Mechanisms of action of Rab proteins, key regulators of intracellular vesicular transport

Introduction

Rab proteins are GTPases that act as key regulators in intracellular vesicular transport. Like other GTPases of the Ras family, they switch between the inactive (GDP-bound) and active (GTP-bound) forms. Also in common with many other GTPases, they associate with membranes via their C-terminal isoprenoid moieties, in the case of Rabs via one, or most commonly two geranylgeranyl groups. Rabs are involved in a cycle of attachment to and detachment from membranes in a mechanism summarized in Figure 1. In a general model, prenylated Rabs are delivered to membranes, and GDP is replaced by GTP (activation) under the influence of GEFs (guanine nucleotide exchange factors), allowing interaction with effectors to occur. After production of vesicles, activated Rabs interact with, for example, motor proteins involved in transport, and with tethering factors that bring two membranes in close vicinity. After fusion with the target membrane has occurred, Rabs are inactivated by GTP hydrolysis catalyzed by GAPs (GTP activating proteins), before extraction of GDP-bound Rabs by GDI (GDP dissociation inhibitor) (Hutagalung and Novick, 2011). To fill in the details of this scheme, a combination of biochemical, chemical biological and biophysical methods has been used over the past one or two decades. Questions addressed concern the mechanism of prenylation, delivery and targeting of Rab proteins to specific membranes, the extraction of Rabs from membranes to allow another cycle of activity to begin, and in recent work the role of post-translational modifications that are involved in regulation of those processes.

Figure 1:

Cycling of Rab proteins.

After ribosomal synthesis, an unprenylated Rab initially binds to the Rab escort protein (REP), which presents the Rab to Rab geranylgeranyltransferase (RabGGTase, consisting of an α-subunit and the catalytic β-subunit). After prenylation, the Rab can be delivered to a target membrane, where it is activated by a guanine nucleotide exchange factor (GEF) that catalyzes the exchange of bound GDP by GTP. In the active GTP-bound form, Rabs interact with effector molecules and regulate different steps in vesicular traffic (e.g. budding, transport of vesicles, tethering and fusion). Finally, the Rab is deactivated by a GTPase activating protein (GAP) that catalyzes GTP hydrolysis and can then be extracted by GDP dissociation inhibitor (GDI). The inactive GDP-bound Rab is thus kept in a soluble complex with GDI in the cytosol and can be delivered to a target membrane for another round of vesicular transport.

Investigation of Rab delivery to and extraction from membranes

In earlier work, the semisyntheses of C-terminally lipid-modified GTPases were described, in particular RabGTPases, that were generated using the method of native chemical ligation (Alexandrov et al., 2002; Durek et al., 2004; Brunsveld et al., 2005; Rak et al., 2005; Wu and Goody, 2010). Such semisynthetic proteins were subsequently used for numerous investigations on Rab structures and mechanisms. An early example was the determination of the structure of a mono-geranylgeranylated Rab (Ypt1) in complex with GDI (Rak et al., 2003) followed by the structure of the native doubly prenylated complex (Pylypenko et al., 2006). This led to an understanding of the mechanism of solublization of prenylated Rabs by GDI (and in a related study by REP; Rak et al., 2004) as a basis for understanding the role of these interactions as well as providing important insights into genetic diseases caused by mutations and reduced activities of these essential Rab regulators. More recently, such semisynthetic proteins were used for quantitative examination of the kinetics and thermodynamics of interactions with GDI and REP molecules. Application of Rabs with a fluorescent NBD-farnesyl instead of the native geranygeranyl group allowed the first determination of the very high affinities of prenylated Rabs for REP and GDI and led to an explanation of the fact that there are two structurally and mechanistically very similar proteins, REP and GDI, rather than one protein or type of protein that combines the required properties of both (Wu et al., 2007). The reason for this is that REP needs to have a high affinity for unprenylated Rabs to allow presentation to RabGGTase for the prenylation reaction, with no significant gain in affinity after prenylation. In contrast, GDI binds unprenylated Rabs weakly but prenylated Rabs very strongly. This is the basis for an essential property of GDI, i.e. the ability to extract Rabs from membranes (Wu et al., 2007). In a model for the extraction mechanism, GDI initially binds weakly to the GTPase domain of Rab, and the increase in affinity when the lipidated tail also docks onto Rab provides the thermodynamic driving force for extraction. If REP showed a similarly large increase in affinity, the interaction free energy in the prenylated form would be so high that it would not be able to deliver Rabs to membranes.

Subsequent work investigated the dependence of the affinity between Rabs and GDI/REP on the nucleotide binding state of the GTPase (Wu et al., 2010). Using C-terminally lipidated and labeled Rabs, it was shown that replacement of GDP by GTP leads to loss of affinity of Rabs to REP/GDI by three orders of magnitude. This explains why the active (GTP) form of Rabs is stabilized in the membrane, because GDI cannot extract it due to the low affinity to the GTP-bound state. This immediately leads to the notion that GEF-mediated GDP/GTP exchange serves as the thermodynamic driving force for Rab membrane delivery and GEFs might play an important role in the targeting of Rab proteins to a membrane to which the cognate GEFs are localized.

The mechanism of Rab targeting

The question of targeting of Rabs to specific membranes was initially probed 25 years ago, leading to a model that hypervariable C-terminal sequences function as the sole determinant (Chavrier et al., 1991). However, this hypothesis was questioned more recently (Ali et al., 2004), and it has been apparent for some time that the initial model is not or at least only partially correct. The work discussed above, together with other work described below on the influence of bacterial effector molecules on Rab localization, led to experiments designed to test the hypothesis that GEFs are involved in and perhaps responsible for guiding Rabs to their initial target membranes. The concept was to ectopically mislocalize RabGEFs to membranes where they are not normally found and to determine whether this then leads to mislocalization of their cognate Rabs to the same membrane (Blumer et al., 2013). The cell organelle chosen for this study was the mitochondrion, since there is relatively little Rab activity at the outer mitochondrial membrane. The system of Rab and cognate GEF initially chosen for this purpose was Rab5/Rabex-5. The principle was to locate FRB (FKBP-rapamycin binding domains) to mitochondria using a mitochondrial localization sequence from the Listeria monocytogenes protein ActA and to induce dimerization with a Rabex-5-FKBP fusion construct using a rapamycin analog (rapalog). The latter construct initially localized to endosomes, as did Rab5. Addition of the rapalog led to translocation of Rabex-5 followed by Rab5a to mitochondria, suggesting that Rabex-5 was able to recruit Rab5a to the same membrane. This was dependent on prenylation of Rab5a, and on intact GEF activity of Rabex-5. Similar results were obtained with two other Rab/GEF pairs, namely Rab8/Rabin8 and Rab1/DrrA (see below for details of the Rab1-DrrA interaction; DrrA is a GEF secreted into the cytoplasm of infected cells by Legionella pneumophila). In work of another group, similar results were obtained using the GEF BLOC-3 with its cognate Rabs Rab32/38 (Gerondopoulos et al., 2012). These results led to the conclusion that GEFs are important and essential factors for the targeting of Rabs to specific membranes or membrane compartments. This mechanism can only apply if RabGEFs are localized to membranes, and this appears to be the case for a large number of RabGEFs (Goody et al., 2014), possibly for all of them except for Mss4, which is probably not a bona fide RabGEF even if it shares some properties with GEFs (Itzen et al., 2006; Wixler et al., 2011).

Controls used in the work describing mislocalization of Rabs as a result of mislocalization of their cognate GEFs were able to exclude many possible artifacts, but an additional approach underlined the conclusions from this work. This involved the use of GTP and GDP derivatives that are covalently attached to Rab proteins. Derivatives of guanine nucleotides were prepared that harbor an acrylamide residue attached via a linker to the exocyclic amino group of the guanine base (Wiegandt et al., 2015); see Figure 2. Positions were identified in Rab sequences that could be substituted by cysteine without affecting the basic properties of the Rab:GDP and Rab:GTP complexes. The covalent adducts were shown to be structurally highly similar to their corresponding non-covalently bound complexes, and covalently bound Rab-GTP analog adducts interacted normally with effector molecules. In addition, covalently bound Rab-GDP complexes were found to interact normally with GDI and to be prenylated efficiently in their complexes with REP. However, Microinjection of covalent Rab5-GDP adducts into cells did not lead to correct membrane localization (i.e. to early endosomes) but mainly to the Golgi apparatus, while microinjection of the Rab5-GppNHp adduct did not lead to membrane localization due to poor prenylation of the active (i.e. GTP-like) state. Preprenylation of the Rab5-GppNHp covalent adduct led to membrane localization, but to the whole endomembrane system with no concentration at early endosomes. As neither the GDP-locked nor the GTP-locked protein localized correctly in these experiments, this allows the interpretation that it is the GEF reaction itself which leads to targeting rather than the mere presence of the Rab protein in the GTP state.

Figure 2:

GTPases can be covalently locked in their active or inactive state using modified nucleotides.

(A) Guanine nucleotides modified at the exocyclic amino group of the guanine base including a short linker and a reactive group (acrylamide, highlighted in the red box) that can covalently react with a cysteine. (B) Structures of the yeast Rab protein Ypt7 after covalent reaction with a GDP-analog (left, pdb ID 4PHF) or a GTP-analog (right, pdb ID 4PHG). A cysteine at position 35 was artificially introduced to allow formation of the covalent bond (the Rab protein is shown in cartoon representation in gray with switch I colored red and switch II colored blue, the covalently bound nucleotide is shown as sticks (carbon – white, oxygen – red, nitrogen – blue, phosphorus – orange, sulfur – yellow), Mg²⁺ as a green sphere; the 2Fo-Fc electron density map around the nucleotide in black is depicted at 1σ, the Fo-Fc electron density map in green before addition of the linker is depicted at 2.5 σ).

Because of the earlier evidence on the role of the hypervariable C-terminus in Rab targeting, a further approach was undertaken to examine this question. Using the previously developed semisynthetic methods, different Rabs were prepared with C-termini harboring polyethylene glycol (PEG) linker in place of much of the hypervariable region (Li et al., 2014). In all constructs, a three amino acid motif important for the interaction with GDI and REP was retained (C-terminal interaction motif, or CIM; Rak et al., 2003, 2004; Wu et al., 2009). The region N-terminal to the CIM was replaced by Gly-Gly-Ser repeats in some constructs. It was found that such constructs could all be prenylated after microinjection into cells, as long as the distance between the CIM and the C-terminal cysteines as well as the distance between the GTPase core structure and the CIM were kept at a proper length. Such Rab1 and Rab5 constructs were found to localize to the correct membranes, showing that the sequence of the C-terminus is not important for targeting of these GTPases, against the prediction that would have been made from early work (Chavrier et al., 1991) but in line with the results obtained on redirecting these proteins to mitochondria (Blumer et al., 2013). However, in the case of Rab35, correct localization to the plasma membrane was not seen using constructs in which the polybasic C-terminus was replaced by uncharged structural elements, suggesting a required interaction with the negatively charged membrane. For Rab7, constructs in which the sequence of the region N-terminal to the CIM was modified did not localize properly. This is probably due to a known interaction of this region with the effector RILP, which interacts partially with this part of the sequence (Wu et al., 2005). This suggests a role of effector molecules in the localization of Rabs, a model that has previously been suggested for Rab9 (Aivazian et al., 2006). The results are interpreted as showing that correct targeting of Rabs to donor membranes is governed by a complex interplay of several or all of a number of factors including GEFs, effectors and interaction of the C-terminus with the membrane.

Postulating a role of effector interactions with Rabs as a source of membrane localization raises certain problems. Thus, the logic of the manner in which Rabs operate is that they are recruited to a specific membrane in the active form, and can then interact with effectors. If the Rab effector interaction is important for stabilization at the membrane, the Rab molecule will not be available for other interactions, but perhaps this is not in fact required. Thus, the effectors themselves could have, and probably do have in the general case, multiple domains/functions and can act as adapters for multiple downstream effects. A stabilization of Rab attachment to the membrane could be achieved if the effector also interacts with the membrane, but the question then arises as to why Rabs are required to recruit them. A possible mechanism of effector stabilization of Rab membrane attachment might arise from the fact that many Rab effectors are divalent, in most cases because of dimerization of the effectors, and can bind two identical Rab molecules. This is in fact the situation with the RILP effector of Rab7 discussed above (Wu et al., 2005). The generation of a heterotetrameric complex (2× Rab, 2× effector) could indeed lead to an increase of affinity for membranes, as the complex now has two doubly prenylated tails instead of one in a heterodimeric Rab:effector complex. Membrane-binding domains (such as FYVE or C2 domains) present in some effector molecules might additionally strengthen the interaction with membranes. At the same time, the effector would be located to the membrane and could fulfill its downstream function via additional domains. Intriguingly, there is now an example of effectors for the Rab8 family that are monomeric, but can bind two Rab molecules simultaneously.

Bivalent, monomeric Rab effectors (bMERBS; bivalent Mical/EHBP Rab binding domains)

In characterizing a Rab effector domain found in a number of proteins of the Mical (molecules interacting with CasL) and EHBP (Eps15-homology domain binding proteins) families, it was found that while some of the domains bound to Rabs (mainly members of the Rab8 family) in a 1:1 stoichiometric ratio, some other combinations of these domains with certain Rabs led to complexes with a 1:2 (effector domain:Rab) ratio (Rai et al., 2016). This was puzzling, as at least one of the domains (from Mical-3) exists as a dimer in the absence of Rabs, suggesting the possibility of 2:2 or possibly 2:1 (effector:Rab) complexes. Solution of the 3-D structures revealed that a single effector domain can indeed interact simultaneously with 2 Rabs in the 1:2 complexes (e.g. between Mical-1 and Rab10; Figure 3). The effector domain structure consists of one long central helix flanked by shorter N-terminal and C-terminal helices that fold onto the central helix, and the 1:2 structure showed that one Rab-binding site is generated by helix 1 and the first part of helix 2, while the second site is composed of the second part of helix 2 and helix 3. The 2 binding sites are similar, but not identical and not even immediately obvious by sequence alignment, and in-depth analysis showed that this structure arose by gene duplication. The binding sites do not have identical affinity to identical Rabs, with the C-terminal site (K_d values in the 10–100 nm range) binding Rabs more strongly than the N-terminal site (K_d values in the micromolar range). In fact, the presence of two sites with different affinities might also concentrate Rabs in certain regions of a membrane by quasi-irreversible binding to the high-affinity site, and recruitment and release of another Rab molecule via the second (low-affinity) site. Once the second Rab is recruited, it could then bind another bivalent effector, finally leading to efficient formation of highly populated Rab microdomains.

Figure 3:

The bMERB domain of Mical-1 binds two Rabs simultaneously.

Two molecules of Rab10 (gray cartoon, switch I – red, switch II – blue) bind Mical-1 (green cartoon) at different sites within its bMERB domain. The two binding sites are very similar and have been suggested to have evolved by duplication of a common ancestor.

Despite the discussion of possible roles of bivalent effectors above, the significance is currently still unclear. In addition to the possible role in binding two Rabs as a mechanism for stabilizing Rabs attached to membranes, there are also other possibilities. Previous studies showed that some bMERB domain containing effector proteins form an auto-inhibited state in which the bMERB domain binds to the actin-binding CH-domain and inhibit its interaction with actin. Thus, bMERB domains might contain one Rab-binding site for recruitment and one for release of the auto-inhibition (Schmidt et al., 2008; Sakane et al., 2010; Sun et al., 2016). It is also possible that the domains might simultaneously bind two different Rabs and act as sorting hubs in endosomal trafficking.

Learning about vesicular transport from bacteria

In addition to the original mechanism that was proposed for Rab targeting, i.e. the role of the hypervariable C-terminus, a suggestion made almost 20 years ago was that a type of molecule referred to as a GDF (GDI displacement factor) might be involved in delivery of Rabs to membranes and also in targeting (Dirac-Svejstrup et al., 1997). The rational behind this was that the GDI-Rab interaction is very stable, and that an agent might be needed that catalyzes or expedites the dissociation reaction to allow Rab attachment to a membrane. A molecule that appears to have these properties is Pra1 (Yip3 in yeast), a membrane protein occurring in the Golgi and in endosomes (Sivars et al., 2003). Evidence was presented that Yip3 is able to act as a GDF towards Rab5, Rab7 and Rab9, which are all endosome-associated Rabs, and was able to mediate their membrane attachment. However, there is no information on the mechanism of this effect, and other examples of eukaryotic proteins with such properties have not been found. It was therefore of great interest when it was reported that DrrA (or SidM), one of the ca. 300 so-called effector molecules injected into the cytoplasm of cells infected by L. pneumophila, has the properties of a GDF as well as those of a GEF towards Rab1 and is important for the localization of host-cell Rab to the Legionella containing vacuoles (LCVs) (Ingmundson et al., 2007; Machner and Isberg, 2007), which are compartments generated in infected cells by the bacteria for their own replication. Subsequently, it was shown that while DrrA does indeed have GEF properties towards Rab1, it does not have discrete GDF properties, i.e. it does not directly accelerate the rate of dissociation of the GDI:Rab1:GDP complex to then allow exchange of GDP by GTP, catalyzed by the GEF activity of DrrA (Schoebel et al., 2009). The explanation for the apparent GDF activity observed in earlier work is that Rab:GDP dissociating spontaneously from its complex with GDI will be acted upon by the GEF activity of DrrA, replacing GDP by GTP and preventing rebinding of Rab to GDI because of the loss of affinity in the GTP state as described above (Wu et al., 2010). This is in fact a general property of GEF proteins, i.e. the ability to mediate dissociation of Rab:GDI and Rab:REP complexes by catalyzing GTP/GDP exchange on spontaneously dissociating Rab:GDP. The evidence presented on DrrA was of importance in the design of the experiments described above on Rab targeting (Blumer et al., 2013).

Whether the rate of dissociation of Rab complexes with GDI or REP needs to be accelerated by additional factors in vivo is still not clear. Spontaneous dissociation rate constants of such complexes appear to vary over at least two orders of magnitude, so that at least in some cases additional catalysis of this effect might not be required. In addition, it has been suggested that the membrane itself might provide additional ‘assistance’ (Goody et al., 2014). At the thermodynamic level, it is clear that there is a driving force in the direction of dissociation because of stabilization of the lipid by attachment to the membrane, but this does not mean that there is active acceleration of Rab:GDI complex dissociation. A model has been presented that suggests a possible mechanism for this acceleration based on the notion that the relatively weak interaction of the lipid moiety itself for GDI (Zhao et al., 2016), transient production of a state in which only the globular part of the Rab molecule is bound to GDI might allow the lipid to interact with a membrane, and subsequent release of the globular region followed by specific GEF activity at the membrane would result in generation of the membrane-bound active Rab state (Goody et al., 2014; Figure 4).

Figure 4:

Rab membrane targeting and removal.

(A) Rab proteins bind to GDI via their C-terminal prenyl groups and their G-domain. For delivery to a target membrane, the prenyl groups might, in a first step, spontaneously dissociate from their GDI binding pocket and instead bind to a target membrane. Once stably bound, they can completely (but reversibly) dissociate from GDI. If their corresponding GEF is present at this site, the Rab can be activated by nucleotide exchange, thus preventing rebinding of GDI due to its low affinity for the GTP-bound Rab protein. Alternatively, post-translational modifications (PTMs) might be used in an analogous manner as PTMs have been shown to decrease the affinity of the modified Rab protein for GDI in many cases (see main text). (B) Once bound to a membrane, Rab proteins can be stabilized by different mechanisms. These include multivalent effector proteins that bind separate Rab molecules simultaneously, thus increasing the number of membrane inserted prenyl groups within the complex. Some Rab effectors, conversely, contain separate membrane-binding domains that can stabilize the Rab at the membrane. For some Rab proteins (for example Rab35) a positively charged C-terminus might also contribute to stable membrane association. (C) Finally Rab proteins can be extracted from a membrane by GDI after catalysis of GTP hydrolysis by a GAP. If a post-translational modification is also involved in stabilization of the membrane attached form, this will have to be reversed, presumably by specific activities as already identified in the case of covalent modifications by Legionella.

The properties of DrrA described so far are all derived from the central GEF domain in the overall domain structure of the molecule. Crystallization of a construct containing this domain and the C-terminal lipid binding domain together with kinetic studies allowed a characterization of the interaction of this domain with phosphatidylinositol-4-phosphate (Schoebel et al., 2010). The affinity between the lipid head group and the DrrA C-terminus (K_d=ca. 20 nm) is the highest yet measured for such an interaction. Apart from its physiological relevance in ensuring recruitment of DrrA even at low concentrations, the high affinity has been exploited for generating probes for detecting pools of phosphatidylinositol-4-phosphate (Hammond et al., 2014). This left the N-terminal domain of DrrA, a domain without obvious sequence homology to other proteins, without a known function. Crystallization and structure determination of part of this domain showed a hitherto unexpected structural homology to nucleotide transferring enzymes, in particular to the C-terminal domain of glutamine synthetase adenylyl transferase, which catalyzes the transfer of AMP from ATP to a tyrosine residue. It was therefore tested whether DrrA displays such an activity towards Rab1, and this proved to be the case (Muller et al., 2010). The adenylylated residue in Rab1b was Tyr77 (Tyr80 in Rab1a), and it was shown that this modification has a significant effect on interaction of Rab1 with its partner molecules, perhaps most significantly with GDI (Muller et al., 2010; Oesterlin et al., 2012). This effect would lead to stabilization of Rabs in their membrane-bound form, since they would be resistant to extraction by GDI. In addition to this, adenylylation occurs with high preference in the GTP-bound form, and GTP hydrolysis by GAP activity is strongly inhibited in the adenylylated form, both with a cellular GAP (TBC1D20) and the Legionella GAP LepB (Muller et al., 2010). Subsequently, a Legionella effector protein (SidD) was identified that is able to remove the AMP residue from adenylylated Rab1 (Neunuebel et al., 2011; Muller et al., 2012). This is consistent with the observation that Rab1 is removed from the surface of LCVs at later stages in the cycle of Legionella reproduction, suggesting that the timing of events involves initial adenylylation of Rab1 and stabilization in the active form, followed by deadenylation by SidD expressed or delivered at a later time point, thus allowing GTP hydrolysis catalyzed by LepB followed by extraction of Rab1:GDP by GDI.

Although the work described here allows a description of processes occurring with Rab1 at the LCV surface, the exact role of these events is not yet clear. A particularly puzzling point is that while adenylylated Rab1 is in the ‘active’ form (because of bound GTP), the interaction with effectors, at least in the case of the cellular effector Mical-3, is inhibited (Muller et al., 2010). Interestingly, there is a secreted Legionella protein that has some of the required properties of a GTPase effector and localizes to the LCV surface (Conover et al., 2003). However, the affinity between Rab1 and LidA is so high that it could not be measured, neither in the GTP nor the GDP form (Schoebel et al., 2011). In particular, the dissociation rate constants of the complexes are extremely slow. Rab8 is also bound extremely tightly, and the structure of the LidA:Rab8:GppNHp complex was determined, revealing a very large interaction surface area (Schoebel et al., 2011), as also seen in the corresponding Rab1 complex (Cheng et al., 2012). Rab6 is bound more weakly, thus allowing the affinity to be measured. The K_d value was 30 pM in the GppNHp form, with a dissociation rate constant of ca. 10⁻⁴ s⁻¹. Even in the GDP form, the K_d value was ca. 5 nm, which is stronger than any other known interaction of a Rab with an effector in the active (GTP) state. The ca. 100-fold lower affinity in the inactive state means that the interaction can be classed as an effector interaction in the classical sense, but as the K_d values for Rab1 and Rab8 can only be estimated as below 10 pM, possibly considerably lower in the GTP state, the question arises as to the biological role of such extraordinarily tight binding. In contrast to the cellular effector Mical-3, LidA can still bind to adenylylated Rabs, in the case of Rab6 about 100-fold more weakly than to the unmodified Rab. Interestingly, DrrA as an adenylylating agent appears to act on Rabs that interact with LidA, suggesting a cooperation of the 2 interactions, but a detailed comparison of all Rabs has not yet been made.

In searching for evidence of adenylylation of Rab1 in infected cells, a further modification of the same region of Rab proteins was discovered, namely phosphocholination of Ser79 of Rab1a (Ser76 of Rab1b), the residue immediately preceeding the tyrosine adenylylated by DrrA (Mukherjee et al., 2011). The nucleotide substrate for this reaction was shown to be CDP-choline, and further studies showed that both the active and the inactive forms of Rab1 are phosphocholinated, with a preference for the inactive (GDP-bound) form, in contrast to the high preference for the active state in the case of adenylylation (Goody et al., 2012). The consequences of phosphocholination were similar to those of adenylylation, except that there was only a minor effect on the effector interaction with Mical-3. As in the case of adenylylation, there was a major effect on the interaction with GDI, which was essentially abolished and would therefore lead to stabilization of membrane attachment, and in this case preferentially in the inactive form due to the preference of the phosphocholination reaction for Rab:GDP. Similarly to the situation with adenylylation, L. pneumophila also secretes a protein that can reverse the modification, in this case removing the phosphocholine group hydrolytically (Tan et al., 2011; Goody et al., 2012). The similarity with adenylylation also extends to the present lack of understanding of the biological purpose of the modification, apart from the general feature that Rabs are stabilized in their membrane-bound form.

Legionella produces other proteins that interact with host cell Rab proteins. One of these, LepB, is a GAP for Rab1 that localizes to LCVs (Ingmundson et al., 2007). In terms of amino acid sequence, it is not related to eukaryotic GAPs, suggesting there might be structural and mechanistic differences compared to the TBC domain GAPs characterized so far. Determination of the 3-D structure showed this to be the case (Mihai Gazdag et al., 2013), with a basic fold unrelated to classical GAPs (Rak et al., 2000; Pan et al., 2006). The mechanism of GTP hydrolysis is also different in the two classes of proteins. TBC domain GAPs use a dual finger mechanism to catalyze GTP hydrolysis, with the GAP contributing an arginine and a glutamine residue to the active site (Pan et al., 2006; Gavriljuk et al., 2012). This is in contrast to the well-established mechanism for Ras and RhoGAPs, which involve a single arginine finger (Rittinger et al., 1997; Scheffzek et al., 1997). The TBC domains supply additionally a catalytic glutamine residue (trans glutamine), which plays a similar role to the catalytic glutamine (cis glutamine) from Ras or Rho, even though the cis glutamine is present in Rabs. As for Ras and RhoGAPs, LepB does not contribute a trans glutamine, and it is the cis glutamine that plays an essential role in catalysis (Mihai Gazdag et al., 2013). LepB displays very high activity (k_cat/K_m=3.6×10⁶m⁻¹s⁻¹; Rothmeier et al., 2013), underlining a common theme in Legionella effectors that interact with Rab proteins, i.e. that their activity or affinity is generally higher than that of human counterparts, allowing effective competition in infected cells. Intriguingly, other bacteria (a pathogenic E. coli strain and Shigella flexneri) have RabGAPs that have a fold that does not resemble TBC domains but operate via a TBC-like dual finger mechanism (Dong et al., 2012).

While the results obtained from investigations of the manner in which bacterial proteins are able to manipulate host Rabs to their own advantage provided many insights into the mechanism of Rab action, it was initially not clear whether similar modifications, or at least modifications that have similar consequences, applied to physiological and indeed other pathological mechanisms besides those used by bacteria. However, while there is no evidence that adenylylation or phosphocholination of the switch II region of Rabs play any role at all in eukaryotes, there is emerging evidence that phosphorylation of this region does occur, potentially with similar consequences to those already described for the modifications described and discussed above. Thus, LRRK2 (leucine-rich repeat kinase 2), which is often mutated (activated) in Parkinson’s disease, can phosphorylate the switch II region in a number of Rabs including Rab8 and Rab10 at a conserved threonine located 4 residues N-terminal to the serine modified by phosphocholination by AnkX from Legionella (Steger et al., 2016). This appears to inhibit the interaction with GDI, as reported for the adenylylation and phosphocholination reactions (Goody et al., 2012; Oesterlin et al., 2012), resulting in increased stability of Rabs in the membrane attached state, which was already suggested to be a potentially general mechanism (Oesterlin et al., 2012). The same residue in Rab1 is phosphorylated by TGF-β activated kinase 1 (TAK1), which plays an important role in innate immunity (Levin et al., 2016). Thus, this region appears to be a hot spot for post-translational modification of Rab proteins of high physiological and pathological relevance. The stabilizing influence on the Rab membrane interaction is included in the model of Figure 4 together with GEF stabilization by generating the GTP-bound state. If the activity leading to the modification reaction is localized to a specific membrane, this will also be a targeting mechanism.

Conclusion

The studies summarized here, which represent the results of combined efforts on understanding Rab function using a variety of methods including structural biology, chemical biology and kinetics, have led to significant advances in our understanding of Rab delivery to membranes, including targeting, extraction from membranes and the role of post-translational modifications in the switch II region of Rab proteins. These modifications were first discovered in bacterial effectors that manipulate eukaryotic Rabs, and their characterization has led to the concept of stabilization of Rabs in membranes by post-translational modification. In this sense, these modifications can be added to the targeting mechanisms defined in the course of the work summarized in this review, including the decisive role of GEFs. Post-translational modifications, in particular phosphorylation, of Rabs occur in eukaryotes, as shown in recent work, suggesting that such modifications might be a general delivery/targeting mechanism to add to those identified and characterized earlier (Oesterlin et al., 2012).