Brefeldin A (BFA) is a fungal toxin that blocks secretion by preventing the assembly of coat protein components onto donor membranes. Early experiments suggested that the effects of BFA were due to inhibition of Arf nucleotide exchange (12), and subsequent work has revealed that a subset of Arf GEFs is indeed sensitive to the drug (see below). In an unusual mechanism, BFA does not bind to the GEF itself; rather it binds tightly to the Arf–GDP–Sec7 domain complex, sequestering the GEFs in an abortive reaction intermediate (13). Recent crystallographic studies revealed the structural basis for this phenomenon; BFA inserts into a hydrophobic cavity that is only present in the Arf–GDP–Sec7 complex in the early stages of the exchange reaction (prior to nucleotide release) and prevents the conformational changes necessary to bring the catalytic glutamate into contact with the bound nucleotide (11,14). Importantly, only a subset of the Arf GEFs is sensitive to BFA (see below). Although residues that confer BFA sensitivity to specific GEFs have been mapped, Arfs also contribute to the BFA-binding site (13,15,16). Both class I and class II Arfs can accommodate BFA in their interface with BFA-sensitive GEFs, but Arf6 cannot (17).
Recently, a second small molecule inhibitor was identified by structure-based screening that inhibits the catalytic activity of both BFA-sensitive and BFA-insensitive GEFs (18). This molecule, LM11, binds to a pocket near the ARNO/Arf1 interface and like BFA prevents conformational changes required for nucleotide exchange. Functional assays suggest that LM11 inhibits the activation of class I and class II Arfs, but not Arf6 (18).
Mammals express three BFA-sensitive Arf GEFs, GBF1 (19), BIG1 and BIG2 (20). A fourth isoform, BIG3, has been reported in databases but appears to lack the catalytic glutamate residue and may therefore have a noncatalytic function. Interestingly, GBF1 has two orthologs in yeast, Gea1p and Gea2p (21). Conversely, mammalian BIG1 and BIG2 have a single ortholog in yeast, Sec7p, after which the catalytic domain was named. All three isoforms localize largely to the Golgi apparatus where they mediate the activation of both class I and class II Arfs, but appear to function in different Golgi subcompartments. For the sake of simplicity, we will discuss only the mammalian GEFs here, but an excellent review that discusses the yeast and plant GEFs can be found in (22).
GBF1 associates primarily with cis-Golgi elements and endoplasmic reticulum-Golgi intermediates [vesicular-tubular clusters (VTCs)] where it interacts with the tethering protein p115 and is responsible for the Arf-dependent recruitment of the COPI vesicle coat (23,24). As noted above, treatment of cells with BFA induces dissociation of coat protein I (COPI) from Golgi membranes; however, overexpression of GBF1 antagonizes this effect (19). Conversely, expression of a catalytically inactive GBF1 mutant (E794K) induces COPI dissociation and BFA-like effects on Golgi organization (25). Several laboratories have recently used fluorescence recovery after photobleaching to show that GBF1 cycles rapidly on and off Golgi membranes, and that its residence time is actually shorter than that of Arf1 (23,26,27). This is consistent with the idea that the GEF dissociates rapidly from Arf-GTP following nucleotide exchange, while the active Arf remains associated with the membrane until GTP hydrolysis occurs. As might be expected, BFA stabilizes the association of GBF1 with Arf on Golgi membranes but prevents its activation, thereby inhibiting COPI assembly.
The aptly named BIG1 (209 kD) and BIG2 (202 kD) were first identified as BFA-sensitive GEFs by Moss, Vaughan and colleagues using biochemical purification methods (28). Although originally isolated as a complex, it now seems that BIG1 and BIG2 have distinct subcellular localizations and functions. In contrast to the localization of GBF1 in the cis-Golgi/VTCs, both BIG1 and BIG2 associate with later Golgi compartments (29). Like GBF1, whose overexpression prevents BFA-induced dissociation of COPI, overexpression of BIG2 prevents redistribution of the clathrin adaptor complex, adaptor protein 1 (AP-1) from the trans Golgi network (TGN) following BFA treatment (30). Similarly, expression of a catalytically inactive mutant leads to loss of both AP-1 and the monomeric clathrin adaptor Golgi associated, gamma adaptin homologous, Arf interacting protein 1 (GGA1) from TGN membranes and induces their tubulation, without affecting the distribution of COPI (31). Thus it appears that BIG2 regulates the Arf-dependent recruitment of AP-1 and GGA adaptor complexes to the TGN. The role of BIG1 in the TGN has not yet been explicitly examined.
BIG2 has also been found in association with perinuclear recycling endosomes, and, as in the TGN, expression of a catalytically inactive mutant induces tubulation of endosomal membranes (32). Remarkably, knockdown of BIG2, but not BIG1 by RNA interference (RNAi) affects the morphology of recycling endosomes and slows recycling of transferrin, again indicating that BIG1 and BIG2 have distinct functions (33). The presence of BIG2 on recycling endosomes is consistent with the sensitivity of this compartment and the adaptor proteins associated with it to BFA, and with the recent observation that both class I and class II Arfs function in transferrin recycling (34). An Arabidopsis ortholog of GBF1, GNOM, also localizes to endosomes where it mediates recycling of auxin transporters (35).
Little is known about how GBF1, BIG1 and BIG2 are targeted to their specific membrane domains, although the N-terminal 560 amino acids of BIG1 appears to be necessary for its association with the Golgi (36). Some clues may be found in sequences outside the catalytic domain. Recent analysis of 42 sequences from different organisms identified five regions of distinct homology that are shared among all GBF1/BIG family members (Figure 1). Near the N-terminus is a region of approximately 150 amino acids referred to as the DCB domain based on its role as a dimerization and cyclophilin binding motif in Arabidopsis GNOM (37). Immediately C-terminal to the DCB domain is the so-called Homology Upstream of Sec7 (HUS) domain, which contains a highly conserved motif N(Y/F)DC(D/N), which Mouratou et al. have named the HUS box (38). The function of this motif is not known, but the fact that it is shared among GBF1, BIG1 and BIG2 argues that it is not involved in targeting to specific membrane compartments. Three additional domains referred to as HDS (Homology Downstream of Sec7) domains have also been identified, but the roles of these domains in GBF1/BIG function remain unknown.
Recent evidence indicates that GBF1 is recruited to ER exit sites and Golgi membranes through an interaction with Rab1b (39). The Rab1-binding site was mapped to the N-terminal 380 amino acids of GBF1, a region that contains the DCB domain, but not the HUS domain. It will be interesting to determine if the DCB domain is necessary for Rab1 binding, or if this is mediated by other structural features of the N-terminal region.
Mutations in BIG2 have been associated with a human hereditary disorder, autosomal recessive periventricular heterotopia with microcephaly (40). Patients with this disease exhibit a severe malformation of the cerebral cortex, which is thought to be caused by a failure of neurons to migrate from the proliferative zone into the cortex. While one disease-causing allele results in a severely truncated protein product, a second allele leads to a point mutation (E209K) in the DCB domain, highlighting the importance of this domain for BIG2 function (40). Because both secretion and recycling are intimately coordinated with cell motility, it seems likely that the impaired migration of periventricular neurons into the cerebral cortex is because of a defect in one or both of these processes. It will be interesting in the future to determine the specific contribution(s) of BIG2 to cell motility in both neuronal and non-neuronal cells.
Recent evidence also indicates that the BFA-sensitive GEFs can be subverted by picornaviruses to inhibit secretion and promote viral replication. The 3A protein of coxsackievirus was found to bind to GBF1 and to inhibit its GEF activity. Interestingly, 3A protein does not bind to the Sec7 domain, but rather to an N-terminal region upstream of the DCB domain, and the mechanism of inhibition remains unknown (41). A second viral protein, 3CD, binds both BIG1 and BIG2 (42), and its expression stabilizes the association of Arf1 with membranes. Curiously, in another study, infection with the related poliovirus was found to enhance cellular Arf1 activity in a 3CD-dependent manner (43), suggesting that, in contrast to the inhibition of GBF1 by 3A, the interaction of 3CD with BIGs may actually enhance their activity. The mechanisms by which this occurs, and how this leads to enhanced viral replication are still poorly understood, but will be a fruitful area for future research.
Perhaps the best characterized of the Arf GEFs are the ARNO/cytohesin class, which we will refer to here simply as cytohesins. All vertebrates express four cytohesin isoforms, cytohesin-1, cytohesin-2/ARNO, cytohesin-3/Grp1/ARNO3 and cytohesin-4, while only one isoform is present in Drosophila, one in C. elegans and none in yeast. The four vertebrate isoforms are closely related (68% identity among the four human proteins), relatively small (45–50 kD) and share a common domain organization consisting of an N-terminal coiled-coil domain, a central Sec7 domain in tandem with a phosphoinositide-binding PH domain and a short C-terminal extension rich in positively charged amino acids (Figure 1). While both cytohesin-2 and cytohesin-3 appear to be ubiquitously expressed, cytohesin-1 is primarily (but not exclusively) found in leukocytes, and cytohesin-4 is more leukocyte specific (44).
Early work demonstrated that, in contrast to the Golgi-associated GEFs, the cytohesins are distributed primarily to the cell periphery (45), and can be acutely recruited to the plasma membrane in response to phosphatidylinositol 3 (PI3)-kinase signaling (46,47). Intriguingly, cytohesins 1, 2 and 3 are expressed in two splice isoforms that differ only in the insertion of a single glycine residue in the β1/β2 loop of the PH domain. Remarkably, these variants have distinct phosphoinositide-binding specificities; isoforms lacking the insertion (diglycine form) exhibit a strong selectivity for PtdIns(3,4,5)P3, while those containing the insertion (triglycine form) have comparable affinity for both PtdIns(3,4,5)P3 and PtdIns(4,5)P2 (PIP2) (48). Because PIP2 is abundant in the plasma membrane at steady state while PIP3 levels are acutely regulated, one might predict that expression of one variant in preference to another could have significant biological consequences. While this hypothesis remains to be tested directly, it is noteworthy that each of the four cytohesins is expressed in the brain with distinct diglycine to triglycine ratios: Cytohesin-1 and cytohesin-2 are expressed predominantly in the triglycine form, whereas cytohesin-3 and cytohesin-4 are predominantly in the diglycine form (44).
Two recent studies suggest that membrane recruitment of cytohesins may also be regulated by the interaction of small GTPases with the PH domain. Both Arf6 (49) and the Arf-like protein Arl4 (50) were found to bind the PH domain of ARNO in a GTP-dependent manner. Although these proteins bound to a surface of the domain distinct from the phosphoinositide-binding site, membrane recruitment still required interaction of the domain with inositol phospholipids, suggesting that both protein–protein and protein–lipid interactions are necessary for this process.
Ten years after their discovery, the substrate specificity of the cytohesins in vivo remains controversial. In vitro, the cytohesins are promiscuous, although they act most efficiently on class I Arfs. However, in vivo they colocalize preferentially with Arf6 in the cell periphery and ARNO/cytohesin-2 has been shown to preferentially activate endogenous Arf6 in intact cells (51). Because the assays that measure Arf activation in vivo do not resolve spatially distinct populations, and the large pool of active Arf1 in the Golgi could obscure activation of a smaller non-Golgi pool, it remains possible that the cytohesins act on both Arf1 and Arf6 in the cell periphery. As described above, Donaldson et al. have proposed a model whereby activated Arf6, through interaction with the PH domains of cytohesins, helps recruit these proteins to the plasma membrane where they then activate Arf1 in a GTPase cascade (49). However, overexpression of ARNO clearly leads to enhanced activation of endogenous Arf6 in cells (51), indicating that Arf6 may be both a substrate and an effector of the cytohesins.
The biological functions that have been ascribed to the cytohesins are diverse. Cytohesin-1 was initially characterized as an activator of β2-integrin-mediated adhesion (hence the name cytohesin) (52). These early studies have been reinforced by subsequent demonstration that cytohesin-1 has an important role in attachment and migration of leukocytes, through mechanisms that require both catalytic and noncatalytic functions of the protein (53). ARNO/cytohesin-2 has also been shown to promote both the migration of epithelial cells (51) and the outgrowth and branching of neurites (54,55) through a mechanism that requires its catalytic activity. In both contexts, activation of Arf6 by ARNO leads to the downstream activation of the Rho family GTPase Rac1, which is necessary for both cell migration and neurite outgrowth. Ginsberg and colleagues recently reported that ARNO/cytohesin-2 is part of a GTPase-signaling cascade in which integrin-mediated activation of R-Ras leads to recruitment of ARNO through an interaction with the R-Ras effector protein RLIP76/RalBP1 (56). The resulting activation of Arf6 again leads to downstream activation of Rac1 and the resulting extension of lamellipodia. The other cytohesins may also participate in integrin-mediated signaling, but this remains to be explored.
ARNO has also been implicated in several distinct aspects of vesicular transport. These include the docking and fusion of secretory granules in adrenal chromaffin (57) and neuroendocrine cells (58), the endocytosis/desensitization of a subset of G-protein-coupled receptors through an interaction with β-arrestins (59), and the regulation of postendocytic trafficking via an interaction with the vacuolar adenosine triphosphatase (ATPase). Marshansky and colleagues have shown that endosome acidification leads to recruitment of both ARNO/cytohesin-2 and its substrate Arf6 to endosomal membranes (60). More recently, they discovered that ARNO binds directly to the a2 subunit of the vacuolar ATPase, while Arf6 interacts with the c-subunit, and that both interactions require an acidic pH in the endosomal lumen (61). Presumably, recruitment and activation of Arf6 to the acidifying endosome leads to the assembly of carrier vesicles that facilitate membrane recycling and/or endosome maturation, although this has not been shown directly. In this context, it appears that the vacuolar-ATPase acts as a luminal pH sensor, promoting vesicle formation by bringing Arf6 and ARNO together in space on the cytoplasmic face of the endosome.
Recently, several reports have implicated cytohesins as key regulators of insulin receptor signaling. As noted above, invertebrates express only one cytohesin, and mutants in the Drosophila cytohesin steppke (step) are significantly smaller than their wild-type counterparts (62). A similar phenotype can result from defects in insulin signaling, and Hoch and colleagues found that indeed insulin signaling is impaired in steppke mutants. In parallel experiments, an inhibitory RNA aptamer, SecinH3, that inhibits cytohesins but not other Arf GEFs, was shown to dramatically attenuate insulin signaling in both insect and mammalian model systems (63). Surprisingly, both genetic and biochemical analyses indicated that cytohesins act very far upstream in the insulin signaling pathway. Although autophosphorylation of the insulin receptor itself is not affected by cytohesin inhibition, subsequent tyrosine phosphorylation of insulin receptor substrate-1 (IRS-1) the next step in the pathway, was shown to be significantly impaired (63). Knockdown of either cytohesin-2 or cytohesin-3 (but not cytohesin-1), or of Arf6 (but not Arf1) also inhibited insulin signaling in HepG2 cells. Finally, treatment of mice with SecinH3 led to hepatic insulin resistance, demonstrating that cytohesins are important regulators of insulin signaling in vivo (63). Because insulin resistance is a hallmark of type 2 diabetes, it will now be important to establish the prevalence of genetic lesions in cytohesins in patients with the disease. It will also be important to determine how Arf6 activation by cytohesins promotes the assembly of the insulin receptor/IRS-1 complex, and subsequent downstream signaling.
Although phosphoinositide signaling is an important component in the recruitment of cytohesins to the membrane, it is clearly not the only determinant. In the past few years, a number of proteins have been identified that interact with the N-terminal coiled-coil domain of cytohesins, including cytohesin interacting protein (Cytip, also known as CASP and Cybr) (64), the related protein tamalin/Grp1-associated scaffold protein (GRASP) (65), and interacting protein for cytohesin exchange factors (IPCEF) (66). These proteins function as adaptors, linking the cytohesins to other proteins. For example, tamalin has been shown to bind to metabotrobic glutamate receptors and the neurotrophin receptor TrkC via an N-terminal PDZ domain, and associates with multiple cytohesins via its C-terminus (65,67). As noted above, cytohesins appear to act at a point in the insulin signaling pathway that precedes the recruitment of PI3-kinase, suggesting that their association with the insulin receptor is not dependent upon local phosphoinositide synthesis. Proteins such as Cytip, tamalin and IPCEF may therefore provide an independent mechanism for coupling receptor signaling to Arf activation.
The EFA6 family is comprised of four mammalian proteins, EFA6A, EFA6B, EFA6C and EFA6D (68), one representative in nematodes (Y55D9A.1) and one in Drosophila melanogaster (CG6941). Yeast do not express an EFA6-like protein. Members of the EFA6 family are characterized by a variable N-terminal region, a central Sec7 domain paired with a PH domain, and a C-terminal domain containing a coiled-coil motif.
In contrast to the cytohesins, which are relatively promiscuous in their substrate specificity, the EFA6 family is highly selective for Arf6, even in vitro (69). Another distinction is in the PH domain, which unlike the cytohesin PH domain, appears to interact selectively with PtdIns(4,5)P2. In keeping with this phospholipid specificity, EFA6 isoforms localize primarily to the plasma membrane at steady state (68). A third distinguishing feature of this family is that expression of EFA6 isoforms promotes the reorganization of cortical actin into structures resembling microvilli. The molecular basis for this actin remodeling is not yet understood, but is not dependent on nucleotide exchange activity; instead it requires the C-terminal coiled-coil domain that is present in all EFA6 isoforms (68). In polarized epithelial cells, EFA6A has been found associated predominantly with the apical plasma membrane, although its potential involvement in the formation of apical microvilli was not explored (70). However, during establishment of junctional complexes, ectopic expression of EFA6A was shown to enhance the rate of tight junction formation. Furthermore, junction disassembly induced by calcium depletion was inhibited by EFA6 expression, in a manner that depended on both nucleotide exchange activity and the C-terminal coiled-coil domain, suggesting a role for this protein in junction maintenance (70).
EFA6A is expressed predominantly in the brain, with smaller transcripts (presumably splice variants) expressed in intestinal tissues (68). EFA6C transcripts have also been detected primarily in the brain (71). EFA6B is expressed more widely with highest levels in the placenta, pancreas, spleen and thymus. Recent reverse transcriptase polymerase chain reaction analysis indicates that EFA6D is indeed ubiquitous, appearing in every tissue examined (71). Interestingly, while EFA6A, EFA6C and EFA6D transcripts are all relatively enriched in the brain, they exhibit distinct regional distributions (71). Although it has been difficult to localize these proteins in neurons by morphological methods, both EFA6A and EFA6D cofractionate with postsynaptic density markers, suggesting that they are concentrated in synapses.
In cultured hippocampal neurons, ectopic expression of EFA6A has been shown to promote the formation of dendritic spines, in a manner that is dependent on nucleotide exchange activity (72). Conversely, shRNA-mediated knockdown of EFA6A inhibits spine formation. These data suggest that, although EFA6 family members become concentrated in mature synapses, they also have a role in synapse formation and plasticity.
The BRAGs are characterized by the presence of an IQ-like domain, a central catalytic domain, an adjacent PH domain and at least one coiled-coil domain. Mammalian genomes contain three BRAG genes, BRAG1 (IQsec2), BRAG2 (IQsec1/Arf GEP100) and BRAG3 (IQsec3/synArfGEF), each of which is expressed in at least two splice isoforms, while C. elegans and Drosophila each contain only one BRAG gene. BRAG2/Arf GEP100 has been characterized in vitro and found to be selective for Arf6, and as the name implies, is resistant to inhibition by BFA (73). Curiously, although these proteins contain PH domains, nucleotide exchange was shown to be unaffected by the presence of phosphoinositides, suggesting that the PH domain may not interact with inositol phospholipids (73). However, this remains to be determined directly. Similarly, although all BRAGs possess an IQ domain (which often, but not always bind calmodulin), it is not yet known if this domain indeed binds calmodulin. In this regard, in vitro assays suggest that calmodulin has no effect on catalytic activity (73), and mutants in the IQ domain of the Drosophila ortholog do not appear to alter its biological activity (see below).
In mammals, BRAG1 and BRAG3 seem to be expressed primarily in the brain (74,75). BRAG2 is enriched in the brain but is also expressed in a variety of other tissues (73). Like EFA6A and EFA6D, BRAG1 and BRAG3 cofractionate with markers of postsynaptic densities, and colocalize with synaptic markers in cultured hippocampal neurons (74). Interestingly, the longer splice variant of both BRAG1 and BRAG3 contains a C-terminal type I PDZ-binding motif, which in the case of BRAG3 has been shown to interact in vitro with the postsynaptic proteins PSD95, SAP97 and Homer (75). Although these findings clearly suggest a synaptic function for BRAG proteins, a specific role in neurotransmission is yet to be demonstrated.
In non-neuronal cells, BRAG2 has been shown to play a regulatory role in the endocytosis of a subset of cargo proteins. RNAi-mediated depletion of endogenous BRAG2 in Hela cells led to the accumulation of β1 integrins on the cell surface, and a corresponding increase in attachment of cells to fibronectin (76). Similarly, BRAG2 depletion in HepG2 cells led to increased surface levels of E-cadherin, and inhibited cadherin endocytosis induced by hepatocyte growth factor (HGF) (77). In contrast, BRAG2 depletion had no effect on surface levels of transferrin receptor suggesting that BRAG2 modulates endocytosis in a cargo-selective manner (76).
In Drosophila, the single BRAG gene is expressed in three splice isoforms, A, B and C, all of which contain a C-terminal PDZ-binding motif. Olson and colleagues first identified dBRAG (which they called Loner) in a screen for mutations affecting muscle development (78). These authors found that loner mutants had profound defects in larval myoblast fusion, an effect that could be phenocopied by expression of dominant negative Arf6 in the myoblast lineage, indicating that Arf6 is activated by Loner in this context. Loner/dBRAG was recruited to sites of cell–cell contact by the homophilic adhesion molecules Dumbfounded (Duf) and Roughest (Rst), which serve as adhesion receptors in myoblast fusion. Importantly, loner mutants lacking either catalytic activity or the PH domain failed to restore myoblast fusion, indicating that both components are necessary for Loner function. However, as noted above, complementation of the loner defect did not require an intact IQ motif, suggesting that it is not required in this context. Although the precise roles of Loner and Arf6 in myoblast fusion have not as yet been determined, they are at least partly responsible for the proper localization of Rac, which is necessary for fusion (78).
Drosophila BRAG was also isolated in a separate screen for mutants in neuronal pathfinding, where it was given the name schizo (79). In schizo mutants, neurons fail to cross the embryonic midline, apparently because of enhanced repulsive signaling by the transmembrane protein Slit. The schizo phenotype could be suppressed by reducing the expression of Slit, suggesting that Schizo may regulate the surface levels of Slit on midline glial cells (79). This would be in keeping with the proposed role of mammalian BRAG2 in regulating the endocytosis of adhesive receptors such as integrins and E-cadherin.
F-boxes are modules of approximately 40 amino acids that mediate the incorporation of proteins that bear them into multisubunit ubiquitin–ligase complexes. F-box only protein 8, which is found in all vertebrates, contains an N-terminal F-box and a C-terminal Sec7 domain, but no other recognizable motifs (Figure 1). This protein remains completely uncharacterized, however its structure predicts that it can interact with both Arfs and ubiquitin ligases, and may facilitate the ubiqutination and subsequent degradation of Arfs. Alternatively, it may be recruited to sites of active protein ubiquitination, and promote local Arf activation at such sites. This is clearly an area that will benefit from future exploration.