A cell biologist peering at an electron micrograph, or a biochemist fractionating pulverized cells, is faced with the problem of distinguishing between diverse membrane-bound organelles. Proteins floating in the cytoplasm of the cell often face a similar problem: they have to specifically recognize one organelle among all the others. Many of the proteins that mediate traffic between organelles are ‘peripheral’ membrane proteins that lack transmembrane domains and instead are recruited directly to the cytosolic surface of the organelles on which they act. Such proteins include the various coat proteins that generate transport vesicles, the motor proteins that move vesicles and organelles along the cytoskeleton and the ‘tethering factors’ that attach the vesicles to their destination organelles before fusion (Fig. 1)1.

Figure 1: Schematic representation of the steps of vesicle transport.
figure 1

a, Coat proteins are recruited to the cytosolic face of the donor membrane and induce the formation of a vesicle. The coat recruits SNAREs and transmembrane receptors bound to their cargo. b, After uncoating, motor protein can be recruited to enable the vesicle to travel along microtubules or actin filaments. c, Once at its destination, the vesicle becomes tethered to the acceptor membrane, probably by long coiled-coil proteins or multimeric tethering complexes. d, The SNAREs on the vesicle and acceptor membrane form a complex which drives membrane fusion and hence delivery of the contents of the vesicle.

The accurate functioning of membrane traffic requires that these peripheral proteins bind only to one specific organelle. In some cases they recognize a specific ‘integral’ membrane protein anchored in the bilayer of the relevant organelle. For instance, secretion is initiated when cytosolic ribosomes dock onto the receptors and the translocation machinery that are integral to the membrane of the endoplasmic reticulum (ER). Membrane traffic also involves other integral membrane proteins such as the SNAREs (soluble N-ethylmaleimide-sensitive factor attachment protein receptors) that mediate bilayer fusion, and the cargo receptors that collect soluble proteins from the organelle lumen into transport vesicles (Fig. 1).

However, the majority of membrane-traffic components are peripheral membrane proteins that recognize the correct organelle by binding to either specific lipids, such as phosphoinositides, or to activated forms of GTPases. These lipids and GTPases are usually present on only a subset of internal membranes and hence provide each organelle with a unique identity that allows it to be recognized by the many proteins that act on its cytosolic surface.

The reason that short-lived molecules such as phosphoinositides or activated GTPases are so widely used as internal spatial signals probably reflects the ease with which their subcellular distribution can be controlled. In contrast to integral membrane proteins, which have to be inserted into the ER and then trafficked through a series of compartments to their site of action, lipids can be synthesized, or GTPases activated, only at the site at which they are required. This confers the advantage of accuracy, but also plasticity, allowing vesicles to rapidly lose the identity of the organelle from which they have budded. In addition, it is believed that some organelles themselves can change their identity. Examples include the early endosomes or the cisternae of the Golgi, which continuously mature into later compartments as new ones are generated to replace them.

Although this plasticity has enabled internal organelles to function efficiently, it also seems to be an Achilles' heel for eukaryotic cells that is exploited by invading pathogens. Many pathogenic bacteria and protozoa replicate inside our cells in membrane-bound compartments where they can hide from immune surveillance. The pathogens form and maintain these compartments by altering organelle identity and subverting normal membrane-traffic events. This alteration can be achieved by simply perturbing cytoplasmic GTPases and lipid species, and it does not require the removal or addition of integral membrane proteins. Thus the plasticity of organelle identity may have left eukaryotic cells particularly vulnerable to invasion.

For these reasons the regulation of organelle-specific GTPases and lipids has emerged as an issue relevant to both fundamental cell biology and medicine. In this review, we summarize what is known about how GTPases are activated and lipids are synthesized to create organelle identity. We also mention recent work suggesting that transient attachment of fatty acids contributes to the recruitment of peripheral membrane proteins. We finish by describing how these systems are subverted by invading pathogens.

Organelle-specific GTPases

Two main classes of small GTPase, the Rab and Arf families, contribute to defining the identity of organelles2. These GTPases are molecular switches that can alternate between a GTP-bound active form and a GDP-bound inactive form3. The latter is cytosolic, whereas the active form is associated with membranes. These key characteristics enable them to recruit peripheral membrane proteins only when associated with specific membranes. The overall role of the Rab and Arf GTPases in defining organelle identity is similar, but they differ in the way that they achieve their membrane targeting, and so we will discuss them separately.

Rab GTPases

There are more than 60 different Rab proteins in mammals. Some of these are ubiquitously expressed whereas others are cell-type specific and are associated with specialized organelles4. There are fewer Rabs in lower eukaryotes, which suggests that additional specialist Rabs emerged as new cell types arose during evolution. All Rabs so far characterized are associated with specific organelles, where they act as targeting determinants for a wide variety of protein, including molecular motors and tethering factors2,5. For example, the ubiquitous Rab6 localizes to the trans-Golgi where it recruits Bicaudal D, an accessory protein for the microtubule motor dynein6,7, and TMF1, a coiled-coil protein involved in membrane traffic8. Another ubiquitous Rab, Rab5, recruits many proteins to early endosomes including EEA1, a tethering factor for endosome fusion9. By contrast, Rab27a is expressed only in secretory cells such as melanocytes and recruits the motor myosin-Va to the membranes of melanosomes where it mediates anchoring to the actin cytoskeleton10.

Rab GTPases have a central position in determining when and where peripheral membrane proteins are recruited to organelles. But how are Rab GTPases directed to the correct membrane? Rabs are modified by the attachment of prenyl groups, usually two of them, to the carboxy terminus of the protein. This lipid anchor, although essential for membrane association, is clearly not enough to determine the subcellular localization of these GTPases because it is shared by many different Rabs11,12. The linker between the lipid anchor and the rest of the protein, the ‘hypervariable region’, shows the highest level of sequence divergence between Rabs and have been proposed to act as a signal for targeting13. However, recent analysis of several Rabs has shown that this region is neither necessary nor sufficient for correct targeting, and so it does not seem to be a general signal for Rab targeting14. Instead, current evidence suggests that the correct recruitment of Rab GTPases is governed by the proteins that regulate their GDP/GTP cycle (Fig. 2a), because some of these are themselves found only on specific organelles.

Figure 2: Recruitment of the Rab and Arf family GTPases to membranes.
figure 2

a, Rab GTPases. Rab–GDP forms a complex in the cytosol with GDI. GDF displaces GDI from Rab–GDP, and the Rab is anchored at the membrane by its C-terminal prenyl groups. There, a GEF activates the Rab by exchange of GDP for GTP. This induces a conformational change in the switch 1 and 2 regions of the GTPase (indicated here by a change in colour) that enables the Rab–GTP to bind to its effectors. A GAP stimulates the hydrolysis of GTP and the Rab is retrieved by GDI to the cytosol once again (not shown). b, Arf GTPases. In the cytoplasm, the N-terminal amphipathic helix of Arf–GDP is tucked into a hydrophobic pocket. The N-terminal myristoyl group binds reversibly to membranes where a GEF activates the Arf. The exchange of GDP for GTP induces a change not only in switch 1 and 2 but also in the interswitch loop which moves to displace the N-terminal helix out of its pocket. Arf–GTP then binds tightly to membranes through the hydrophobic residues of the N-terminal helix and the myristoyl anchor, and recruits effectors. A GAP reverses the process and returns the Arf to the cytosol.

Rab proteins bound to GDP form a complex in the cytosol with a protein called the GDP-dissociation inhibitor (GDI). This is thought to mask the prenyl groups and so prevent random association with membranes. A set of membrane proteins known as GDI displacement factors (GDFs) are thought to catalyse the dissociation of Rab from GDI at the target membrane, resulting in the anchoring of the prenyl groups in the lipid bilayer11,12. Once on the membrane, the Rab is activated by replacement of the GDP by GTP by a guanine nucleotide exchange factor (GEF). As with other Ras superfamily GTPases, GTP binding induces a conformational change confined to two main segments, called the switch 1 and switch 2 regions, contributing to the creation of an interface for effector proteins to bind specifically to the active GTP-bound conformation. The Rabs themselves hydrolyse GTP very inefficiently, and, in vivo, this process is stimulated by GTPase-activating proteins (GAPs), after which free GDI can then extract the inactive Rab from the membrane.

According to this cycle, GDFs and GEFs are the main determinants for the localization of active Rabs. However, the relative importance of these two classes of protein for specificity is unclear. Several small membrane proteins from the well-conserved Yip1, Yip2/Yop1 or Pra1/Yip3 families bind promiscuously in vitro to prenylated Rabs and have been proposed to act as GDFs15,16. One of these, Pra1, catalyses the displacement of Rab9 from GDI in vitro17,18. However, the Yip1, Yip2 and Pra1 families have many fewer members than there are Rabs, which suggests that these proteins are, at most, only partly responsible for the specific targeting of Rabs. Indeed, all eight members of the human Yip1 family are located at the ER and Golgi19. Moreover, if a GDF dissociates a Rab from GDI on a membrane where the relevant GEF is absent then the GDP-bound Rab would be rapidly removed from the membrane by GDI. Thus the importance of the Yip1 and Pra1 families for specificity in vivo remains to be proven. Interestingly, both Yip1 and Yip3 (the yeast Pra1 homologue) have a role in ER-to-Golgi transport that is independent of their interaction with Rabs, suggesting that the families may have other functions20,21,22.

It seems likely that the Rab GEFs make at least some contribution to the organelle-specific recruitment of Rabs. However, despite the large number of Rabs, only a few Rab GEFs have so far been identified, and thus much remains unclear about their role11,12. Nonetheless, the known GEFs seem to be localized to specific organelles, implying that their distribution could determine the distribution of the relevant Rab. Moreover, all known GEFs are peripheral-membrane proteins, and studies of how they are themselves localized are beginning to suggest general principles for organization of internal membranes23,24. An interesting case is that of the Rab protein Sec4, which is present on the trans-Golgi and post-Golgi vesicles in yeast. The GEF for this GTPase, the peripheral membrane protein Sec2, is an effector of another Rab GTPase called Ypt31 (ref. 25). Similarly, it has been suggested that the GEF for Ypt31 is recruited on membranes by yet another Rab, Ypt1, localized on the cis-Golgi26. This ‘Rab cascade’ could be a way to coordinate the sequential generation of spatial landmarks if, as is widely believed, the Golgi cisternae mature from cis to trans through the stack. Another illuminating case is mammalian Rab5 on early endosomes, which seems to be initially recruited by a GEF in the nascent clathrin-coated vesicles, which once budded will fuse to form the early endosome27. However, Rab5 itself binds directly to a complex containing Rabex-5, a second Rab5 exchange factor28. Recruitment by the GTPase of its own GEF clearly provides an amplification mechanism that could sustain recruitment of Rab5 to membranes once endosomes have formed.

Finally, it is worth noting the contribution of the Rab GAPs, which exert a negative effect on Rab distribution. A number of these have been identified, and their biological importance for membrane traffic has been confirmed by studies in yeast and mammalian cells29,30,31. However, their precise contribution to Rab location, and how their distribution is regulated, are at present poorly understood.


The Arf family of GTPases comprises Sar1, Arf1–6 and a number of Arf-like GTPases that are similar to Arfs but more distantly related. In common with the Rabs, the Arfs are localized to specific organelles in the cell and recruit a wide range of effectors (Fig. 3). Sar1 recruits the COPII vesicle coat to the membranes of the ER32, whereas Arf1 recruits COPI and clathrin/adaptor vesicle coats to the Golgi and also other effectors including putative vesicle tethers33,34. Arf6 is localized to the plasma membrane, where it is responsible for the recruitment of the kinase that makes phosphatidylinositol 4,5-bisphosphate (PtdIns(4,5)P2) (see below)35. The roles of Sar1 and Arf1 are well established, but the localization and function of most of the Arf-like proteins are less well characterized. Exceptions are Arl1 and ARFRP1 (Arl3 in yeast), which act in a pathway involved in the recruitment to the trans-Golgi of a set of coiled-coil proteins that share a C-terminal GRIP domain that is directly bound by Arl1–GTP36,37.

Figure 3: Location of GTPases to specific organelles.
figure 3

Examples of GTPases that are localized to specific membranes within eukaryotic cells. Arf family GTPases are shown in purples, and Rabs in greens. Some, such as Arf1, are found in multiple locations (in this case throughout the Golgi stack), but most are restricted to only one organelle. The examples shown are the best-characterized and ubiquitous cases, but there are many more GTPases, especially Rabs, that are less well characterized or are found in specialized or polarized cells. A detailed account of the distribution of all Rabs characterized so far can be found elsewhere2,102.

Like Rabs, Arf GTPases carry a lipid moiety, which in the case of Arfs is an amino-terminal myristoyl group. Again, this modification is not sufficient for specific targeting as it binds promiscuously to membranes. Moreover, a few members of this family such as Sar1 or Arl3/ARFRP1 lack this modification38. The specificity of the interaction of Arfs with membranes must therefore be conferred by something else, and in this case it primarily seems to be at the level of the GDP/GTP switch catalysed by their GEFs.

Arf GTPases have the peculiarity of possessing an N-terminal amphipathic helix that follows the myristoyl lipid group in the structure. In the GDP-bound conformation, the hydrophobic residues of this amphipathic helix are masked inside a hydrophobic pocket on the core of the GTPase38. The myristoyl group mediates rapidly reversible and non-specific association with membranes. When bound to membranes, the exchange of GDP for GTP is catalysed by a specific GEF (Fig. 2b). As with Rabs, this exchange induces a conformational change in the GTPase. This involves not only the classic switch 1 and switch 2 regions, but also an intermediate region specific to Arfs termed the ‘interswitch’ by Pasqualato et al.38. In the GTP-bound form, this interswitch slides through the protein and pushes the N-terminal amphipathic helix out of its pocket and forces it to interact with an adjacent bilayer (Fig. 2b)39,40. This mechanism allows the direct coupling of GTP-binding to membrane recruitment, and thereby explains the absence of an equivalent to Rab GDI in the Arf cycle. In the GTP-bound form, Arf family GTPases bind to their specific effectors, until a GAP induces the hydrolysis of GTP, which reverses membrane association and effector binding.

This structural evidence that the activation of Arfs is coupled with their membrane association indicates that their distribution is also controlled by that of the GEFs that activate them. A simple example is that of Sar1, which is recruited to ER membranes upon activation by the GEF Sec12, a transmembrane protein embedded in the ER32. But in the case of Arfs, all GEFs identified so far are peripheral membrane proteins that contain a 200-residue conserved ‘Sec7’ catalytic domain41. These GEFs are recruited to either the Golgi or the plasma membrane. In most cases the mechanism of their recruitment is not fully understood but is likely to involve the contribution of several lipid and/or protein interactions that together define a compartment or a subcompartment. For example, Arf1 is activated by a GEF called GBF1 on the cis-Golgi. GBF1 interacts with the peripheral protein p115 (ref. 42) and also with a transmembrane protein called Gmh1 (ref. 43). Gea2, a yeast homologue of GBF1, also binds to the P-type ATPase Drs2, integral to membranes44. However, none of these interactions by themselves seems to be essential to localize Arf1 to the cis-Golgi. The precise mechanism by which GBF1 and its relatives on the trans-Golgi (BIG1 and BIG2) are recruited remains somewhat mysterious. At the plasma membrane, the levels of Arf6–GTP are regulated in response to external signals. Arf GEFs of the cytohesin/ARNO family contain pleckstrin homology domains that mediate binding to phosphoinositides generated at the plasma membrane by signal-transduction pathways45.

Transmembrane proteins may also contribute to the recruitment of several members of the Arf family. The GDP-bound form of Arf1 has been reported to bind to the cytoplasmic domain of a member of the p24 family of cargo receptors that cycles between the ER and Golgi46. More recently, Honda et al. have suggested that a SNARE called membrin could be a receptor for Arf1 on the cis-Golgi47. Finally, yeast Arl3 and its human counterpart ARFRP1 lack a myristoylation site. They are instead N-terminally acetylated, and this modification is necessary for their binding to a small polytopic membrane protein Sys1 that acts as a receptor on the membranes of the trans-Golgi48,49.

In common with the Rab cycle, GAPs play a key role in the Arf cycle and seem to be particularly important for removing the active GTPases from the membrane leaving a compartment in vesicles. A striking example is that of Sec23, the GAP for Sar1, which is a component of the COPII coat recruited by Sar1 itself32. Sec23 induces Sar1 to hydrolyse GTP after vesicle budding, leading to vesicle uncoating. In the case of COPI, the coat itself is not a GAP; instead ArfGAP1 binds to the coat so as to be recruited into the budding vesicle50. In addition, ArfGAP1 has the unusual property of being highly sensitive to the degree of membrane curvature: the rate of ArfGAP1-stimulated hydrolysis of GTP on Arf1 increases with the curvature of the lipid bilayer51 (see also the review by McMahon and Gallop, p.590, in this issue). This leads to a model in which polymerization of the COPI coat increases membrane curvature, which in turn increases the rate of hydrolysis of GTP on Arf1, coupling budding to removal of Arf1 and hence coat disassembly.

Organelle-specific phosphoinositides

The second major class of molecule that contributes to the unique identity of organelles is specific lipid species, especially phosphoinositides. These are forms of phosphatidylinositol (PtdIns) with phosphate attached by specific kinases to the 3, 4 or 5 positions of the inositol ring (Fig. 4a). Two of the phosphoinositides are second messengers and are synthesized only in response to external signals (PtdIns(3,4)P2 and PtdIns(3,4,5)P3). However, the majority are constitutively present in cells and are generally found on only one or a small subset of organelles (Fig. 4b). They are recognized by peripheral-membrane proteins that usually bind specifically to that particular phosphoinositide52,53.

Figure 4: Location of phosphoinositides to specific organelles.
figure 4

a, Phosphoinositides are generated by phosphorylation of phosphatidylinositol (PtdIns) on the 3, 4 or 5 positions of the inositol ring. All seven possible combinations of phosphorylation occur in vivo. PtdIns is shown with a C18:1 unsaturated chain, but others can also occur in vivo especially C20:4 arachidonate. b, A schematic illustration of the major sites of intracellular accumulation of the well-characterized phosphoinositides. This summary is primarily based on studies in yeast and mammalian tissue culture cells and it remains to be seen how much variation occurs in specialized cells. PtdIns(3,5)P2 has not been localized in mammalian cells, but the 5-kinase is found on late endosomes, and in yeast the lipid accumulates on the vacuole56. Further discussion of phosphoinositide distribution can be found in recent reviews52,53.

Perhaps the best characterized is PtdIns(3)P, which is present on early endosomes, and is recognized by a wide range of peripheral membrane proteins that have key roles in endosomal function. These include motor proteins, tethering proteins and the machinery that downregulates receptors by means of membrane invagination and the formation of multivesicular bodies. Many of these proteins recognize PtdIns(3)P through one of two small folds: the FYVE domain and the PX domain54,55. PtdIns(3)P is also used as a substrate for the synthesis of PtdIns(3,5)P2, which is found on later endocytic compartments. Although the importance of PtdIns(3,5)P2 for late endosomal function is clear from genetic studies in yeast, its effectors have been more elusive. Indeed it is not clear if it is localized to late endosomes, lysosomes or both56. However, two proteins have recently been identified that bind specifically to PtdIns(3,5)P2, including a coated vesicle component, and so these issues should soon be clarified57,58.

Phosphoinositides are also important in the exocytic pathway, with PtdIns(4)P present on the Golgi, where it is recognized by vesicle coat proteins and by the pleckstrin homology domains of proteins that deliver lipids to the Golgi59,60,61. PtdIns(4)P is also found on the plasma membrane, where it is a substrate for the synthesis of PtdIns(4,5)P2. The proteins that are specifically localized to the trans-Golgi and bind PtdIns(4)P seem to also recognize the Golgi GTPase Arf1 (ref. 62). This combinatorial recognition of two determinants allows more restricted targeting than would be seen with either alone, and it is emerging as a more general feature of targeting of peripheral membrane proteins.

PtdIns(4)P and PtdIns(4,5)P2 at the plasma membrane can be phosphorylated by 3-kinases to generate signalling lipids, and PtdIns(4,5)P2 can also be cleaved to generate the second messengers diacylglycerol (DAG) and Ins(1,4,5)P3. However, PtdIns(4,5)P2 is also a major landmark for proteins that need to find the plasma membrane. As such, it is recognized by several key proteins involved in the formation of clathrin-coated pits, including the AP2 adaptor, as well as by many proteins that regulate the actin cytoskeleton63,64 (for a full discussion of PtdIns(4,5)P2 see the review by McLaughlin and Murray, p. 605 in this issue). PtdIns(4,5)P2 has also been proposed to cluster to define subdomains within the plasma membrane65, but this has recently been challenged by detailed in vivo imaging66.

The role of phosphoinositides as key landmarks for subcellular organization raises the obvious question of how their restricted distribution is established. Analogous to the GTPases described above, the answer is that this is achieved by a combination of localized synthesis by specific kinases and rapid turnover by phosphatases that prevent the lipid spreading between compartments. Strikingly, all phosphoinositide (PI) kinases are themselves peripheral membrane proteins, and, at least in some cases, their location is determined by organelle-specific GTPases23,24. For instance, the endosomal PtdIns-3-OH kinase (PI(3)K) Vps34 that makes PtdIns(3)P is regulated, in part, by the endosomal GTPase Rab5 (ref. 67), whereas the PI(5)K that makes PtdIns(4,5)P2 is recruited to the plasma membrane by Arf6 (ref. 35). In addition, organelle-specific PI kinases seem to be major targets for regulatory pathways that adjust the constitutive level of lipid to suit the changing needs of the cell. Thus the endosomal PI(3)K Vps34 is part of a large complex that contains a protein kinase and additional components that seem to recruit the kinase to membrane structures formed during autophagy68. Likewise, the PI(5)K that generates PtdIns(3,5)P2 on later endocytic compartments interacts with further proteins and is apparently regulated by osmotic stress56.

In addition to the kinases, phosphatases are important for restricting the distribution of particular lipids. For instance, the phosphatase synaptojanin is recruited to clathrin-coated pits at the plasma membrane and seems to be important for preventing PtdIns(4,5)P2 from entering the endocytic pathway53,69. Another example is the PI 4-phosphatase Sac1, which is localized in the ER in both yeast and mammals and prevents PtdIns(4)P from accumulating on ER membranes70,71. In addition, yeast Sac1 relocates to the Golgi when nutrients are limiting, where it reduces the level of this PtdIns(4)P and hence the rate of secretion70,71.

Finally, it should be noted that the spatial regulation of phosphoinositides is complicated by their metabolic inter-relationship. Thus, some phosphoinositides must exist before others can be generated. Degradation of one can also generate another. This means that the same lipid can be generated by multiple routes; for instance, PtdIns(3)P on endosomes or phagosomes can be generated both by the action of a PI(3)K on PtdIns or by dephosphorylation of the signalling lipids PtdIns(3,4)P2 and PtdIns(3,4,5)P3 coming in from the plasma membrane72. Moreover, it is striking that the basic exocytic routes and endocytic routes are characterized by monophosphoryated lipids (PtdIns(3)P and PtdIns(4)P) being elaborated further by 5-phosphorylation, which might reflect an underlying organizing principle of membrane traffic that emerged early in evolution53.

Further roles for lipids in protein targeting

In the context of lipids and organelle identity it is also worth mentioning the role of DAG and phosphatidylserine (PtdSer). DAG is generated during signal-transduction events at the plasma membrane and transiently recruits a number of signalling proteins, most notably protein kinases C. However, it is also constitutively produced by lipid metabolism on Golgi membranes and seems to be recognized by a small number of Golgi proteins, although how these proteins distinguish Golgi and plasma membrane pools of DAG remains unclear73.

PtdSer is an acidic phospholipid found throughout the membranes of the cell. It is normally present in both leaflets of the bilayer, except in the plasma membrane, and perhaps in endocytic compartments. In these membranes it is asymmetrically distributed as it is all translocated to the leaflet facing the cytoplasm. It seems that the resulting twofold increase in PtdSer levels in the cytoplasmic leaflet is important for a number of peripheral proteins to be specifically recruited to the plasma membrane through electrostatic interactions with stretches of basic residues74. The asymmetric distribution of PtdSer requires specific translocases, and, interestingly, a number of putative translocases are required for membrane traffic events at the plasma membrane or endosomes75,76.

Finally, an exciting area of study to have emerged recently is that of the direct covalent attachment of proteins to lipids as a mechanism to recruit proteins to particular organelles. One unusual but striking example is Atg8, a key regulator of autophagy, which is recruited to autophagosomes by reversible attachment to phosphatidyl-ethanolamine77. However, perhaps the most widespread example is palmitoylation. Many peripheral membrane proteins are modified by the addition of one or more palmitoyl (C16:0) fatty acyl groups that contribute to the attachment of the protein to the cytosolic face of the membrane78. Analysis of the biology of palmitoylation was hampered until recently by the lack of identified protein palmitoyltransferases (PATs). The recent identification of two PATs in yeast, and the realization that they share a ‘DHHC’ motif found in a larger family of membrane proteins of unknown function, suggest that palmitoylation may be more specific than previously suspected79. This means that the distribution of particular PATs could control the distribution of their substrates. Thus PATs in the Golgi seem to attach a number of proteins such as Ras and nitric oxide synthase to Golgi membranes79. The palmitoylated proteins can then travel on post-Golgi vesicles to the plasma membrane. In addition, recent studies have shown that proteins can be depalmitoylated and hence released from the membrane to travel to new locations80,81.

Identity theft

Many pathogenic bacteria invade host cells and in the process form membrane-bound compartments (similar to the cellular organelles) in which they replicate. These bacteria have developed a number of tactics to disguise these compartments so that they seem to be benign parts of the cell. This allows them to avoid the cell's defence mechanisms and instead receive vesicles delivering lipids and nutrients. Bacteria steal the identity of cellular organelles by manipulating the lipids and GTPases that define organelle identity82,83.

Bacterial pathogens enter mammalian cells by two main pathways84. One route takes advantage of professional phagocytic cells such as macrophages that take up microorganisms into phagosomes. In a second route, bacteria force their way inside non-phagocytic cells by using ‘secretion systems’ to inject virulence factors across the plasma membrane of the host cell85. These virulence proteins trigger localized alteration of the cytoskeleton and the engulfment of the bacteria. For example, Salmonella induces extensive actin rearrangements and membrane ruffling at the site of bacteria–host-cell contact. The bacteria inject SigD/SopB, a phosphatidylinositol phosphatase, to degrade the plasma-membrane lipid PtdIns(4,5)P2 that normally recruits proteins that organize cortical actin86.

Irrespective of the route of entry, bacteria would normally follow the endocytic pathway, which consists of the maturation of the membrane-bound compartment to a late endosome state followed by fusion with lysosomes, leading to digestion of the trapped microbe (Fig. 5). But many pathogens have developed strategies to avoid this fate. Some bacteria such as Listeria and Shigella simply lyse their plasma-membrane-derived compartment and escape into the cytoplasm87. However, most species remain in their compartment, or vacuole, and prevent its maturation to avoid fusion with lysosomes (Fig. 5). For example, compartments containing Mycobacterium tuberculosis retain determinants of early endosomal organelles such as Rab5 and exclude the late endosomal determinant Rab7 (ref. 88). In addition, mycobacteria secrete an enzyme SapM that dephosphorylates the PtdIns(3)P on its compartment, thereby inhibiting movement down the endocytic route toward fusion with lysosomes89. By contrast, Salmonella arrests maturation at a later stage. It does not remove PtdIns(3)P from the membrane of its compartment and so allows fusion with early endosomes to increase compartmental size. Instead, the secreted bacterial phosphatase SopB is necessary for the generation and persistence of this PtdIns(3)P, perhaps by degrading PtdIns(3,5)P2 and arresting the progression of the compartment towards fusion with lysosomes90.

Figure 5: Identity theft by invading pathogens.
figure 5

Bacteria are taken up by phagocytic cells actively seeking to destroy invaders or can force non-phagocytic cells to engulf them. In either case they end up in a phagosome — an intracellular membrane-bound compartment that has many of the characteristics of an early endosome. This compartment matures and fuses with lysosomes, leading to the digestion of the bacteria (left). However, many pathogenic bacteria can evade this fate by using their secretion systems to transfer virulence factors into the cytosol of the host cells that perturb the pathways that generate organelle identity (right). These secreted virulence factors arrest the maturation of the compartment (green stars), and once the endocytic pathway has been evaded, further proteins are secreted (purple stars) to set up interactions with the secretory pathway to attract the delivery of host-cell proteins and lipids while the bacteria replicate.

Once they have evaded destruction by lysosomal hydrolases, the invading bacteria then set up interactions with the secretory pathway to acquire nutrients (Fig. 5). For example, vacuoles containing Legionella or Brucella intercept ER-derived vesicles91. Legionella secretes a protein called RalF out of its vacuole, which contains a domain closely related to the Sec7 domain of the GEFs for Arf1 (ref. 92). RalF is sufficient to recruit Arf1 to the membrane of the vacuole and thus perhaps mimic the vesicle-docking machinery on the cis-Golgi. However, RalF is not essential for bacterial replication and redirection of ER-derived material, and so other, as yet unidentified, bacterial proteins must be involved. Chlamydia- and Salmonella-containing vacuoles reach a perinuclear position in the host cell by movement along microtubules, where they recruit different Rabs and intercept vesicles carrying proteins and lipids trafficking from the Golgi apparatus93,94. The positioning and the structure of such vacuoles are the result of a fine-tuning between the recruitment of both the plus-end motor kinesin and the minus-end motor dynein on their membrane mediated in part by Rab7 (refs 9496).

It thus seems that bacteria use a combination of strategies to perturb, both spatially and temporally, the pathways that generate membrane identity. This enables them to recruit many host-cell peripheral membrane proteins and use them for their own malevolent purposes. Interestingly, there is recent evidence that viruses can use similar strategies. For example, vaccinia virus recruits motor proteins to its external membrane to travel to the plasma membrane97, and poliovirus recruits Arf GTPases to its replication sites98. This field has made rapid progress, but the fact that many of the bacterial proteins secreted across vacuolar membranes do not yet have assigned functions indicates that there remains much to be learnt.

Future questions

From the work of many laboratories it is now clear that the organization of the internal membranes of eukaryotic cells is dependent on the generation of spatially restricted GTPases and lipids. Nonetheless, there still remains much to be learnt about organelle identity. The inherent appeal of simplicity makes the notion of a GTPase or lipid precisely restricted to one organelle very attractive. However, living cells are rarely simple, and there are clear cases where determinants are found in more than one place, such as PtdIns(4)P on the trans-Golgi and the plasma membrane, or Arf1 throughout the Golgi stack. In some cases this may allow specialized sets of proteins to be recruited to multiple locations, or it may simply reflect that complete precision requires the combinatorial recognition of multiple determinants. For any given protein, it may therefore be important to search for additional binding partners even after one has been discovered. Indeed, further factors such as membrane shape could also contribute.

Nonetheless, most GTPases and phosphoinositides have a restricted distribution and there must be mechanisms that account for this. It is clear that there are still a number of missing components, in particular the GEFs for many members of the Rab and Arf families. However, it is apparent from the GTPase regulators and phosphoinositide enzymes so far identified that most are themselves peripheral membrane proteins and hence must recognize organelle-specific determinants. Thus, perhaps the major challenge is to understand what ultimately initiates these pathways that lead to the generation of localized determinants for recruiting effectors.

The impossibility of organelle identity being defined by an infinite series of peripheral proteins raises the issue of the importance of integral membrane proteins. The fact they must start their lives in the ER makes them seem a liability as spatial determinants. Indeed it is striking that the ER itself is an exception as an organelle because most of the known peripheral membrane proteins bind directly to integral membrane proteins99, with just one GTPase (Sar1) and no specific lipids found on this organelle. The SNAREs are integral membrane proteins found in many organelles that have been proposed to contribute to the specificity of membrane fusion, although this has been questioned by their promiscuous assembly in vitro100,101. Determining whether the SNAREs contribute to the specificity of membrane traffic, and how they interact with the peripheral proteins, is a key question for the future. Nonetheless, other integral membrane proteins could contribute elsewhere if they were activated when in the right location by changes in luminal pH, Ca2+ or levels of lipids such as cholesterol and sphingolipids, which might reflect progress through the secretory or endocytic routes. Alternatively, integral membrane proteins could be recognized in combination with other determinants, including other integral proteins, if they converge on the same organelle by cycling through a different itinerary. It is even possible that organelles are themselves sensitive to their location in the cell.

There are many areas that still need to be investigated and debated. However, it is already clear that as more is learnt about organelle identity, we can also hope to understand better the action of the microbial virulence factors that subvert identity, and perhaps one day even use this to design therapeutic strategies to combat these pathogens.