Professor Gunnar von Heijne, Center for Biomembrane Research, Department of Biochemistry and Biophysics, Stockholm University, SE-106 91 Stockholm, Sweden. (fax: +46 8 15 36 79; e-mail: firstname.lastname@example.org).
Chances are that you have come across membrane proteins many times in your professional life: ion channels, aquaporins, G-protein-coupled receptors, drug resistance proteins. But it is also quite likely that you have never bothered to think about what the implications are of being a membrane protein, as opposed to a soluble protein. What is special about membrane proteins in terms of structure and function, how many membrane proteins are out there, how are they made in the cell? Welcome to the membrane protein universe!
Why do cells need membrane proteins? The answer is but a long list: to transport ions, metabolites and larger molecules such as proteins and RNA across their membranes, to send and receive chemical signals, to propagate electrical impulses, to attach to neighbouring cells or the extracellular matrix, to anchor enzymes and other proteins to specific locations in the cell, to regulate intracellular vesicular transport, to control membrane lipid composition, to organize and maintain the shape of organelles and the cell itself , etc.
It is difficult to imagine a cell without membrane proteins, and the appearance of the first transmembrane peptide must have been a very important step in the evolution of life. Before membrane proteins could make the cell membrane leaky to specific classes of ions and small molecules, the barrier properties of lipid bilayers probably greatly restricted how fast the protocell could obtain nutrients from its surroundings, and hence its growth rate. On the other hand, it does not take much to make a simple transmembrane peptide: just a string of ∼20 predominantly hydrophobic amino acids. To the extent that the protocell membrane was something akin to artificial self-assembled liposomes, such a peptide would be able to insert spontaneously into the lipid bilayer. Further refinement of the amino acid sequence would then rapidly lead to self-interacting peptides that could form larger aggregates with the ability to create more or less substrate-selective pores across the membrane. From there on, the road would lie open to the complex membrane proteomes that we see today.
What membrane proteins look like
The main technique used to determine the three-dimensional structure of membrane proteins is X-ray crystallography. For various technical reasons, it is much more difficult to solve structures of membrane proteins than of water-soluble proteins, and the number of known membrane protein structures represents <1% of all known protein structures. Still, progress in the field is rapid and it seems that we are entering a phase where the number of known membrane protein structures will grow exponentially over the next couple of decades .
Membrane proteins live their lives in a highly anisotropic environment, Fig. 1 (top), which changes from 55 mol L−1 H2O on one side of the membrane to essentially 0 mol L−1 H2O in the centre of the lipid bilayer, and then back to 55 mol L−1 H2O. And all this over only a 50 Å distance . In addition, the part of the membrane occupied by the lipid headgroups (a roughly 15 Å thick layer in each leaflet) is chemically complex with ample possibilities for electrostatic, hydrogen bond and van der Waals-type interactions.
Given this complexity, one might be tempted to assume that the three-dimensional structures of membrane proteins would be equally complex. This is not the case, however. Instead, on a basic architectural level, membrane proteins are in fact simpler and less variable than water-soluble proteins.
So far, only two distinct architectural principles have been discovered in the membrane protein universe: the α-helix bundle and the β-barrel  (Fig. 1, bottom). On a general level, these two basic architectures reflect the need for the protein backbone to satisfy its full hydrogen-bonding potential when buried in a nonaqueous environment. With no water molecules around, these hydrogen bonds can only form within the protein backbone itself, and the two obvious ways to do this is for the membrane-buried parts of the polypeptide chain to fold up either as α-helices or to form a large β-barrel.
The β-barrel proteins are found only in the outer membrane of Gram-negative bacteria and in organelles such as mitochondria and chloroplast, where they serve mostly as passive (but substrate-specific) diffusion pores for small molecules. Being exposed on the surface of bacterial cells, many β-barrel proteins are used as the initial sites of attachment by bacteriophages and bacterial toxins.
In contrast, helix-bundle proteins are found in all cellular membranes, and are vastly more numerous and functionally much more diverse than the β-barrel proteins. It is amongst the helix-bundle proteins that we find nearly all medically important membrane proteins: receptors, transporters, channels, enzymes, etc. For these reasons, I will focus on the helix-bundle proteins.
Helix-bundle membrane proteins come in all sizes and shapes: from proteins with only a single transmembrane α-helix connecting often quite large water-soluble domains on opposite sides of the membrane, to proteins with 20 or more transmembrane helices forming a tightly packed bundle in the membrane. Even larger assemblies of transmembrane helices can be formed from interacting subunits – the record at present is held by cytochrome c oxidase from bovine mitochondria that has 28 transmembrane α-helices distributed amongst 13 subunits  (Fig. 2).
A closer look at the know three-dimensional membrane protein structures reveals some common features . First, as expected, the transmembrane helices are composed mainly of apolar (hydrophobic) amino acids such as leucine, isoleucine, alanine, valine and phenylalanine. These residues are particularly frequent on the lipid-exposed surface of the protein, mediating intimate contacts with the membrane lipids. Sprinkled amongst the apolar residues are polar and even charged residues, often selected for important functional reasons and tending to face the inside of the helix bundle rather than the surrounding lipid. Finally, two normally rare aromatic residues – tryptophan and tyrosine – are strongly enriched near the ends of the transmembrane helices, in the parts of the protein that are embedded in the lipid headgroup regions.
A very interesting feature appears when one shifts the attention from the transmembrane helices themselves to the parts of the protein that are exposed outside the membrane: there is a very strong tendency for positively charged amino acids (lysine and arginine) to be enriched in the parts that face the cytoplasm (i.e. the inside of the cell) with a corresponding reduction in the content of Arg and Lys in parts that face the extra-cytoplasmic compartment (i.e. the outside of the cell) [6, 7]. This skewed distribution is often referred to as the ‘positive-inside rule’, and it has been shown to be valid for the quasi-totality of the helix-bundle membrane proteins, regardless of organism (Archea, Eubacteria, Eukaryota) and membrane system (endoplasmic reticulum, Golgi apparatus, plasma membrane, mitochondrial inner membrane, chloroplast inner and thylakoid membranes) [8–11].
In the context of membrane proteins, ‘topology’ refers to the way in which a polypeptide chain ‘weaves’ back and forth across the membrane, and a topology model is simply a diagram showing the approximate locations of the transmembrane helices in the protein chain and their in/out orientation across the membrane (Fig. 3). The positive-inside rule simply says that the side of the topology model with the higher number of Lys and Arg residues will face the cytosol (or the equivalent compartment in mitochondria and chloroplasts). By protein engineering, it has been shown that the positive-inside rule indeed is a powerful determinant of membrane protein topology, and there are now many examples where the orientation of a protein (or a part of a protein) in the membrane has been manipulated successfully by the addition or removal of positively charged residues . Cases have also been found where evolution has played the same trick, creating homologous membrane proteins with the same number of transmembrane helices but adopting opposite orientations in the membrane .
How membrane proteins are made
At first sight, the biosynthesis of membrane proteins might be thought a simple matter of hydrophobic polypeptide chains coming off the ribosome and then spontaneously integrating into the nearest membrane. Not so. Membrane proteins are not designed to survive long in an aqueous environment, and would readily precipitate into a hydrophobic goo should they ever find themselves alone outside the ribosome.
To avoid this sad fate, all cells have evolved sophisticated machineries that recognize and handle the membrane proteins as they emerge from the ribosome, ultimately ensuring their proper integration into the intended target membrane. The best understood of these machineries is the so-called SRP-Sec61 system that serves to integrate proteins into the membrane of the endoplasmic reticulum in eukaryotic cells (and, equivalently, into the plasma membrane of prokaryotes). In broad outline, the SRP-Sec61 system works as follows [14, 15] (Fig. 4).
1As the nascent polypeptide chain emerges from the ribosome, the first hydrophobic segment (or, in some proteins, an N-terminal ‘signal peptide’ that is later proteolytically removed from the protein) triggers the tight binding of the signal recognition particle (SRP) to the ribosome. SRP is a multifunctional complex composed of an RNA backbone and associated proteins. One of these proteins (SRP54) has a deep pocket on its surface into which the hydrophobic segment binds. This in turn triggers a conformational change in the SRP, causing it both to prevent the ribosome from binding incoming tRNAs (and thereby temporarily halting translation) and priming it for interacting with the SRP receptor, a protein anchored in the ER membrane.
2The ribosome–SRP complex finds an empty SRP receptor on the ER membrane. In a finely orchestrated ‘courtship dance’, the SRP and its receptor then together bring about the hydrolysis of two bound GTP molecules (one on SRP, one on the SRP receptor) leading to further conformational changes and the eventual discharge of SRP from the ribosome. This unmasks a hitherto occluded binding site on the ribosome, allowing it to sit down on top of the Sec61 translocon, a protein-conducting channel in the ER membrane.
3The Sec61 translocon, now positioned directly under the nascent-chain tunnel in the ribosome, opens up such that the growing polypeptide chain can start threading its way across the membrane. But the Sec61 channel has one further trick up its sleeve: if a sufficiently apolar segment appears in the translocating nascent chain, the Sce61 channel opens a ‘lateral gate’ in the channel wall, oriented towards the lipid bilayer, through which the apolar polypeptide segment moves out sideways to finally form a properly integrated transmembrane α-helix.
This is how a protein with a single transmembrane helix is thought to be made; multi-spanning membrane proteins also use the Sec61 translocon in a not completely understood way to attain the correct topology .
Other machineries responsible for the integration of membrane proteins into the mitochondrial and chloroplast membranes exist, but much less is known about how they work [17, 18]. The basic idea is the same, however: to protect the nascent membrane protein from contact with water on its way from the ribosome to the target membrane.
How many membrane proteins are there?
Direct identification of membrane proteins using standard proteomics techniques is difficult, and such studies have only provided superficial inventories of membrane proteomes so far [19, 20]. Fortunately, however, helix-bundle membrane proteins are rather easy to identify by computational means, and good prediction programmes exist . Computational studies now agree on the answer to the question of how many different membrane proteins there are in typical genomes: between 20% and 30% of all predicted genes encode helix-bundle membrane proteins . This number is not much affected if β-barrel proteins are included, as they have been estimated to constitute at most a few percentage of the total proteome [23, 24]. Contrary to what one might have imagined based on the fact that so few three-dimensional membrane protein structures are known, membrane proteins clearly represent a major fraction of the protein universe.
When membrane proteins are sorted according to predicted functional class, further patterns emerge (Fig. 5). In unicellular micro-organisms such as Escherichia coli and the yeast Saccharomyces cerevisiae, proteins involved in the transport of small molecules easily account for 40–50% of all membrane proteins [25, 26]. In multicellular eukaryotes, proteins involved in signal transduction across membranes play a more important role: in the human genome, the so-called G-protein-coupled receptors (GPCRs) alone account for some 5% of all protein-encoding genes (i.e. around 15% of all human membrane proteins are GPCRs) .
Membrane proteins and disease
Like all other parts of a cell, malfunctioning membrane proteins can cause a wide range of diseases. This is not the place to review such diseases in detail, but let me mention just a few general groups. Neurological and cardiac diseases caused by defective ion channels perhaps come first to mind [28–31]. Colour blindness is caused by nonfunctional photoreceptors (rhodopsins) in the eye , membrane proteins that belong to the large class of GPCRs. Water-channel proteins (aquaporins) are critical not only in the kidney but also in, e.g., stroke . Cystic fibrosis is caused by misfolding mutations in the CFTR protein, a chloride transporter in the lung . Finally, mutations in membrane proteins involved in protein import into organelles such as peroxisomes and mitochondria underlie many serious (but fortunately rare) heritable disorders .
One reason for the recent surge in membrane protein research is undoubtedly its relevance for the pharma industry. According to recent estimates , 30% of the human targets for small-molecule drugs currently on the market are GPCRs, 7% are ion channels, 4% are transporters, and other receptors and cell-surface proteins account for an additional 5%. Finally, many kinases and phosphatases – two other very important classes of drug targets – are membrane proteins. In sum, more than half of all current drug targets are membrane proteins.
When it comes to antibiotics, membrane proteins figure prominently amongst proteins involved in the development of resistant bacteria . Not only do β-barrel proteins in the outer membrane of Gram-negative bacteria control the entry of drugs into the cell to a significant extent, but, even more importantly, drug efflux pumps in the inner membrane can clear the cell of antibiotics. Efficient means to inhibit such pumps would thus be a very important part of the antibiotics arsenal.
Portraits of medically important membrane proteins
In these early days of membrane protein structural biology, much of the interest has been focused on proteins of medical relevance. This does not mean that structural biologists have chosen human proteins as targets; rather, the common approach has been to work on bacterial homologues of interesting human proteins. In order to increase the chances that well-diffracting crystals suitable for structure determination by X-ray crystallography are obtained, one typically starts out with a large collection (20–40) of homologous proteins from a range of different bacteria, expresses these in E. coli, and purifies those that express the best and behave nicely from a biochemical point of view. Finally, very large crystallization screens involving tens of thousands of conditions (choice of detergent, pH, precipitant, additives, temperature, etc.) often need to be carried out in order to find good crystals – if such can be found at all!
Despite this very cumbersome process, the three-dimensional structures of many extremely interesting membrane proteins are now known to atomic resolution. To give some general feel for what can be learned from these structures, here are a few examples.
G-protein-coupled receptors are high on the target lists of all major pharma companies and have hence received a lot of attention. GPCRs generally act by capturing external signals (a photon, a neurotransmitter, an extracellular hormone, an odourant, a pheromone) whereupon they undergo a conformational change that triggers the activation of a cytoplasmic G-protein. This ultimately leads to an appropriate biochemical response in the target cell.
Unfortunately, GPCRs have proved especially difficult to purify and crystallize, and to date only one high-resolution structure is available from this very large protein family [38, 39]. This structure is of bovine rhodopsin, a retinal protein with a bound chromophore that can absorb light (Fig. 6). Whilst certainly interesting in itself, this particular structure is not what the pharma industry would like to see, as rhodopsin has only limited sequence similarity with the medically more relevant ligand-activated GPCRs.
The aquaporin water channels are widespread throughout nature. In humans they are found in, e.g., the kidney where they are needed to resorb water from the primary urine . Aquaporins have six transmembrane helices that form an hourglass-shaped internal space into which dip two ‘re-entrant loops’, one from each side of the membrane (Fig. 7). There is a narrow channel running down the protein, just wide enough for a water molecule to pass. The electrostatic field in the channel is such that protons (or, to be precise, H3O+ molecules) cannot penetrate, and the water molecules line up in such a way in the channel that protons cannot ‘hop’ from one to the other across the membrane . Aquaporins therefore conduct water at a high rate but are not leaky to protons and other ions.
Needless to say, ion channels of different kinds are absolutely essential for proper signalling in the nervous system and between nerve cells and muscle cells. The structures of two common kinds of ion channels – K+ channels and acetylcholine receptor channels – are known to date.
All K+ channels (and probably also most Na+ channels) have a common ‘inverted tee-pee’ architecture (Fig. 8). The upper part of the channel hosts the ‘selectivity filter’, a piece of structure designed to impart ion selectivity onto the channel. The lower part controls the gating of the channel, and can be connected to various kinds of ‘sensor domains’ that make the channel gate in response to different types of stimuli (voltage, Ca2+ levels, small ligands).
The selectivity filter in K+ channels works according to a beautifully simple principle: molecular mimicry . In essence, oxygen atoms in the walls of the selectivity filter are held at precisely the right distances relative to one another to mimic the positions of the water oxygens in the solvation shell that surrounds the K+ ion when in solution. The K+ ion can therefore easily slip out of its solvation shell and hop into the selectivity filter, whereas the smaller Na+ ion finds the selectivity filter too roomy and hence prefers to stay in its somewhat tighter solvation shell. This misfit between Na+ and the selectivity filter results in a 1000-fold difference in permeability between K+ and Na+ ions.
The gate also works according to a very simple mechanical principle. Movable sensor domains are connected to the base of the channel, where they can pull or push on the transmembrane helices to make the gate open or close. As an example, in the MthK Ca2+-gated channel from Methanobacterium thermoautotrophicum, a Ca2+-binding sensor domain undergoes a large conformational change when Ca2+ binds, which is transmitted to the gate helices  (Fig. 9). In voltage-gated channels such as the Shaker channel from rat brain, a membrane-embedded, highly charged voltage-sensor domain is instead connected to the gate helices . Thanks to its net positive charge the sensor domain can move relative to the channel in response to changes in the membrane potential, thereby opening and closing the gate.
The acetylcholine receptor has a very different structure (Fig. 10), with a large funnel extending into the synaptic cleft, a double-layered transmembrane channel domain and a ‘fenestrated’ cytoplasmic vestibule [45, 46]. The acetylcholine-binding site is located in the wall of the extracellular funnel, and ligand binding triggers a conformational change that is transmitted into the membrane domain where it causes rotations of some of the pore-lining transmembrane helices, thereby opening the ion channel. Both the funnel wall and the cytoplasmic vestibule are negatively charged, presumably explaining at least in part the channel's preference for cations over anions.
ATP-binding cassette (ABC) transporters are perhaps best known for their role in the development of drug resistance, both in eukaryotes (e.g. during cancer therapy) and prokaryotes (antibiotic resistance). They use the energy of ATP hydrolysis to export (or sometimes import) small-molecule substrates across the cell membrane. A recent example of an ABC transporter from the bacterium Lactobacillus lactis is shown in Fig. 11 . The structure is of the protein in its ATP-bound state, where the two ATP-biding domains in the lower part of the structure are tightly packed against each other, forcing the transmembrane domain to open its large substrate-biding cavity towards the cell exterior. Upon hydrolysis of ATP, the conformation is thought to switch to one where the substrate-biding cavity is instead open towards the interior of the cell, ready to suck up drug molecules or other substrates from the cytoplasm or the inner leaflet of the lipid bilayer. The basic mechanism is thus one of ’alternating access and release’, where ATP binding and hydrolysis drives the protein between the inward- and outward-facing states, thereby making it possible to push substrate molecules across the membrane.
Another class of small-molecule transporters use the energy stored in a transmembrane proton gradient to push substrates against their concentration gradient. The best-studied representative of such proton-driven transporters is lactose permease from E. coli (Fig. 12), which is used by the bacterium to concentrate lactose from the environment . Also here, there is an alternating access-and-release mechanism at work, but the transition between the outward- and inward-facing conformations is driven by the concerted biding of lactose and one proton to nearby binding sites in the protein .
The ATP synthase
Finally, the arguably most impressive membrane protein ‘machine’ understood at the atomic level is the ATP synthase. In eukaryotic cells, ATP synthase is located in the inner mitochondrial membrane where it uses the energy stored in the transmembrane proton gradient to drive synthesis of ATP from ADP and phosphate.
The enzyme is composed of three parts: a rotor in the membrane, a catalytic domain facing the mitochondrial matrix, and a stator that provides a fixed connection between the membrane and the catalytic domain [50, 51] (Fig. 13). The rotor in the membrane can pick up protons from one side of the membrane and allow them to pop out on the other side, but only if it rotates in the plane of the membrane in the process. This rotation is transmitted to the ‘central stalk’– a long helical structure that is connected to the rotor at one end and penetrates through a tunnel in the centre of the catalytic domain all the way to the top. The central stalk is slightly bent, and will push on different parts of the catalytic domain as it rotates. This pushing and pulling controls the conformation of the active sites (three in all) in the catalytic domain, forcing them through a cycle in which they first bind ADP and phosphate and then bring the two molecules close enough together for ATP to form. The stator fixes the catalytic domain in the membrane, preventing it from rotating with the central stalk.
Conclusions and outlook
As I hope this brief review has made clear, membrane proteins are not only numerous but also of central biological and medical importance. Today, we have a good idea of what is in a membrane proteome, we understand how membrane proteins are manufactured in reasonable detail, and we are getting the first atomic resolution glimpses of what membrane proteins look like and how they work.
Nevertheless, there are still large vistas of ignorance that need to be explored: How do membrane proteins fold in the membrane? Can the structure of membrane proteins be predicted from the amino acid sequence alone? Which membrane proteins form complexes and which do not? How are we to improve the success rate of membrane protein structure determination? Which are the best membrane protein drug targets? How do membrane proteins evolve? So far, we have only scratched the surface of the world of membrane proteins…
Conflict of interest statement
No conflict of interest was declared.
Research in the author's laboratory was funded by grants from the Swedish Research Council, the Marianne and Marcus Wallenberg Foundation, the Swedish Foundation for Strategic Research and the Swedish Cancer Foundation.