kholodenko
		Professor and Director of Computational Cell Biology, Department of Pathology, Anatomy and Cell Biology, Thomas Jefferson University, JAH, 1020 Locust St, Philadelphia, Pennsylvania 19107, USA e-mail: boris.kholodenko@jefferson.edu

		Vivek Malhotra
		Professor of Cell and Developmental Biology, Cell and Developmental Biology, UC San Diego, Pacific Hall, Rm 2222A, La Jolla, California 92093-0347, USA e-mail: malhotra@biomail.ucsd.edu

		Jennifer Lippincott-Schwartz
		Chief, Section on Organelle Biology, Cell Biology and Metabolism Branch, National Institute of Child Health and Human Development, National Institutes of Health, Building 18T, Room 101, 18 Library Drive, Bethesda, Maryland 20892-5430, USA e-mail: jlippin@helix.nih.gov

		Suzanne Pfeffer
		Professor and Chairman of Biochemistry, Department of Biochemistry, 279 Campus Drive B400, Stanford, California 94305-5307, USA e-mail: pfeffer@stanford.edu

		Sandra L. Schmid
		Professor and Chairman, Department of Cell Biology, The Scripps Research Institute, 10550 North Torrey Pines Road, La Jolla, California 92037, USA e-mail: slschmid@scripps.edu

		Ritva Tikkanen
		Assistant Professor of Biochemistry, Institute of Biochemistry II, Hs. 75, University Clinic of Frankfurt, Theodor-Stern-Kai 7, D-60590 Frankfurt am Main, Germany e-mail: tikkanen@biochem2.de

		Scott D. Emr
		Professor of Cellular and Molecular Medicine, Department of Cellular and Molecular Medicine & Howard Hughes Medical Institute, UCSD School of Medicine, Palade Laboratories Rm 318, 9500 Gilman Drive, La Jolla, California 92093-0668, USA e-mail: semr@ucsd.edu

		Nicoletta Kahya
		Biotechnological Centre, Tatzberg 47-51, D-01307 Dresden, Germany e-mail: kahya@mpi-cbg.de

		Patricia Bassereau
		Directrice de Recherche CNRS, PhysicoChimie Curie , Section de Recherche de l'Institut Curie, 11 Rue Pierre et Marie Curie, 75231 Paris Cedex 5, France e-mail: patricia.bassereau@curie.fr

		Daniel Axelrod
		Professor of Physics and Research Scientist in Biophysics, University of Michigan, 930 North University, Ann Arbor, Michigan 48109, USA e-mail: daxelrod@umich.edu

		Ernest M. Wright
		Mellinkoff Professor in Medicine, Department of Physiology, David Geffen School of Medicine at UCLA, 53-263 CHS, Box 951751, Los Angeles, California 90095-1751, USA e-mail: ewright@mednet.ucla.edu

		Richard A. Bond
		Associate Professor of Pharmacology, Department of Pharmacological and Pharmaceutical Sciences, University of Houston, College of Pharmacy, 521 Science and Research Bldg 2, Houston, Texas 77204-5037, USA e-mail: rabond@uh.edu

		Frederick R. Maxfield
		Professor and Chair, Department of Biochemistry, Weill Medical College, Cornell University, 1300 York Avenue, New York, New York 10021, USA e-mail: frmaxfie@med.cornell.edu

References

1.	Lee, A. G. How lipids affect the activities of integral membrane proteins. Biochim. Biophys. Acta 1666, 62–87 (2004).
2.	Gould, G. W. , McWhirter, J. M. , East, J. M. & Lee, A. G. Effects of diet on the function of sarcoplasmic reticulum. Biochem. J. 245, 751–755 (1987).
3.	Los, D. A. & Murata, N. Membrane fluidity and its roles in the perception of environmental signals. Biochim. Biophys. Acta 1666, 142–157 (2004).
4.	Gennis, R. B
5.	Van Voorst, F. & De Kruijff, B. Role of lipids in the translocation of proteins across membranes. Biochem. J. 347, 601–612 (2000).
6.	Palsdottir, H. & Hunte, C. Lipids in membrane protein structures. Biochim. Biophys. Acta 1666, 2–18 (2004).
7.	Hilditch, T. P. (1956).
8.	Sinensky, M. Homeoviscous adaptation — a homeostatic process that regulates the viscosity of membrane lipids in. Escherichia coli. Proc. Natl Acad. Sci. USA 71, 522–525 (1974).
9.	Hazel, J. R. Thermal adaptation in biological membranes: is homeoviscous adaptation the explanation?. Annu. Rev. Physiol. 57, 19–42 (1995).
10.	Abe, F. & Iida, H. Pressure-induced differential regulation of the two tryptophan permeases Tat1 and Tat2 by ubiquitin ligase Rsp5 and its binding proteins, Bul1 and Bul2. Mol. Biol. Cell 23, 7566–7584 (2003).
11.	Jans, D. A (1997).
12.	Thompson, G. A. Jr. Lipid molecular species remodeling and membrane fluidity. Biochem. Soc. Trans. 17, 286–289 (1989).
13.	Kasai, R. Molecular control of membrane properties during temperature acclimation. Membrane fluidity regulation of fatty acid desaturase action?. Biochemistry 15, 5228–5233 (1976).
14.	Gilberto, R. Unravelling the signal-transduction network in B lymphocytes. Nature 420, 708–710 (2002).
15.	Thewke, D. , Kramer, M. & Sinensky, M. S. Transcriptional homeostatic control of membrane lipid composition. Biochem. Biophys. Res. Commun. 273, 1–4 (2000).
16.	Ciapa, B. & Whitaker, M. Two phases of inositol polyphosphate and diacylglycerol production at fertilisation. FEBS Lett. 195, 347–351 (1986).
17.	Hagar, R. E. , Burgstahler, A. D. , Nathanson, M. H. & Ehrlich, B. E. Type III InsP3 receptor channel stays open in the presence of increased calcium. Nature 396, 81–84 (1998).
18.	Saunders, C. M PLCζ : a sperm-specific trigger of Ca2+ oscillations in eggs and embryo development. Development 129, 3533–3544 (2002).
19.	Schroit, A. J. , Madsen, J. W. & Tanaka, Y. In vivo recognition and clearance of red blood cells containing phosphatidylserine in their plasma membranes. J. Biol. Chem. 260, 5131–5138 (1985).
20.	Fadok, V. A. Exposure of phosphatidylserine on the surface of apoptotic lymphocytes triggers specific recognition and removal by macrophages. J. Immunol. 148, 2207–2216 (1992).
21.	Schlegel, R. A. & Williamson, P. Phosphatidylserine, a death knell. Cell Death Diff. 8, 551–563 (2001).
22.	Pelham, H. R. B. & Munro, S. Sorting of membrane proteins in the secretory pathway. Cell 75, 603–605 (1993).
23.	Munro, S. Localization of proteins to the Golgi apparatus. Trends Cell Biol. 8, 11–15 (1998).
24.	Finkelstein, A. Water and nonelectrolyte permeability of lipid bilayer membranes. J. Gen. Physiol. 68, 127–135 (1976).
25.	Finkelstein, A. Nature of the water permeability increase induced by antidiuretic hormone (ADH) in toad urinary bladder and related tissues. J. Gen. Physiol. 68, 137–143 (1976).
26.	Dunham, P. B. , Cass, A. , Trinkaus, J. P. & Bennett, M. V. L. Water permeability of fundulus egg. Biol. Bulletin 139, 420–421 (1970).
27.	Hess, S. T Quantitative electron microscopy and fluorescence spectroscopy of the membrane distribution of influenza hemagglutinin. J. Cell Biol.
28.	Kozlov, M. M. , Leikin, S. L. , Chernomordik, L. V. , Markin, V. S. & Chizmadzhev, Y. A. Stalk mechanism of vesicle fusion. Intermixing of aqueous contents. Eur. Biophys. J. 17, 121–129 (1989).
29.	Kozlov, M. M. & Markin, V. S. Possible mechanism of membrane fusion. Biofizika 28, 255–261 (1983).
30.	Kozlovsky, Y. & Kozlov, M. Stalk model of membrane fusion: solution of energy crisis. Biophys. J. 88, 882–895 (2002).
31.	Kozlovsky, Y. & Kozlov, M. Membrane fission: model for intermediate structures. Biophys. J. 85, 85–96 (2003).
32.	Shemesh, T. , Luini, A. , Malhotra, V. , Burger, K. N. & Kozlov, M. M. Prefission constriction of Golgi tubular carriers driven by local lipid metabolism: a theoretical model. Biophys. J. 85, 3813–3827 (2003).
33.	Andelman, D. 603–642 (1995).
34.	Parsegian, V. A. & Rand, R. P. 643–690 (1995).
35.	Evans, E. 723–754 (1995).
36.	Chernomordik, L. V. & Kozlov, M. M. Protein–lipid interplay in fusion and fission of biological membranes. Annu. Rev. Biochem. 72, 175–207 (2003).
37.	Chernomordik, L. , Kozlov, M. & Zimmerberg, J. Lipids in biological membrane fusion. J. Mem. Biol. 146, 1–14 (1995).
38.	Kozlovsky, Y. & Kozlov, M. Membrane fission: model for intermediate structures. Biophys. J. 85, 85–96 (2003).
39.	Almeida, P. F. F. & Vaz, W. L. C. 305–357 (1995).
40.	McConnell, H. M. & Radhakrishnan, A. Condensed complexes of cholesterol and phospholipids. Biochim. Biophys. Acta 1610, 159–173 (2003).
41.	Simons, K. & Vaz, W. L. Model systems, lipid rafts, and cell membranes. Annu. Rev. Biophys. Biomol. Struct. 33, 269–295 (2004).
42.	Straume, M. & Litman, B. J. Equilibrium and dynamic structure of large, unilamellar, unsaturated acyl chain phosphatidylcholine vesicles. Higher order analysis of 1,6-diphenyl-1,3,5-hexatriene and 1-[4-trimethylammonio]-6-phenyl-1,3,5-hexatriene anisotropy decay. Biochemistry 26, 5113–5120 (1987).
43.	Straume, M. & Litman, B. J. Equilibrium and dynamic bilayer structural properties of unsaturated acyl chain phosphatidylcholine-cholesterol-rhodopsin recombinant vesicles and rod outer segment disk membranes as determined from higher order analysis of fluorescence anisotropy decay. Biochemistry 27, 7723–7733 (1988).
44.	Straume, M. Interconversions of metarhodopsins I and II: a branched photointermediate decay model. 29, 9135–9142 (1990).
45.	Straume, M. & Litman, B. J. Influence of cholesterol on equilibrium and dynamic bilayer structure of unsaturated acyl chain phosphatidylcholine vesicles as determined from higher order analysis of fluorescence anisotropy decay. Biochemistry 26, 5121–5126 (1987).
46.	Mitchell, D. Modulation of metarhodopsin formation by cholesterol-induced ordering of bilayers. Biochemistry. 29, 9143–9149 (1990).
47.	Mitchell, D. C. , Straume, M. & Litman, B. J. Role of sn-1-saturated, sn-2-polyunsaturated phospholipids in control of membrane receptor conformational equilibrium: effects of cholesterol and acyl chain unsaturation on the metarhodopsin I–metarhodopsin II equilibrium. Biochemistry. 31, 662–670 (1992).
48.	Hamosh, M. & Salem, N. Jr. Vol. 74, 106–120 (1998).
49.	Mitchell, D. C. Why is DHA essential for nervous system function?. Biochem. Soc. Trans. 26, 365–370 (1998).
50.	Birch, E. E. A randomized controlled trial at early dietary supply of long-chain polyunsaturated fatty acids and mental development in infants. Dev. Med. Child Neurol. 42, 174–181 (2000).
51.	Polozova, A. & Litman, B. J. Cholesterol dependent recruitment of di22:6-PC by a G-protein-coupled receptor into lateral domains. Biophys. J. 79, 2632–2643 (2001).
52.	Watts, A. , Volovski, I. D. & Marsh, D. Rhodopsin–lipid associations in bovine rod outer segment membranes. Identification of immobilized lipid by spin-labels. Biochemistry 18, 5006–5013 (1979).
53.	Albert, A. D. & Yeagle, P. L. Phospholipid domains in bovine retinal rod outer segment disk membranes. Proc. Natl Acad. Sci. USA 80, 7188–7191 (1983).
54.	Hessel, E. Light-induced reorganization of phospholipids in rod disc membranes. J. Biol. Chem. 276, 2538–2543 (2001).
55.	Holthuis, J. C. & Levine, T. Lipid traffic: floppy drives and a superhighway. Nature Rev. Mol. Cell Biol. 6, 209–220 (2005).
56.	Kol, M. A. , De Kroon, A. I. P. M. , Killian, J. A. & de Kruijff, B. Transbilayer movement of phopholipids in biogenic membranes. Biochemistry 43, 2673–2681 (2004).
57.	Pomorski, T. , Holthuis, J. C. , Herrmann, A. & van Meer, G. Tracking down lipid flippases and their biological functions. J. Cell Sci. 117, 805–813 (2004).
58.	Cronan, J. E. Bacterial membrane lipids: where do we stand?. Annu. Rev. Microbiol. 57, 203–224 (2003).
59.	Marsh, D. & Pali, T. The protein–lipid interface: perspectives from magnetic resonance and crystal structures. Biochim. Biophys. Acta 1666, 118–141 (2004).
60.	Misquitta, L. V. Membrane protein crystallization in lipidic mesophases with tailored bilayers. Structure 12, 2113–2124 (2004).
61.	Luecke, H. Structure of bacteriorhodopsin at 1.55 Å resolution. J. Mol. Biol. 291, 899–911 (1999).
62.	Tsukihara, T. The whole structure of the 13-subunit oxidized cytochrome c oxidase at 2.8 Å. Science. 272, 1136–1144 (1996).
63.	Ferguson-Miller, S. Mammalian cytochrome c oxidase, a molecular monster subdued. Science 272, 1125– (1996).
64.	Grigorieff, N. Electron-crystallographic refinement of the structure of bacteriorhodopsin. J. Mol. Biol. 259, 393–421 (1996).
65.	Membrane Proteins of Known Structure [online], <http:// blanco.biomol.uci.edu/Membrane_Proteins_xtal.html>.
66.	Marsh, D. Lipid-binding proteins: structure of the phospholipid ligands. Protein Sci. 12, 2109–2117 (2003).
67.	Standfuss, J. & K¨hlbrandt, W. The three isoforms of the light-harvesting complex II: spectroscopic features, trimer formation, and functional roles. J. Biol. Chem. 279, 36884–36891 (2004).
68.	McDermott, G. Crystal structure of an integral membrane light-harvesting complex from photosynthetic bacteria. Nature 374, 517–521 (1995).
69.	Watts, A. 351–370
70.	Sternberg, B. The essential role of Halobacterium halobium lipids in 2D-array formation of bacteriorhodopsin. Biochim. Biophys. Acta. 1108, 21–30 (1992).
71.	Sternberg, B. , Watts, A. & Cejka, Z. Lipid induced modulation of the protein packing in two-dimensional crystals of bacteriorhodopsin. J. Struct. Biol. 110, 196–204 (1993).
72.	Watts, A. Bacteriorhodopsin: the mechanism of 2D-array formation and the structure of the retinal in the protein. Biophys. Chem. 55, 137–151 (1994).
73.	Cartailler, J. & Luecke, H. X-ray crystallographic analysis of lipid–protein interactions in the bacteriorhodopsin purple membrane. Annu. Rev. Biophy. Biomol. Struct. 32, 285–310 (2003).
74.	Jordan, P. Three-dimensional structure of cyanobacterial photosystem I at 2.5 Å resolution. Nature 411, 909–917 (2001).
75.	Jones, M. R. Protein–lipid interactions in the purple bacterial reaction centre. Biochim. Biophys. Acta 1565, 206–214 (2002).
76.	Gether, U. Different binding epitopes on the NK1 receptor for substance-P and a nonpeptide antagonist. Nature 362, 345–348 (1993).
77.	Elling, C. E. , Nielsen, S. M. & Schwartz, T. W. Conversion of antagonist-binding site to metal-ion site in the techykinin NK-1 receptor. Nature 374, 74–77 (1995).
78.	Schambye, H. T. Differentiation between binding-sites for angiotensin-II and nonpeptide antagonists on the angiotensin-II type-1 receptors. Proc. Natl Acad. Sci. USA 91, 7046–7050 (1994).
79.	Landin, J. S. , Katragadda, M. & Albert, A. D. Destabilization of rhodopsin and opsin by proteolytic cleavage rod outer segment disk membranes. Biochemistry 40, 11176–11183 (2001).
80.	Lee, A. G. How lipids interact with an intrinsic membrane protein: the case of the calcium pump. Biochim. Biophys. Acta 1376, 381–390 (1998).
81.	Killian, J. A. Hydrophobic mismatch between proteins and lipids in membranes. Biochim. Biophys. Acta 1376, 401–416 (1998).
82.	Gil, T. Theoretical analysis of protein organization in lipid membranes. Biochim. Biophys. Acta 1376, 245–266 (1998).
83.	Watts, A. 687–740 (1993).
84.	Watts, A. Protein–lipid interactions. Nature 294, (1981).
85.	Lee, A. G. Lipid–protein interactions in biological membranes: a structural perspective. Biochim. Biophys. Acta. 1612, 1–40 (2003).
86.	Sheets, E. D. , Holowka, D. & Baird, B. Membrane organization in immunoglobulin E receptor signaling. Curr. Opin. Chem. Biol. 3, 95–99 (1999).
87.	Janes, P. W. , Ley, S. C. , Magee, A. I. & Kabouridis, P. S. The role of lipid rafts in T cell antigen receptor (TCR) signalling. Semin. Immunol. 12, 23–34 (2000).
88.	Kusumi, A. , Koyama-Honda, I. & Suzuki, K. Molecular dynamics and interactions for creation of signal-induced stabilized rafts from small unstable steady-state rafts. Traffic 5, 213–230 (2004).
89.	Bowie, J. U. Border crossing. Nature 433, 367–369 (2004).
90.	Killian, J. A. & von Heijne, G. How proteins adapt to a membrane-water interface. Trends Biochem. Sci. 25, 429–434 (2000).
91.	Deol, S. D. , Bond, P. J. , Domene, C. & Sansom, M. S. Lipid–protein interactions of integral membrane proteins: a comparative simulation study. Biophys. J. 87, 3737–3749 (2004).
92.	Driscoll, P. C. Solving the FYVE domain–PtdIns(3)P puzzle. Nature Struct. Biol. 8, 287–290 (2001).
93.	Kutateladze, T. & Overduin, M. Structural mechanism of endosome docking by the FYVE domain. Science 291, 1793–1796 (2001).
94.	Baraldi, E. Structure of the PH domain from Bruton’s tyrosine kinase in complex with inositol 1,3,4,5 tetrakisphosphate. Structure 7, 449–460 (1999).
95.	Misra, S. & Hurley, J. H. Crystal structure of a phosphatidylinositol 3-phosphate-specific membrane-targeting motif, the FYVE domain of Vps27p. Cell 97, 657–666 (1999).
96.	Lietzke, S. E. Structural basis of 3-phosphoinositide recognition by pleckstrin homology domains. Mol. Cell 6, 385–394 (2000).
97.	Bravo, J. The crystal structure of the PX domain from p40(phox) bound to phosphatidylinositol 3-phosphate. Mol. Cell 8, 829–839 (2001).
98.	Zhou, C. Z. Crystal structure of the yeast Phox homology (PX) domain protein Grd19p complexed to phosphatidylinositol-3-phosphate. J. Biol. Chem. 278, 50371–50376 (2003).
99.	Hurley, J. H. & Misra, S. Signaling and subcellular targeting by membrane-binding domains. Annu. Rev. Biophys. Biomol. Struct. 29, 49–79 (2000).
100.	McAuley, K. E. Structural details of an interaction between cardiolipin and an integral membrane protein. Proc. Natl Acad. Sci. USA 96, 14706–14711 (1999).
101.	Chernomordik, L. V. Lysolipids reversibly inhibit Ca2+-, GTP- and pH-dependent fusion of biological membranes. FEBS Lett. 318, 71–76 (1993).
102.	Chernomordik, L. An early stage of membrane fusion mediated by the low pH conformation of influenza hemagglutinin depends upon membrane lipids. J. Cell Biol. 136, 81–93 (1997).
103.	Basanez, G. Bax-type apoptotic proteins porate pure lipid bilayers through a mechanism sensitive to intrinsic monolayer curvature. J. Biol. Chem. 277, 49360–49365 (2002).
104.	Zimmerberg, J. Are the curves in all the right places?. Traffic 1, 366–368 (2000).
105.	Chernomordik, L. , Chanturiya, A. , Green, J. & Zimmerberg, J. The hemifusion intermediate and its conversion to complete fusion: regulation by membrane composition. Biophys. J. 69, 922–929 (1995).
106.	Kozlovsky, Y. , Zimmerberg, J. & Kozlov, M. M. Orientation and interaction of oblique cylindrical inclusions embedded in a lipid monolayer: a theoretical model for viral fusion peptides. Biophys. J. 87, 999–1012 (2004).
107.	Helfrich, W. Elastic properties of lipid bilayers: theory and possible experiments. Z. Naturforsch. 28, 693–703 (1973).
108.	Leikin, S. , Kozlov, M. M. , Fuller, N. L. & Rand, R. P. Measured effects of diacylglycerol on structural and elastic properties of phospholipid membranes. Biophys. J. 71, 2623–2632 (1996).
109.	Hamm, M. & Kozlov, M. Tilt model of inverted amphiphilic mesophases. Eur. Phys. J. B 6, 519–528 (1998).
110.	Hamm, M. & Kozlov, M. Elastic energy of tilt and bending of fluid membranes. Eur. Phys. J. E 3, 323–335 (2000).
111.	Rand, R. P. & Fuller, N. L. Structural dimensions and their changes in a reentrant hexagonal-lamellar transition of phospholipids. Biophys. J. 2127–2138 (1994).
112.	Evans, E. & Skalak, R (1980).
113.	Seddon, J. M. & Templer, R. H. 97–160 (1995).
114.	Arbuzova, A. , Murray, D. & McLaughlin, S. MARCKS, membranes, and calmodulin: kinetics of their interaction. Biochim. Biophys. Acta. 1376, 369–379 (1998).
115.	Cantor, R. S. The lateral pressure profile in membranes: a physical mechanism of general anesthesia. Biochemistry 36, 2339–2344 (1997).
116.	Van den Brink-van der Laan, E. , Chupin, V. , Killian, J. A. & De Kruijff, B. Small alcohols destabilize the KcsA tetramer via their effect on the membrane lateral pressure. Biochemistry 43, 5937–5942 (2004).
117.	Lee, A. G. Lipids and their effects on membrane proteins: evidence against a role for fluidity. Prog. Lipid Res. 30, 323–348 (1991).
118.	Cantor, R. S. Lipid composition and the lateral pressure profile in bilayers. Biophys. J. 76, 2625–2639 (1999).
119.	Attard, G. S. , Templer, R. H. , Smith, W. S. , Hunt, A. N. & Jackowski, S. Modulation of CTP: phosphocholine cytidylyltransferase by membrane curvature elastic stress. Proc. Natl Acad. Sci USA 97, 9032–9036 (2000).
120.	Botelho, A. V. , Gibson, N. J. , Thurmond, R. L. , Wang, Y. & Brown, M. F. Conformational energetics of rhodopsin modulated by nonlamellar-forming lipids. Biochemistry 41, 6354–6368 (2002).
121.	Yeagle, P. L (1987).
122.	Yeagle, P. L. 121–146 (1988).
123.	Yeagle, P. L. Cholesterol and the cell membrane. Biochim. Biophys. Acta Biomembrane Rev. 822, 267–287 (1985).
124.	Boesze-Battaglia, K. & Albert, A. Cholesterol modulation of photoreceptor function in bovine rod outer segments. J. Biol. Chem. 265, 20727–20730 (1990).
125.	Levine, T. P. , Wiggins, C. A. & Munro, S Inositol phosphorylceramide synthase is located in the Golgi apparatus of. Saccharomyces cerevisiae. Mol. Biol. Cell 11, 2267–2281 (2000).
126.	Edidin, M. Patches, posts and fences: proteins and plasma membrane domains. Trends Cell Biol. 2, 376–380 (1992).
127.	Gawrisch, K. 147–172 (2005).
128.	Gaus, K. Visualizing lipid structure and raft domains in living cells with two-photon microscopy. Proc. Natl Acad. Sci. USA 100, 15554–15559 (2003).
129.	Fujiwara, T. , Ritchie, K. , Murakoshi, H. , Jacobson, K. & Kusumi, A. Phospholipids undergo hop diffusion in compartmentalized cell membrane. J. Cell Biol. 157, 1071–1082 (2002).
130.	Murase, K. Ultrafine membrane compartments for molecular diffusion as revealed by single molecule techniques. Biophys. J. 86, 4075–4093 (2004).
131.	Kusumi, A. Paradigm shift of the plasma membrane concept from the two-dimensional continuum fluid to the partitioned fluid: high-speed single-molecule tracking of membrane molecules. Annu. Rev. Biophys. Biomol. Struct. 17, 3–21 (2005).
132.	de Kruijff, B. (ed.) Reviews on biomembranes special issue: lipid–protein interactions. Biochim. Biophys. Acta (1998).
133.	Choe, H. & Anderson, R. Unique molecular-species composition of glycerolipids of frog rod outer segments. Exp. Eye Res. 51, 159–165 (1990).
134.	Niu, S.-L. Reduced G protein-coupled signaling efficiency in retinal rod outer segments in response to n-3 fatty acid deficiency. J. Biol. Chem. 279, 31098–31104 (2004).
135.	Bell, M. & Dick, J. Molecular-species composition of the major diacyl glycerophospholipids from muscle, liver, retina and brain of cod. Gadus morhua). Lipids. 26, 565–573 (1991).
136.	Stinson, A. , Wiegand, R. & Anderson, R. Fatty-acid and molecular-species compositions of phospholipids and diacylglycerols from rat retinal membranes. Exp. Eye Res. 52, 213–218 (1991).
137.	Aveldano, M. I. Phospholipid species containing long and very long polyenoic fatty acids remain with rhodopsin after hexane extraction of photoreceptor membranes. Biochemistry 27, 1229–1239 (1988).
138.	Nie, S. & Zare, R. N. Optical detection of single molecules. Annu. Rev. Biophys. Biomol. Struct. 26, 567–596 (1997).
139.	Kusumi, A. Cell biology: hopping fences. Nature 433, 680– (2005).
140.	Valiyaveetil, F. I. , Zhou, Y. & MacKinnon, R. Lipids in the structure, folding and function of the KcsA K+ channel. Biochemistry 41, 10771–10777 (2002).
141.	Wang, J. , Gambhir, A. , McLaughlin, S. & Murray, D. A computational model for the electrostatic sequestration of PI(4,5)P2 by membrane-adsorbed basic peptides. Biophys. J. 86, 1969–1986 (2004).
142.	Kawasaki, K. , Yin, J. J. , Subczynski, W. K. , Hyde, J. S. & Kusumi, A. Pulse EPR detection of lipid exchange between protein-rich raft and bulk domains in the membrane: methodology development and its application to studies of influenza viral membrane. Biophys. J. 80, 738–748 (2001).
143.	Anderson, R. G. W. & Jacobson, K. A role for lipid shells in targeting proteins to caveolae, rafts, and other lipid domains. Science 296, 1821–1825 (2002).
144.	Pike, L. Lipid rafts: bringing order to chaos. J. Lipid Res. 44, 655–667 (2003).
145.	Gambhir, A. Electrostatic sequestration of PIP2 on phospholipid membranes by basic/aromatic regions of proteins. Biophys. J. 86, 2188–2207 (2004).
146.	McLaughlin, S. , Wang, J. , Gambhir, A. & Murray, D. PIP(2) and proteins: interactions, organization, and information flow. Annu. Rev. Biophys. Biomol. Struct. 31, 151–175 (2002).
147.	Mitaku, S Interaction stabilizing tertiary structure of bacteriorhodopsin studied by denaturation experiments. Proteins 22, 350–362 (1995).
148.	Ulmschneider, M. B. & Sansom, M. S. P. Amino acid distributions in integral membrane protein structures. Biochim. Biophys. Acta Biomembr. 1512, 1–14 (2001).
149.	Hessa, T. Recognition of transmembrane helices by the endoplasmic reticulum translocon. Nature 433, 377–381 (2005).
150.	Ubarretxena-Belandia, I. & Engelman, D. M. Helical membrane proteins: diversity of functions in the context of simple architecture. Curr. Opin. Struct. Biol. 11, 370–376 (2001).
151.	Ridder, A. N. J.A. Importance of hydrophobic matching for spontaneous insertion of a single-spannning membrane protein. Biochemistry 41, 4946–4952 (2002).
152.	Van Dalen, A. , Van der Laan, M. , Driessen, A. J. M. , Killian, J. A. & de Kruijff, B. Components required for membrane assembly of newly synthesized K+ channel KcsA. FEBS Lett. 511, 51–58 (2002).
153.	Schneider, D. & Engelman, D. M. Motifs of two small residues can assist but are not sufficient to mediate transmembrane helix interactions. J. Mol. Biol. 343, 799–804 (2004).
154.	Muth, T. R. & Caplan, M. J. Transport protein trafficking in polarized cells. Annu. Rev. Cell. Dev. Biol. 19, 333–366 (2003).
155.	Edidin, M. The state of lipid rafts: from model membranes to cells. Annu. Rev. Biophys. Biomol. Struct. 32, 257–283 (2003).
156.	Balamuth, F. , Leitenberg, D. , Unternaehrer, J. , Mellman, I. & Bottomly, K. Distinct patterns of membrane microdomain partitioning in TH 1 and TH 2 cells. Immunity 15, 729–738 (2001).
157.	Ikonens, E. & Simons, K. Protein and lipid sorting from the trans-Golgi network to the plasma membrane in polarized cells. Semin. Cell Dev. Biol. 9, 503–509 (1998).
158.	van der Wouden, J. M. , Maier, O. , van IJzendoorn, S. C. & Hoekstra, D. Membrane dynamics and the regulation of epithelial cell polarity. Int. Rev. Cytol. 226, 127–164 (2003).
159.	Heerssen, H. M. & Segal, R. A. Location, location, location: a spatial view of neurotrophin signal transduction. Trends Neurosci. 25, 160–165 (2002).
160.	Howe, C. L. , Valletta, J. S. , Rusnak, A. S. & Mobley, W. C. NGF signaling from clathrin-coated vesicles: evidence that signaling endosomes serve as a platform for the Ras–MAPK pathway. Neuron 32, 801–814 (2001).
161.	Weible, M. W. & Hendry, I. A. What is the importance of multivesicular bodies in retrograde axonal transport in vivo . J. Neurobiol. 58, 230–243 (2004).
162.	Sorkin, A. & Von Zastrow, M. Signal transduction and endocytosis: close encounters of many kinds. Nature Rev. Mol. Cell Biol. 3, 600–614 (2002).
163.	Burke, P. , Schooler, K. & Wiley, H. S. Regulation of epidermal growth factor receptor signalling by endocytosis and intracellular trafficking. Mol. Biol. Cell 12, 1897–1910 (2001).
164.	Simons, K. & Toomre, D. Lipid rafts and signal transduction. Nature Rev. Mol. Cell Biol. 1, 31–39 (2000).
165.	Haugh, J. M. & Lauffenburger, D. A. Physical modulation of intracellular signaling processes by locational regulation. Biophys. J. 72, 2014–2031 (1997).
166.	Bray, D. Signaling complexes: biophysical constraints on intracellular communication. Annu. Rev. Biophys. Biomol. Struct. 27, 59–75 (1998).
167.	Bivona, T. G. Phospholipase Cγ activates Ras on the Golgi apparatus by means of RasGRP1. Nature 424, 694–698 (2003).
168.	Kholodenko, B. N. MAP kinase cascade signaling and endocytic trafficking: a marriage of convenience?. Trends Cell Biol. 12, 173–177 (2002).
169.	Vieira, A. V. , Lamaze, C. & Schmid, S. L. Control of EGF receptor signaling by clathrin-mediated endocytosis. Science 274, 2086–2089 (1996).
170.	Miaczynska, M. , Pelkmans, L. & Zerial, M. Not just a sink: endosomes in control of signal transduction. Curr. Opin. Cell Biol. 16, 400–406 (2004).
171.	Kholodenko, B. N. Four-dimensional organization of protein kinase signaling cascades: the roles of diffusion, endocytosis and molecular motors. J. Exp. Biol. 206, 2073–2082 (2003).
172.	Palade, G. Intracellular aspects of the process of protein synthesis. Science 189, 347–358 (1975).
173.	Bloble, G. & Dobbertsetin, B. Transfer of proteins across membranes. J. Cell Biol. 67, 835–851 (1975).
174.	Burgess, T. L. & Kelly, R. B. Constitutive and regulated secretion of proteins. Ann. Rev. Cell Biol. 3, 243–294 (1987).
175.	Pfeffer, S. R. RabGTPases: specifying and deciphering organelle identity and function. Trends Cell Biol. 11, 487–491 (2001).
176.	Ellgaard, L. & Helenius, A. Quality control in the endoplasmic reticulum. Nature Rev. Mol. Cell Biol. 4, 181–191 (2003).
177.	Mostov, K. , Su, T. & ter Beest, M. Polarized epithelial membrane traffic: conservation and plasticity. Nature Cell Biol. 5, 287–293 (2003).
178.	Rodriguez-Boulan, E. , Musch, A. & Le Bivic, A. Epithelial trafficking: new routes to familiar places. Curr. Opin. Cell Biol. 16, 436–442 (2004).
179.	Gruenberg, J. & Stenmark, H. The biogenesis of multivesicular endosomes. Nature Rev. Mol. Cell Biol. 5, 317–323 (2004).
180.	Ghosh, P. , Dahms, N. M. & Kornfeld, S. Mannose 6-phosphate receptors: new twists in the tale. Nature Rev. Mol. Cell Biol. 4, 202–212 (2003).
181.	Pfeffer, S. Membrane domains in the secretory and endocytic pathways. Cell 112, 507–517 (2003).
182.	Altan-Bonnet, N. , Sougrat, R. & Lippincott-Schwartz, J. Molecular basis for Golgi biogenesis and biogenesis. Trends Cell Biol. 16, 364–372 (2004).
183.	Geuze, H. J. Involvement of the endoplasmic reticulum in peroxisome formation. Mol. Biol. Cell 14, 2900–2907 (2003).
184.	Maxfield, F. & McGraw, T. Endocytic recycling. Nature Rev. Mol. Cell Biol. 5, 121–132 (2004).
185.	Glick, B. S. Can the Golgi form de novo ?. Nature Rev. Mol. Cell Biol. 3, 615–619 (2002).
186.	Munro, S. Organelle identity and the organization of membrane traffic. Nature Cell Biol. 6, 469–472 (2004).
187.	Simonsen, A. The role of phosphoinositides in membrane transport. Curr. Opin. Cell Biol. 13, 485–492 (2001).
188.	Odorizzi, G. , Babst, M. & Emr, S. D. Phosphoinositide signaling and the regulation of membrane trafficking in yeast. Trends Biochem. Sci. 25, 229–235 (2000).
189.	Roth, M. G. Phosphoinositides in constitutive membrane traffic. Physiol. Rev. 84, 699–730 (2004).
190.	Czech, M. P. Dynamics of phosphoinositides in membrane retrieval and insertion. Annu. Rev. Physiol. 65, 791–815 (2003).
191.	Lemmon, M. A. Phosphoinositide recognition domains. Traffic 4, 201–213 (2003).
192.	Hurley, J. H. & Meyer, T. Subcellular targeting by membrane lipids. Curr. Opin. Cell Biol. 13, 146–152 (2001).
193.	Zerial, H. & McBride, H. Rab proteins as membrane organizers. Nature Rev. Mol. Cell Biol. 2, 107–117 (2001).
194.	Ganley, I. , Carroll, K. , Bittova, L. & Pfeffer, S. R. Rab9 regulates late endosome size and requires effector interaction for stability. Mol. Biol. Cell 15, 5420–5430 (2004).
195.	Bock, J. B. , Matern, H. T. , Peden, A. A. & Scheller, R. H. A genomic perspective on membrane compartment organization. Nature 409, 839–841 (2001).
196.	Miaczynska, M. & Zerial, M. Mosaic organization of the endocytic pathway. Exp. Cell Res. 272, 8–14 (2002).
197.	Sonnichsen, B. , De Renzis, S. , Nielsen, E. , Rietdorf, J. & Zerial, M. Distinct membrane domains on endosomes in the recycling pathway visualized by multicolor imaging of Rab4, Rab5, and Rab11. J. Cell Biol. 149, 901–914 (2000).
198.	de Renzis, S. , Sonnichsen, B. & Zerial, M. Divalent Rab effectors regulate the sub-compartmental organization and sorting of early endosomes. Nature Cell Biol. 4, 124–133 (2002).
199.	London, M. & London, E. Ceramide selectively displaces cholesterol from ordered lipid domains (rafts): implications for lipid raft structure and function. J. Biol. Chem. 279, 997–10004 (2004).
200.	Gomez-Mouton, C. Segregation of leading-edge and uropod components into specific lipid rafts during T cell polarization. Proc. Natl Acad. Sci. USA 98, 9642–9647 (2001).
201.	Resh, M. D. Membrane targeting of lipid modified signal transduction proteins. Subcell. Biochem. 37, 217–232 (2004).
202.	Mollenhaue, H. H. & Morre, D. J. Perspectives on Golgi apparatus form and function. J. Electron Microsc. Tech. 17, 2–14 (1991).
203.	Glick, B. S. & Malhotra, V. The curious status of the Golgi apparatus. Cell 95, 883–889 (1998).
204.	Miaczynska, M. APPL proteins link Rab5 to nuclear signal transduction via an endosomal compartment. Cell 116, 445–456 (2004).
205.	Mironov, A. A. Small cargo proteins and large aggregates can traverse the Golgi by a common mechanism without leaving the lumen of cisternae. J. Cell Biol. 155, 1225–1238 (2001).
206.	Hirschberg, K. Kinetic and morphological analysis of secretory protein traffic in living cells. J. Cell Biol. 143, 1485–1503 (1998).
207.	Deitz, S. B. , Rambourg, A. , Kepes, F. & Franzusoff, A. Sec7p directs the transitions required for yeast golgi biogenesis. Traffic 1, 172–183 (2000).
208.	Lanoix, J. J. GTP hydrolysis by Arf-1 mediates sorting and concentration of Golgi resident enzymes into functional COPI vesicles. EMBO J. 18, 4935–4948 (1999).
209.	Majoul, I. , Straub, M. , Hell, S. W. , Duden, R. & Soling, H. D. KDEL-cargo regulates interactions between proteins involved in COPI vesicle traffic: measurements in living cells using FRET. Dev. Cell. 1, 139–153 (2001).
210.	Rein, U. , Andag, U. , Duden, R. , Schmitt, H. D. & Spang, A. ARF-GAP-mediated interaction between the ER–Golgi v-SNAREs and the COPI coat. J. Cell Biol. 157, 395–404 (2002).
211.	Liu, W. , Moriyama, K. , Phair, R. , Duden, R. & Lippincott-Schwartz, J. In vivo dynamics of ARFGAP1 and its functional interaction with Arf1 and coatomer on Golgi membranes. J. Cell Biol.
212.	Lippincott-Schwartz, J. & Patterson, G. Development and use of fluorescent protein markers in living cells. Science 300, 87–91 (2003).
213.	Lippincott-Schwartz, J. , Snapp, E. & Kenworthy, A. Studying protein dynamics in living cells. Nature Rev. Mol. Cell Biol. 2, 444–456 (2001).
214.	Hell, S. W. , Dyba, M. & Jakobs, S. Concepts for nanoscale resolution in fluorescence microscopy. Curr. Opin. Neurobiol. 14, 599–609 (2004).
215.	Zhang, J. , Campbell, R. E. , Ting, A. Y. & Tsien, R. Y. Creating new fluorescent probes for cell biology. Nature Rev. Mol. Cell Biol. 3, 906–918 (2002).
216.	Saxton, M. J. & Jacobson, K. Single particle tracking: application to membrane dynamics. Annu. Rev. Biophys. Biomol. Struct. 26, 373–399 (1997).
217.	de Lange, F. Cell biology beyond the diffraction limit: near-field scanning optical microscopy. J. Cell Sci. 114, 4153–4160 (2001).
218.	Michalet, X. Quantum dots for live cells, in vivo imaging, and diagnostics. Science 307, 538–544 (2005).
219.	Han, M. , Gao, X. , Su, J. & Nie, S. Quantum-dot-tagged microbeads for multiplexed optical coding of biomolecules. Nature Biotechnol. 19, 631–635 (2001).
220.	Sharma, P. Nanoscale organization of multiple GPI-anchored proteins in living cell membranes. Cell 116, 577–589 (2004).
221.	Schwille, P. Fluorescence correlation spectroscopy and its potential for intracellular applications. Cell Biochim. Biophys. 34, 383–408 (2001).
222.	Heinze, K. G. , Koltermann, A. & Schwille, P. Simultaneous two-photon excitation of distinct labels for dual-color fluorescence cross-correlation analysis. Proc. Natl Acad. Sci. USA 97, 10377–10382 (2000).
223.	Stephens, D. J. & Allan, V. J. Light microscopy techniques for live cell imaging. Science 300, 82–86 (2003).
224.	Roux, A. Role of curvature and phase transition in lipid sorting and fission of membrane tubules. EMBO J. 24, 1537–1545 (2005).
225.	Axelrod, D. , Koppel, D. E. , Schlessinger, J. , Elson, E. & Webb, W. W. Mobility measurement by analysis of fluorescence photobleaching recovery kinetics. Biophys. J. 16, 1055–1069 (1976).
226.	Wolf, D. E. Theory of fluorescence recovery after photobleaching measurements on cylindrical surfaces. Biophys. J. 61, 487–493 (1992).
227.	Bulseco, D. A. & Wolf, D. E. Fluorescence correlation spectroscopy: molecular complexing in solution and in living cells. Methods Cell Biol. 72, 465–498 (2003).
228.	Bacia, K. & Schwille, P. A dynamic view of cellular processes by in vivo fluorescence auto- and cross-correlation spectroscopy. Methods 29, 74–85 (2003).
229.	Waharte, F. , Brown, C. M. , Coscoy, S. , Coudrier, E. & Amblard, F. A two-photon FRAP analysis of the cytoskeleton dynamics in the microvilli of intestinal cells. Biophys. J. 88, 1467–1478 (2005).
230.	Frazer, B. H. , Girasole, M. , Wiese, L. M. , Franz, T. & De Stasio, G. Spectromicroscope for the photoelectron imaging of nanostructures with X-rays (SPHINX): performance in biology, medicine and geology. Ultramicroscopy 99, 87–94 (2004).
231.	Johansson, G. A. Exploring the use of soft X-ray microscopy for imaging subcellular structures of the inner ear. J. Microsc. 215, 203–212 (2004).
232.	Ortega, R. , Bohic, S. , Tucoulou, R. , Somogyi, A. & Devès, G. Microchemical element imaging of yeast and human cells using synchrotron X-ray microprobe with Kirkpatrick–Baez optics. Analytical Chem. 76, 309–314 (2004).
233.	Kemner, K. M. Elemental and redox analysis of single bacterial cells by X-ray microbeam analysis. Science 306, 686–687 (2004).
234.	Bastiaens, P. I. & Pepperkok, R. Observing proteins in their natural habitat: the living cell. Trends Biochem Sci. 25, 631–637 (2000).
235.	Bonifacino, J. S. & Lippincott-Schwartz, J. Coat proteins: shaping membrane transport. Nature Rev. Mol. Cell Biol. 4, 409–414 (2003).
236.	Martinez-Menarguez, J. A. , Geuze, H. J. , Slot, J. W. & Klumperman, J. Vesicular tubular clusters between the ER and Golgi mediate concentration of soluble secretory proteins by exclusion from COPI-coated vesicles. Cell 98, 81–90 (1999).
237.	Geuze, H. J. , Slot, J. W. , Strous, G. J. , Lodish, H. F. & Schwartz, A. L. Intracellular site of asialoglycoprotein receptor-ligand uncoupling: double-label immunoelectron microscopy during receptor-mediated endocytosis. Cell 32, 277–287 (1983).
238.	Klumperman, J. Differences in the endosomal distributions of the two mannose 6-phosphate receptors. J. Cell Biol. 121, 997–1010 (1993).
239.	Koster, A. J. & Klumperman, J. Electron microscopy in cell biology: integrating structure and function. Nature Rev. Mol. Cell Biol. Suppl. S6–S10 (2003).
240.	Mironov, A. A. , Polishchuk, R. S. & Luini, A. Visualizing membrane traffic in vivo by combined video fluorescence and 3D electron microscopy. Trends Cell Biol. 10, 349–353 (2000).
241.	Oorschot, V. , de Wit, H. , Annaert, W. G. & Klumperman, J. A novel flat-embedding method to prepare ultrathin cryosections from cultured cells in their in situ orientation. J. Histochem. Cytochem. 50, 1067–1080 (2002).
242.	McNally, J. G. , Karpova, T. , Cooper, J. & Conchello, J. A. Three-dimensional imaging by deconvolution microscopy. Methods 19, 373–385 (1999).
243.	Scalettar, B. A. , Swedlow, J. R. , Sedat, J. W. & Agard, D. A. Dispersion, aberration and deconvolution in multi-wavelength fluorescence images. J. Microscopy 182, 50–60 (1996).
244.	Sako, Y. , Minoghchi, S. & Yanagida, T. Single molecule imaging of EGFR signalling on the surface of living cells. Nature Cell Biol. 2, 168–172 (2000).
245.	Lewis, A. Near-field optics: from subwavelength illumination to nanometric shadowing. Nature Biotechnol. 21, 1378–1386 (2003).
246.	Juiskaitis, R. 21–44 (2003).
247.	Lemasters, J. J. 66–87 (2001).
248.	Labarthet, F. L. & Shen, Y. R. 169–196 (2003).
249.	Axelrod, D. Total internal reflection fluorescence microscopy in cell biology. Methods Enzymol. 361, 1–33 (2003).
250.	Jares-Erijman, E. A. & Jovin, T. M. FRET imaging. Nature Biotechnol. 21, 1387–1395 (2003).
251.	Lakowicz, J. (1999).
252.	Mattheyses, A. , Hoppe, A. & Axelrod, D. Polarized fluorescence resonance energy transfer microscopy. Biophys. J. 87, 2787–2797 (2004).
253.	Murakoshi, H. Single-molecule imaging analysis of Ras activation in living cells. Proc. Natl Acad. Sci. USA 101, 7317–7322 (2004).
254.	Glebov, O. O. & Nichols, B. J. Lipid raft proteins have a random distribution during localized activation of the T-cell receptor. Nature Cell Biol. 6, 238–243 (2004).
255.	Klar, T. A. Fluorescence microscopy with diffraction resolution barrier broken by stimulated emission. Proc. Natl Acad. Sci. USA 97, 8206–8210 (2000).
256.	Gustafsson, M. G. L. Extended resolution fluorescence microscopy. Curr. Opin. Struct. Biol. 9, 627–634 (1999).
257.	McIntosh, R. , Nicastro, D. & Mastronarde, D. New views of cells in 3D: an introduction to electron tomography. Trends Cell Biol. 15, 43–51 (2005).
258.	Eskandari, S. , Wright, E. M. , Kreman, M. , Starace, D. M. & Zampighi, G. A. Structural analysis of cloned plasma membrane proteins by freeze-fracture electron microscopy. Proc. Natl Acad. Sci. USA 95, 11235–11240 (1998).
259.	Zampighi, G. A. Structure of functional single AQP0 channels in phospholipid membranes. J. Mol. Biol. 325, 201–210 (2003).
260.	Zampighi, L. M. , Kavanau, C. L. & Zampighi, G. A. The Kohonen self-organizing map: a tool for the clustering and alignment of single particles imaged using random conical tilt. J. Struct. Biol. 146, 368–380 (2004).
261.	Lanzavecchia, S. Conical tomography of freeze-fracture replicas: a method for the study of integral membrane proteins inserted in phospholipid bilayers. J. Struct. Biol. 149, 87–98 (2005).
262.	Kumar, S. & Hoh, J. H. Probing the machinery of intracellular trafficking with the atomic force microscope. Traffic 2, 746–756 (2001).
263.	Kuo, S. C. Using optics to measure biological forces and mechanics. Traffic 2, 757–763 (2001).
264.	White, J. G. , Squirrell, J. M. & Eliceiri, K. W. Applying multiphoton imaging to the study of membrane dynamics in living cells. Traffic 2, 775–780 (2001).
265.	Holtom, G. R. Achieving molecular selectivity in imaging using multiphoton Raman spectroscopy techniques. Traffic 2, 781–788 (2001).
266.	Elson, E. L. Fluorescence correlation spectroscopy measures molecular transport in cells. Traffic 2, 789–796 (2001).
267.	Edidin, M. Near-field scanning optical microscopy, a siren call to biology. Traffic 2, 797–803 (2001).
268.	Palade, G. E. & Siekevitz, P. Liver microsomes; an integrated morphological and biochemical study. J. Biophys. Biochem. Cytol. 25, 171–200 (1956).
269.	De Camilli, P. & Warren, G. Membranes and sorting, getting there. Curr. Opin. Cell Biol. 11, 423–424 (1999).
270.	Kahya, N. , Scherfeld, D. , Bacia, K. , Poolman, B. & Schwille, P. Probing lipid mobility of raft-exhibiting model membranes by fluorescence correlation spectroscopy. J. Biol. Chem. 278, 28109–28115 (2003).
271.	Devaux, P. F. & Morris, R. Transmembrane asymmetry and lateral domains in biological membranes. Traffic 5, 241–246 (2004).
272.	Kwik, J. Membrane cholesterol, lateral mobility, and the phosphatidylinositol 4,5-biphosphate-dependent organization of cell actin. Proc. Natl Acad. Sci. USA 100, 13964–13969 (2003).
273.	Manno, S. , Takakuwa, Y. & Mohandas, N. Identification of a functional role for lipid asymmetry in biological membranes: phosphatidylserine-skeletal protein interactions modulate membrane stability. Proc. Natl Acad. Sci. USA 99, 1943–1948 (2002).
274.	Zimmerberg, J. & Parsegian, V. A. Polymer inaccessible volume changes during opening and closing of a voltage-dependent ionic channel. Nature 323, 36–39 (1986).
275.	Baumgart, T. , Hess, S. T. & Webb, W. W. Imaging coexisting fluid domains in biomembrane models coupling curvature and line tension. Nature 425, 821–824 (2003).
276.	Kooijman, E. E. Spontaneous curvature of phosphatidic acid and lysophosphatidic acid. Biochemistry 44, 2097–2102 (2005).
277.	Leduc, C. Cooperative extraction of membrane nanotubes by molecular motors. Proc. Natl Acad. Sci. USA 101, 17096–17101 (2004).
278.	Devaux, P. F. Is lipid translocation involved during endoand exocytosis?. Biochimie 82, 497–509 (2000).
279.	Lipowsky, R. & Sackmann, E. (1995).
280.	Evans, E. & Rawicz, W. Entropy-driven tension and bending elasticity in condensed-fluid membranes. Phys. Rev. Lett. 64, 2094–2097 (1990).
281.	Luisi, P. L. & Walde, P. (2000).
282.	Derényi, I. , J¨licher, F. & Prost, J. Formation and interaction of membrane tubes. Phys. Rev. Lett. 88, (2002).
283.	Upadhyaya, A. & Sheetz, M. P. Tension in tubulovesicular networks of Golgi and endoplasmic reticulum membranes. Biophys. J. 86, 2923–2928 (2004).
284.	Limozin, L. & Sackmann, E. Polymorphism of cross-linked actin networks in giant vesicles. Phys. Rev. Let. 89, (2002).
285.	Girard, P. , Julicher, F. & Prost, J. Fluid membranes exchanging material with external reservoirs. Eur. Phys. J. E14, 387–394 (2004).
286.	Manneville, J.-B. , Bassereau, P. , Ramaswamy, S. & Prost, J. Active membrane fluctuations studied by micropipet aspiration. Phys. Rev. E64, (2001).
287.	Mayor, S. & Rao, M. Rafts: scale-dependent, active lipid organization at the cell surface. Traffic 5, 231–240 (2004).
288.	Montal, M. Formation of bimolecular membranes from lipid monolayers. Meth. Enzymol. 32B, 545–554 (1974).
289.	Pautot, S. , Frisken, B. J. & Weitz, D. A. Engineering asymmetric vesicles. Proc. Natl Acad. Sci. USA 100, 10718–10721 (2003).
290.	Lippincott-Schwartz, J. , Roberts, T. H. & Hirschberg, K. Secretory protein trafficking and organelle dynamics in living cells. Annu. Rev. Cell Dev. Biol. 16, 557–589 (2000).
291.	Singer, S. J. & Nicolson, G. L. The fluid mosaic model of the structure of cell membranes. Science 175, 720–731 (1972).
292.	Horrobin, D. F. Modern biomedical research: an internally self-consistent universe with little contact with medical reality?. 2, 151–154 (2003).
293.	Schuck, S. & Simons, K. Polarized sorting in epithelial cells: raft clustering and the biogenesis of the apical membrane. J. Cell Sci. 117, 5955–5964 (2004).
294.	Rodriguez-Boulan, E. Organization of vesicular trafficking in epithelia. Nature Rev. Mol. Cell Biol. 6, 233–247 (2005).
295.	Rodriguez-Boulan, E. & Powell, S. K. Polarity of epithelial and neuronal cells. Annu. Rev. Cell Biol. 8, 395–427 (1992).
296.	Mostov, K. E. , Verges, M. & Altschuler, Y. Membrane traffic in polarized epithelial cells. Curr. Opin. Cell Biol. 12, 483–490 (2000).
297.	Nelson, W. J. & Yeaman, C. Protein trafficking in the exocytic pathway of polarized epithelial cells. Trends Cell. Biol. 11, 483–486 (2001).
298.	Folsch, H. , Ohno, H. , Bonifacino, J. S. & Mellman, I. A novel clathrin adaptor complex mediates basolateral targeting in polarized epithelial cells. Cell 99, 189–198 (1999).
299.	Brown, D. A. & London, E. Functions of lipid rafts in biological membranes. Annu. Rev. Cell Dev. Biol. 14, 111–136 (1998).
300.	Simons, K. & Ikonen, E. Functional rafts in cell membranes. Nature 387, 569–572 (1997).
301.	Munro, S. Lipid rafts, elusive or illusive?. Cell 115, 377–388 (2003).
302.	Haines, T. H. Do sterols reduce proton and sodium leaks through lipid bilayers?. Prog. Lipid Res. 40, 299–324 (2001).
303.	Ramstedt, B. & Slotte, J. P. Membrane properties of sphingomyelins. FEBS Lett. 531, 33–37 (2002).
304.	Paladino, S. Protein oligomerization modulates raft partitioning and apical sorting of GPI-anchored proteins. J. Cell Biol. 167, 699–709 (2004).
305.	Colanzi, A. , Sutterlin, C. & Malhotra, V. Mitosis specific Golgi fragmentation: how and why?. Curr. Opin. Cell Biol. 15, 462–467 (2003).
306.	Warren, G. Membrane partitioning during cell division. Ann. Rev. Biochem. 62, 323–348 (1993).
307.	Charest, A. Oncogenic targeting of an activated tyrosine kinase to the Golgi apparatus. Proc. Natl Acad. Sci. USA 100, 916–921 (2003).
308.	Sutterlin, C. , Hsu, P. , Mallabiabarrena, A. & Malhotra, V. Fragmentation and dispersal of the pericentriolar Golgi complex is required for entry into mitosis in mammalian cells. Cell 109, 359–369 (2002).
309.	Preisinger, C. Plk1 docking to GRASP65 phosphorylated by Cdk1 suggests a mechanism for Golgi checkpoint signalling. EMBO J. 24, 753–765 (2005).
310.	Altan-Bonnet, N. , Phair, R. D. , Polishchuk, R. S. , Weigart, R. & Lippincott-Schwartz, J. Role of Arf1 in mitotic Golgi disassembly, chromosome segregation and cytokinesis. Proc. Natl Acad. Sci. USA 100, 13314–13319 (2003).
311.	Palade, G. E. The organization of living matter. Proc. Natl Acad. Sci. USA 52, 613–634 (1964).
312.	Knodler, L. A. & Steele-Mortimer, O. Taking possession: biogenesis of the Salmonella-containing vacuole. Traffic 4, 587–599 (2003).
313.	Kleijmeer, M. Reorganization of multivesicular bodies regulates MHC class II antigen presentation by dendritic cells. J. Cell Biol. 155, 53–63 (2001).
314.	Barbero, P. , Bittova, L. & Pfeffer, S. R. Visualization of Rab9-mediated vesicle transport from endosomes to the trans Golgi in living cells. J. Cell Biol. 156, 511–518 (2002).
315.	Lamaze, C. Interleukin 2 receptors and detergent-resistant membrane domains define a clathrin-independent endocytic pathway. Mol. Cell 7, 661–671 (2001).
316.	Mayor, S. & Riezman, H. Sorting GPI-anchored proteins. Nature Rev. Mol. Cell Biol. 5, 110–120 (2004).
317.	Nusse, R. Wnts and Hedgehogs: lipid-modified proteins and similarities in signaling mechanisms at the cell surface. Development 130, 5297–5305 (2003).
318.	Rocks, O. An acylation cycle regulates localization and activity of palmitoylated Ras isoforms. Science 10 Feb 2005
319.	Linder, M. E. & Deschenes, R. J. Model organisms lead the way to protein palmitoyltransferases. J. Cell Sci. 117, 521–526 (2004).
320.	Roth, A. F. The yeast DHHC cysteine-rich domain protein Akr1p is a palmitoyl transferase. J. Cell Biol. 159, 23–28 (2002).
321.	Dietrich, L. E. P. & Ungermann, C. On the mechanism of protein palmitoylation. EMBO Rep. 5, 1053–1057 (2004).
322.	Marty, A. & Finkelstein, A. Pores formed in lipid bilayer membranes by nystatin, Differences in its one-sided and two-sided action. J. Gen. Physiol. 65, 515–526 (1975).
323.	Holz, R. & Finkelstein, A. The water and nonelectrolyte permeability induced in thin lipid membranes by the polyene antibiotics nystatin and amphotericin B. J. Gen. Physiol. 56, 125–145 (1970).
324.	Leonard, D. M. Ras farnesyltransferase: a new therapeutic target. J. Med. Chem. 40, 2971–2990 (1997).
325.	Magee, T. & Seabra, M. C. Fatty acylation and prenylation of proteins: what’s hot in fat. Curr. Opin. Cell Biol. 17, 190–196 (2005).
326.	Torgersen, K. M. Release from tonic inhibition of T cell activation through transient displacement of C-terminal Src kinase (Csk) from lipid rafts. J. Biol. Chem. 276, 29313–29318 (2001).
327.	Sim, A. T. & Scott, J. D. Targeting of PKA, PKC and protein phosphatases to cellular microdomains. Cell. Calcium. 26, 209–217 (1999).
328.	Parton, R. G. & Hancock, J. F. Lipid rafts and plasma membrane microorganization: insights from Ras. Trends Cell. Biol. 14, 141–147 (2004).
329.	Hancock, J. F. Ras proteins: different signals from different locations. Nature Rev. Mol. Cell Biol. 4, 373–384 (2003).
330.	Dorn, G. W. & Mochly-Rosen, D. Intracellular transport mechanisms of signal transducers. Annu. Rev. Physiol. 64, 407–429 (2002).
331.	Kholodenko, B. N. , Hoek, J. B. & Westerhoff, H. V. Why cytoplasmic signalling proteins should be recruited to cell membranes. Trends Cell. Biol. 10, 173–178 (2000).
332.	Xiong, W. & Ferrell, J. E. Jr. A positive-feedback-based bistable ‘memory module’ that governs a cell fate decision. Nature 426, 460–465 (2003).
333.	Levchenko, A. , Bruck, J. & Sternberg, P. W. Scaffold proteins may biphasically affect the levels of mitogen-activated protein kinase signaling and reduce its threshold properties. Proc. Natl Acad. Sci. USA 97, 5818–5823 (2000).
334.	Baumann, C. A. CAP defines a second signalling pathway required for insulin-stimulated glucose transport. Nature 407, 202–207 (2000).
335.	Berridge, M. J. , Lipp, P. & Bootman, M. D. The versatility and universality of calcium signalling. Nature Rev. Mol. Cell Biol 1, 11–21 (2000).
336.	Hilgemann, D. W. , Feng, S. & Nasuhoglu, C. The complex and intriguing lives of PIP2 with ion channels and transporters. Sci STKE (2001).
337.	He, C. , Hobert, M. , Friend, L. & Carlin, C. The epidermal growth factor receptor juxtamembrane domain has multiple basolateral plasma membrane localization determinants, including a dominant signal with a polyproline core. J. Biol. Chem. 277, 38284–38293 (2002).
338.	Horton, A. C. & Ehlers, M. D. Neuronal polarity and trafficking. Neuron 40, 277–295 (2003).
339.	Krogsgaard, M. , Huppa, J. B. , Purbhoo, M. A. & Davis, M. M. Linking molecular and cellular events in T-cell activation and synapse formation. Semin. Immunol. 15, 307–315 (2003).
340.	Kusumi, A. , Ike, H. , Nakada, C. , Murase, K. & Fujiwara, T. Single-molecule tracking of membrane molecules: plasma membrane compartmentalization and dynamic assembly of raft-philic signaling molecules. Semin. Immunol. 17, 3–21 (2005).
341.	Lange, Y. , Ye, J. & Steck, T. L. How cholesterol homeostasis is regulated by plasma membrane cholesterol in excess of phospholipids. Proc. Natl Acad. Sci. USA 101, 11664–11667 (2004).
342.	Sheets, E. D. , Holowka, D. & Baird, B. Critical role for cholesterol in Lyn-mediated tyrosine phosphorylation of FceRI and their association with detergent-resistant membranes. J. Cell Biol. 145, 877–887 (1999).
343.	Porter, F. D. Human malformation syndromes due to inborn errors of cholesterol synthesis. Curr. Opin. Pediatr. 15, 607–613 (2003).
344.	Benz, R. , Hallmann, D. , Poralla, K. & Eibl, H. Interaction of hopanoids with phosphatidylcholines containing oleic and ω -cyclohexyldodecanoic acid in lipid bilayer membranes. Chem. Phys. Lipids 34, 7–24 (1983).
345.	Berry, A. M. Hopanoid lipids compose the Frankia vesicle envelope, presumptive barrier of oxygen diffusion to nitrogenase. Proc. Natl Acad. Sci. USA 90, 6091–6094 (1993).
346.	Coetzee, T. , Suzuki, K. & Popko, B. New perspectives on the function of myelin galactolipids. Trends Neurosci. 21, 126–130 (1998).
347.	Hill, W. G. & Zeidel, M. L. Reconstituting the barrier properties of a water-tight epithelial membrane by design of leaflet-specific liposomes. J. Biol. Chem. 275, 30176–30185 (2000).
348.	Mann, R. K. & Beachy, P. A. Novel lipid modifications of secreted protein signals. Annu. Rev. Biochem. 73, 891–923 (2004).
349.	Luzzati, V. 71–123 (1968).
350.	Kozlov, M. M. & Markin, V. S. Possible mechanism of membrane fusion. Biofizika 28, 255–261 (1983).
351.	Kozlov, M. M. , Leikin, S. L. , Chernomordik, L. V. , Markin, V. S. & Chizmadzhev, Y. A. Stalk mechanism of vesicle fusion. Intermixing of aqueous contents. Eur. Biophys. J. 17, 121–129 (1989).
352.	Chernomordik, L. , Kozlov, M. & Zimmerberg, J. Lipids in biological membrane fusion. J. Membr. Biol. 146, 1–14 (1995).
353.	Kozlovsky, Y. , Chernomordik, L. V. & Kozlov, M. M. Lipid intermediates in membrane fusion: formation, structure, and decay of hemifusion diaphragm. Biophys. J. 83, 2634–2651 (2002).
354.	Chernomordik, L. V. & Kozlov, M. M. Protein–lipid interplay in fusion and fission of biological membranes. Annu. Rev. Biochem. 72, 175–207 (2003).
355.	Kozlovsky, Y. & Kozlov, M. Stalk model of membrane fusion: solution of energy crisis. Biophys. J. 88, 882–895 (2002).
356.	Kozlovsky, Y. & Kozlov, M. Membrane fission: model for intermediate structures. Biophys. J. 85, 85–96 (2003).
357.	Gerlich, D. & Ellenberg, J. 4D imaging to assay complex dynamics in live specimens. Nature Cell Biol. S14–S19 (2003).
358.	Denk, W. , Strickler, J. H. & Webb, W. W. Two-photon scanning fluorescence microscopy. Science 248, 73–76 (1990).
359.	Zipfel, W. R. , Williams, R. M. & Webb, W. W. Nonlinear magic: multiphoton microscopy in the biosciences. Nature Biotechnol. 21, 1369–1377 (2003).
360.	Gustafsson, M. G. L. Extended resolution fluorescence microscopy. Curr. Opin. Struct. Biol. 9, 627–34 (1999).
361.	Gustafsson, M. G. L. , Agard, D. A. & Sedat, J. W. 3D wide-field light microscopy with better than 100 nm axial resolution. J. Microsc. 195, 10–16 (1999).
362.	Egner, A. & Hell, S. W. Fluorescence microscopy with super-resolved optical sections. Trends Cell Biol. 15, 207–215 (2005).
363.	Nagorni, M. & Hell, S. W. 4Pi-confocal microscopy provides three dimensional images of the microtubule network with 100- to 150-nm resolution. J. Struct. Biol. 123, 236–247 (1998).
364.	Egner, A. , Jakobs, S. & Hell, S. W. Fast 100-nm resolution three-dimensional microscope reveals structural plasticity of mitochondria in live yeast. Proc. Natl Acad. Sci. USA 99, 3370–3375 (2002).
365.	Dyba, M. & Hell, S. W. Photostability of a fluorescent marker under pulsed excited-state depletion through stimulated emission. Appl. Opt. 42, 5123–5129 (2003).
366.	Peleg, G. , Lewis, A. , Linial, M. & Loew, L. M. Nonlinear optical measurement of membrane potential around single molecules at detected cellular sites. Proc. Natl Acad. Sci. USA 96, 6700–6704 (1999).
367.	Moreaux, L. , Sandre, O. & Mertz, J. Membrane imaging by second-harmonic generation microscopy. J. Opt. Soc. Am. B 17, 1685–1694 (2000).
368.	Cheng, J.-X. , Pautot, S. , Weitz, D. A. & Xie, X. S. Ordering of water molecules between phospholipids bilayers visualized by coherent anti-Stokes Raman scattering microscopy. Proc. Natl Acad. Sci. USA 100, 9826–9830 (2003).
369.	Potma, E. O. & Xie, X. S. Direct visualization of lipid phase segregation in single lipid bilayers with coherent anti-Stokes Raman scattering microscopy. ChemPhysChem 6, 77–79 (2005).
370.	Axelrod, D. Total internal reflection fluorescence microscopy in cell biology. Traffic 2, 764–774 (2001).
371.	Keller, P. , Toomre, D. , Diaz, E. , White, J. & Simons, K. Multicolor imaging of post-Golgi sorting and trafficking in living cells. Nature Cell Biol. 3, 140–149 (2001).
372.	Lippincott-Schwartz, J. , Snapp, E. & Kenworthy, A. Studying protein dynamics in living cells. Nature Rev. Mol. Cell Biol. 2, 444–456 (2001).
373.	Kenworthy, A. K. & Edidin, M. Distribution of glycosylphosphatidylinositol-anchored protein at the apical surface of MDCK cells examined at a resolution of <100 Å using imaging fluorescence resonance energy transfer. J. Cell Biol. 142, 69–84 (1998).
374.	Varma, R. & Mayor, S. GPI-anchored proteins are organized in sub-micron domains at the cell surface. Nature 394, 798–801 (1998).
375.	Xu, Y. , Piston, D. W. & Johnson, C. H. A bioluminescence resonance energy transfer (BRET) system: application to circadian clock proteins. Proc. Natl Acad. Sci. USA 96, 151–156 (1999).
376.	Sharma, P. Nanoscale organization of multiple GPI-anchored proteins in living cell membranes. Cell 116, 577–589 (2004).
377.	Rocheleau, J. V. , Edidin, M. & Piston, D. W. Intrasequence GFP in class I MHC molecules, a rigid probe for the fluorescence anisotropy measurements of the membrane environment. Biophys. J. 84, 4078–4086 (2003).
378.	Benninger, R. K. P. , Önfelt, B. , Neil, M. A. A. , Davis, D. M. & French, P. M. W. Fluorescence imaging of two-photon linear dichroism: cholesterol depletion disrupts molecular orientation in cell membrane. Biophys. J. 88, 609–622 (2005).
379.	Lippincott-Schwartz, J. , Altan-Bonnet, N. & Patterson, G. H. Photobleaching and photoactivation: following dynamics in living cells. Nature Cell Biol. S7–S14 (2003).
380.	Politz, J. C. Use of caged fluorophores to track macromolecular movement in living cells. Trends Cell Biol. 9, 284–287 (1999).
381.	Schwille, P. Fluorescence correlation spectroscopy and its potential for intracellular applications. Cell Biochem. Biophys. 34, 383–408 (2001).
382.	Haustein, E. & Schwille, P. Single-molecule spectroscopic methods. Curr. Opin. Struct. Biol. 14, 531–540 (2004).
383.	Heinze, K. , Koltermann, A. & Schwille, P. Simultaneous two-photon excitation of distinct labels for dual-color fluorescence crosscorrelation analysis. Proc. Natl Acad. Sci. USA 97, 10377–10382 (2000).
384.	Thompson, N. L. , Lieto, A. M. & Allen, N. W. Recent advances in fluorescence correlation spectroscopy. Curr. Opin. Struct. Biol. 12, 634–641 (2002).
385.	Müller, J. D. , Chen, Y. & Gratton, E. Resolving heterogeneity on the single molecular levelwith the photon-counting histogram. Biophys. J. 78, 474–486 (2000).
386.	Petersen, N. O. Analysis of membrane protein cluster densities and sizes in situ by image correlation spectroscopy. Faraday Discuss. 111, 289–305 (1998).
387.	Wiseman, P. W. Spatial mapping of integrin interactions and dynamics during cell migration by image correlation microscopy. J. Cell Sci. 117, 5521–5534 (2004).
388.	Reits, E. A. J. & Neefjes, J. J. From fixed to FRAP: measuring protein mobility and activity in living cells. Nature Cell Biol. 3, (2001).
389.	Weiss, M. Challenges and artifacts in quantitative photobleaching experiments. Traffic 5, 662–671 (2004).
390.	Suhling, K. , French, P. M. W. & Phillips, D. Time-resolved fluorescence microscopy. Photochem. Photobiol. Sci. 4, 13–22 (2005).
391.	Wallrabe, H. & Periasamy, A. Imaging protein molecules using FRET and FLIM microscopy. Curr. Opin. Biotechnol. 16, 19–27 (2005).
392.	Peter, M. & Ameer-Beg, S. M. Imaging molecular interactions by multiphoton FLIM. Biol. Cell 96, 231–236 (2004).
393.	Kusumi, A. Paradigm shift of the plasma membrane concept from the two-dimensional continuum fluid to the partitioned fluid: high-speed single-molecules tracking of membrane molecules. Annu. Rev. Biophys. Biomol. Struct. 34,
394.	Saxton, M. J. & Jacobson, K. Single particle tracking: applications to membrane dynamics. Annu. Rev. Biophys. Biomol. Struct. 26, 373–399 (1997).
395.	Schutz, G. , Kada, G. , Pastushenko, V. & Schindler, H. Properties of lipid microdomains in a muscle cell membrane visualized by single molecule microscopy. EMBO J. 19, 892–901 (2000).
396.	Edidin, M. Near-field scanning optical microscopy, a siren call to biology. Traffic 2, 797–803 (2001).
397.	de Lange, F. Cell biology beyond the diffraction limit: near-field scanning optical microscopy. J. Cell Sci. 114, 4153–4160 (2001).
398.	Dufrêne, Y. F. Using nanotechniques to explore microbial surfaces. Nature Rev. Microbiol. 2, 451–459 (2004).
399.	Hörber, J. K. H. & Miles, M. J. Scanning probe evolution in biology. Science 302, 1002–1005 (2003).
400.	Pralle, A. & Florin, E. L. Cellular membranes studied by photonic force microscopy. Methods Cell Biol. 68, 193–212 (2002).
401.	Pralle, A. , Keller, P. , Florin, E.-L. , Simons, K. & Horber, J. K. H. Sphingolipid-cholesterol rafts diffuse as small entities in the plasma membrane of mammalian cells. J. Cell Biol. 148, 997–1007 (2000).
402.	Leapman, R. D. Novel techniques in electron microscopy. Curr. Opin. Biotechnol. 14, 591–598 (2004).
403.	Robinson, J. M. , Takizawa, T. , Pombo, A. & Cook, P. R. Correlative fluorescence and electron microscopy on ultrathin cryosections: bridging the resolution gap. J. Histochem. Cytochem. 49, 803–808 (2001).
404.	Ren, Y. , Kruhlak, M. J. & Bazett-Jones, D. P. Same serial section correlative light and energy-filtered transmission electron microscopy. J. Histochem. Cytochem. 51, 605–612 (2003).
405.	Oorschot, V. , de Wit, H. , Annaert, W. G. & Klumperman, J. A novel flat-embedding method to prepare ultrathin cryosections from clutured cells in their in situ orientation. J. Histochem. Cytochem. 50, 1067–1080 (2002).
406.	Zimmerberg, J. , Coorssen, J. , Vogel, S. & Blank, P. Sea urchin egg preparations as systems for the study of calcium-triggered exocytosis. J. Physiol. 520, 15–21 (1999).

Acknowledgements
Nature Reviews Drug Discovery would like to thank all the participants who contributed to the questionnaire. We also wish to acknowledge P. J. Bond, M. Caplan, B. de Kruijff, S. Emr, N. Kahya, M. Kozlov, A. Lee, S. Munro and J. Zimmerberg for their invaluable help in preparing the figures

Competing interests statement.
Kai Simons is the founder of Jado Technologies, which is involved in exploiting lipid rafts for drug delivery.