Dissecting the role of the ARF guanine nucleotide exchange factor GBF1 in Golgi biogenesis and protein trafficking

COPI coats assemble by the transfer en bloc of heptameric coatomer complexes from the cytosol to membranes (Waters et al., 1991). The recruitment is catalysed by ADP-ribosylation factors (ARFs), members of the Ras superfamily of small GTPases (reviewed in Randazzo et al., 2000). ARFs cycle between inactive GDP-bound and active GTP-bound states. The intrinsic GDP-GTP exchange activity of ARFs is low, and in vivo this reaction is facilitated by guanine-nucleotide exchange factors (GEFs) that bind to ARF-GDP and catalyse the displacement of GDP. The resulting `empty' ARF binds rapidly to GTP owing to the presence of high cytosolic levels of GTP in cells. The binding of GTP to ARF displaces the GEF (Niu et al., 2005; Szul et al., 2005), and the activated ARF-GTP associates with membranes and initiates COPI recruitment.

The role of COPI in the biogenesis of the secretory pathway and in protein trafficking within mammalian cells has been explored by manipulating COPI recruitment through molecular and pharmacological approaches. Overexpressing a mutant ARF1 (ARF1-T31N) or mutant GBF1 (GBF1-E794K), or treating cells with BFA, causes dissociation of COPI from membranes (Dascher and Balch, 1994; Donaldson et al., 1990; Garcia-Mata et al., 2003; Klausner et al., 1992; Lippincott-Schwartz et al., 1989; Ward et al., 2001). In all cases, the release of COPI from the membrane results in the same phenotype: the complete collapse of the Golgi into the ER and inhibition of protein trafficking. However, other findings do not support the absolute requirement for COPI in the biogenesis of the secretory pathway. CHO cells depleted of ϵ-COP contain morphologically normal Golgi stacks at 34°C (although the Golgi disperses when the temperature is raised to 39.5°C) (Oka et al., 2004). Similarly, in cells treated with 1,3-cyclohexane-bis(methylamine) (Hu et al., 1999) or with BFA in the presence of monensin (Barzilay et al., 2005), COPI dissociates from Golgi membranes, but the Golgi does not collapse into the ER. Furthermore, siRNA-based depletion of ARF1 in conjunction with ARF4 causes COPI dissociation but does not result in the collapse of the secretory pathway (Volpicelli-Daley et al., 2005) – instead, the Golgi exhibits extensive tubulation.

Equally uncertain is the precise role COPI plays in protein trafficking. A function in retrograde trafficking is supported by extensive genetic and biochemical studies showing that COPI components bind to di-lysine motifs found in proteins (such as the KDEL receptor, members of the p24 family and ERGIC53) that cycle between the Golgi and the ER and that functional COPI is required for the efficient retrieval of such proteins from the Golgi (Cosson et al., 1998; Cosson and Letourneur, 1994; Harter and Wieland, 1998; Lanoix et al., 2001; Letourneur et al., 1994). In addition, EM immunolocalization shows that glycosylating enzymes that cycle continuously between cisternae are enriched in COPI vesicles present in the Golgi region (Martinez-Menarguez et al., 2001). However, live imaging of retrograde trafficking of ERGIC53, the KDEL receptor, the Rab6 GTPase and cholera toxin suggests that retrograde trafficking occurs through tubular elements (Girod et al., 1999; Presley et al., 2002; Sciaky et al., 1997; White et al., 1999). In addition, others have shown by EM immunolocalization that glycosylating enzymes are depleted in COPI vesicles (Kweon et al., 2004). More recently, distinct populations of COPI vesicles have been isolated containing retrograde and anterograde cargo, suggesting that COPI might be involved in protein trafficking in both directions of the secretory pathway (Malsam et al., 2005).

Here, we explore the function of COPI in biogenesis of the Golgi and protein trafficking by preventing recruitment of COPI to membranes by removing GBF1, the GEF for ARF (Claude et al., 1999; Garcia-Mata et al., 2003; Kawamoto et al., 2002; Zhao et al., 2002). We report that siRNA-mediated depletion of GBF1 causes COPI dispersal but does not lead to the collapse of the secretory pathway. Instead, it causes tubulation of the cis-Golgi and the coalescence of the cis-Golgi with the ER-Golgi intermediate compartment (ERGIC). Significantly, COPI dissociation caused by GBF1 depletion inhibits the trafficking of only some proteins. Our findings support a model in which a functional secretory pathway can be maintained in the absence of COPI recruitment. However, GBF1-mediated COPI events are essential for some proteins to navigate selectively through the pathway.

A number of human GBF1 isoforms have been described (Claude et al., 2003) (Fig. 1A). Two siRNAs were designed complementary to distinct sequences within human GBF1 and spanning base-pairs 66-86 (#2siRNA) and 1941-1962 (#1siRNA; Fig. 1A). Both siRNAs target all GBF1 isoforms. Each efficiently depletes GBF1 levels in human HeLa cells to <10% after 3 days of incubation (Fig. 1B,C). The depletion is sequence specific as normal levels of GBF1 are detected in cells transfected with reagent alone (`mock') or scrambled (`scr') siRNA of the same composition. The depletion is GBF1 specific as normal levels of p115, β-COP, calnexin, golgin-84 and β-tubulin are observed in GBF1-depleted cells (Fig. 1D). Analogous results were obtained with the #1siRNA (data not shown). To study the effect of GBF1 depletion in multiple cell types, we also generated siRNAs complementary to the #1siRNA sequence within rat GBF1. Similar to the case for HeLa cells, GBF1 is depleted in normal rat kidney (NRK) cells after 3 days (Fig. 1E). The depletion is selective as normal levels of β-COP and β-tubulin are present in the GBF1-depleted NRK cells.

We first explored the effect of GBF1 depletion on COPI distribution. As shown in Fig. 1F, decreasing concentrations of GBF1 are paralleled by changes in the localization of the β-COP component of COPI. In cells containing normal levels of GBF1, β-COP is concentrated on Golgi membranes and on small peripheral ERGIC structures. In cells with lower levels of GBF1, β-COP is progressively more dispersed. In cells lacking detectable GBF1, β-COP occupies a diffuse pattern analogous to that seen in cells treated with BFA (see below and supplementary material Fig. S1B). (In all panels showing GBF1-depleted cells, we selected a field with at least one cell containing GBF1 from among the vast majority (usually ∼95%) of GBF1-depleted cells.) The redistribution of β-COP most likely reflects reduced membrane association as the cellular level of β-COP is not altered by GBF1 depletion (Fig. 1D). The redistribution of β-COP in GBF1-depleted cells suggests that GBF1 is the major determinant of COPI recruitment within the cell. The role of GBF1 in COPI recruitment correlates with the colocalization of GBF1 and COPI on the ERGIC and the cis-Golgi (Garcia-Mata et al., 2003; Kawamoto et al., 2002; Zhao et al., 2002).

The effect of GBF1 depletion on the architecture of ER exit sites, ERGIC and the Golgi was explored by immunofluorescence (Fig. 2). The distribution of ER exit sites was altered in GBF1-depleted cells; ER exit sites are normally concentrated in a perinuclear region of the cell, adjacent to the Golgi complex, but, in GBF1-depleted cells, ER exit sites are more dispersed (Fig. 2A). The distribution of ERGIC elements is also disrupted in GBF1-depleted cells, with ERGIC redistributed into tubular elements that connect peripheral punctate sites (Fig. 2B). The architecture of the cis-Golgi shows extensive tubulation, with the cis-Golgi tethering protein GM130 localizing to an expanded meshwork of tubules (Fig. 2C). By contrast, the distribution of giantin, a medial-Golgi marker in GBF1-depleted cells remains in a perinuclear structure that appears `looser' than a normal Golgi (Fig. 2D). The architecture of the distal Golgi/trans-Golgi network (TGN), as reflected by the localization of golgin-245, a TGN marker (Derby et al., 2004), is also less perturbed, with golgin-245 localizing in a perinuclear pattern in cells with extensive cis-Golgi tubulation (Fig. 2E). The preferential tubulation of the cis-Golgi suggests that this compartment is regulated by GBF1-mediated ARF-COPI coating events, consistent with the previously described concentration of GBF1 and COPI on cis-Golgi elements (Oprins et al., 1993).

GBF1 interacts directly with the p115 tethering factor (Garcia-Mata and Sztul, 2003), and we examined the localization of p115 in GBF1-depleted cells. As shown in Fig. 2F, p115 is dispersed into a tubular pattern that is similar to that of GM130 tubulation, but more punctate. The membrane association of p115 in GBF1-depleted cells is consistent with our previous findings that p115 and GBF1 target to membranes independently (Garcia-Mata and Sztul, 2003).

The phenotypes observed in the GBF1-depleted cells differ significantly from those in cells treated with BFA or expressing the inactive GBF1-E794K mutant of GBF1 (supplementary material Fig. S1C). In those cells, preventing COPI recruitment to membranes causes the dispersion of ER exit sites, the relocation of ERGIC53 to peripheral punctate sites and the complete collapse of the Golgi complex. An analogous collapse of the Golgi has been described in reports from other laboratories (Lin et al., 1999; Lippincott-Schwartz et al., 1989; Mardones et al., 2006; Ward et al., 2001).

The extent of Golgi tubulation in GBF1-depleted cells correlates directly with GBF1 levels, with `loose' GM130-labelled tubules in cells with relatively normal levels of GBF1, more tubulated GM130 structures in cells containing lower levels of GBF1, and a meshwork of GM130 tubules connecting peripheral elements in cells with undetectable GBF1 (Fig. 2G). The tubules appear to originate from the Golgi and extend into the cell periphery by aligning along microtubules (supplementary material Fig. S2A). Comparing the microtubules in GBF1-depleted cells with those in control cells (supplementary material Fig. S2B) shows that the microtubule cytoskeleton is not visibly disturbed by GBF1 depletion. The actin cytoskeleton is also unchanged in GBF1-depleted cells (data not shown), suggesting that the disruption in Golgi architecture is not due to alterations in the cytoskeleton.

To ensure that the observed phenotypes correlate with selective removal of GBF1 and not off-target effects, we characterized the morphology of cells silenced with the other siRNA (#2siRNA). As shown in Fig. 2H, analogous tubulation of GM130 is present.

A phenotype of extensive Golgi tubulation has been reported in cells in which COPI recruitment is inhibited by removing ARF1 in conjunction with ARF4 by siRNA treatment (Volpicelli-Daley et al., 2005). Removing other combinations of ARF proteins (ARF1 with ARF3, or ARF1 with ARF5) does not dissociate COPI, suggesting that ARF1 and ARF4 are the major regulators of COPI recruitment in the cell. The similarity in phenotypes caused by ARF1-ARF4 depletion and GBF1 depletion suggests that ARF1 and ARF4 represent GBF1 substrates in vivo. In support of this, previous studies have suggested that GBF1 interacts with ARF1 in vivo (Niu et al., 2005; Szul et al., 2005), and in vitro studies showed that GBF1 can catalyse GDP-GTP exchange on a mixture of ARF1 and ARF3 (Claude et al., 1999). To characterize more directly the ARF proteins that interact with GBF1 in vivo, we performed co-immunoprecipitation studies in the presence of BFA. BFA binds to a complex of an ARF protein bound to its cognate GEF, and together they form a stable tertiary complex (Peyroche et al., 1999; Renault et al., 2003). Hence, BFA stabilizes a complex of GBF1 and its cognate ARF proteins. As shown in Fig. 3, GBF1 coprecipitates with ARF1 and ARF4. We stress that cellular levels of ARF3 are much higher than those of ARF4, further strengthening the specificity of the interaction. Our data suggest that ARF1 and ARF4 are the major GBF1 substrates in vivo and that GBF1-mediated activation of ARF1 and ARF4 facilitates COPI recruitment at the ER-Golgi interface.

The cellular responses to GBF1 depletion appear to be conserved among mammalian cells, and Golgi tubulation analogous to that seen in HeLa cells is observed in GBF1-depleted rat NRK cells (Fig. 4A). A reconstruction of a series of confocal images of a single cell shows the tubular meshwork to be extensively connected and largely planar (supplementary material Movie S1).

Previous studies have shown that disruption of COPI recruitment by BFA causes the extension of tubular elements that originate from the Golgi and connect to ER exit sites (Mardones et al., 2006). The destination of the tubules in GBF1-depleted cells was examined by colocalizing GM130 with markers for ER exit sites and the ERGIC. In GBF1-depleted cells, GM130 tubules do not target to ER exit sites (Fig. 4B). A panel of focal planes through a cell shows separation of the tubules and ER exit sites at all levels within the cell. Instead, GM130 tubules appear to connect ERGIC elements distributed throughout the cell (Fig. 4C). GM130 and ERGIC53 colocalize in the punctate structures and in some tubules, but ERGIC53 is found predominantly in the puncta, whereas GM130 is found preferentially in the tubules.

The tubules exhibit dynamic formation and collapse when characterized in real time by simultaneously imaging GRASP65-GFP and ERGIC58-YFP (Fig. 4D; supplementary material Movie S2). GRASP65 is the membrane receptor for GM130 (Barr et al., 1998) and shows extensive tubulation in GBF1-depleted cells. ERGIC58 is the rat homologue of human ERGIC53. For ease of visualization, the yellow in this figure was digitally changed to red after image acquisition. Time-lapse imaging shows the peripheral puncta containing ERGIC58 to be relatively immobile and long lived (Fig. 4E). Evaluation of 15 such puncta shows four disappearing within 5 minutes of imaging, three persisting for up to 10 minutes, and eight remaining after 10 minutes of imaging. The restricted mobility is in agreement with previous studies showing the ERGIC to be long lived and stationary in control cells (Ben-Tekaya et al., 2005). By contrast, GRASP65 appears to flux through the puncta and rapidly move into tubules that emanate from individual ERGICs (Fig. 4F). Often, a single ERGIC structure extends tubules in opposite directions (arrowheads). The formation of the tubules correlates with a change in the color of the ERGIC from yellow (containing GRASP65 and ERGIC58) to red (containing mostly ERGIC58). The tubules often form transient molecular bridges between ERGIC outposts (Fig. 4G). The tubules are relatively unstable and rapidly form and disappear. Furthermore, the tubular network undergoes extensive bidirectional movements along microtubule tracks.

COPI recruitment is required for the anterograde trafficking of VSV-G as preventing COPI recruitment by BFA treatment (Donaldson et al., 1990), expression of the dominant-negative E794K mutant of GBF1 (Garcia-Mata et al., 2003) or expression of dominant-negative ARF mutants (Dascher and Balch, 1994) result in arrest of VSV-G in pre-Golgi compartments. We analysed the trafficking of newly synthesized VSV-G in control and GBF1-depleted cells. Control or GBF1-depleted cells were transfected with the temperature-sensitive ts045VSV-G mutant tagged with GFP at the non-permissive temperature of 42°C. This results in the synthesis of misfolded VSV-G that is retained within the ER in control and GBF1-depleted cells (Fig. 5A). In this experiment, GBF1-depleted cells are identified by GM130 tubulation. After shifting the cells to the permissive temperature of 32°C for 1 hour, VSV-G is detected within the Golgi of control cells (Fig. 5B). This agrees with previously published kinetics of VSV-G trafficking (Presley et al., 1997). By contrast, in GBF1-depleted cells, VSV-G remains largely within the ER, with some VSV-G entering the GM130 puncta and tubules (Fig. 5B). After 2 hours at the permissive temperature, VSV-G is detected within the Golgi and is also visible on the plasma membrane in control cells (Fig. 5C). By contrast, in GBF1-depleted cells after 2 hours at the permissive temperature, a proportion of VSV-G is still retained within the ER, and some VSV-G colocalizes within the punctate and tubular GM130 elements (Fig. 5C). Importantly, VSV-G is not detected on the plasma membrane of GBF1-depleted cells even after 12 hours at the permissive temperature (data not shown). This suggests that trafficking of VSV-G is inhibited in GBF1-depleted cells.

The inhibition of VSV-G transport was confirmed by pulse-chase experiments. Newly synthesized VSV-G is sensitive to Endo H and is processed to an Endo-H-resistant intermediate form R₁ that is subsequently processed to a terminally glycosylated R_t form (Peter et al., 1994; Pind et al., 1994). In pulse-chase studies, ∼34% of VSV-G was terminally glycosylated in control cells after a 2 hour chase, but only ∼4% of VSV-G was terminally glycosylated in GBF1-depleted cells (Fig. 5D). Our immunofluorescence and biochemical findings confirm an essential role for COPI in the trafficking of VSV-G.

The exogenously expressed VSV-G is produced from a CMV promoter at significantly higher levels than endogenous cellular proteins. It is therefore possible that the trafficking requirements of VSV-G differ from those of endogenous proteins. Therefore, we analysed the trafficking of an endogenous type I transmembrane protein, E-selectin ligand 1 (ESL-1). This broadly expressed surface glycoprotein is a major ligand for E-selectin and has been implicated in leukocyte extravasation by facilitating leukocyte-endothelial cell interactions (Steegmaier et al., 1997; Wild et al., 2001). ESL-1 is synthesized in HeLa cells and localizes in thin cell-cell connections that contain fibronectin (Fig. 5E).

ESL-1 turns over with a t_1/2 of ∼24 hours (Steegmaier et al., 1997; Wild et al., 2001), suggesting that continuous delivery of newly synthesized ESL-1 to the plasma membrane is required for the steady-state plasma membrane localization. The surface delivery of endogenous ESL-1 was compared in control cells and in GBF1-depleted cells. In control cells cultured with scrambled siRNA for 3 days, ESL-1 is detected on plasma membrane projections in a pattern analogous to that in untreated cells (compare Fig. 5E with F). By contrast, in cells cultured with siRNA against GBF1 for 3 days, ESL-1 is not on the plasma membrane. Instead, ESL-1 appears to be retained intracellularly, with some ESL-1 localized in punctate and tubular GM130 structures (Fig. 5G). In this experiment, GBF1-depleted cells are identified by GM130 tubulation. The levels of total ESL-1 appear lower in GBF1-depleted cells, suggesting that inhibition in trafficking results in ESL-1 degradation. In support, western blotting shows reduced levels of ESL-1 in GBF1-depleted cells (Fig. 5H). The western blot also documents GBF1 depletion.

The inhibition in ESL-1 trafficking was confirmed biochemically. In control cells, ESL-1 is synthesized as a core-glycosylated Endo-H-sensitive 155-kDa form (Fig. 5I, red arrowhead). A fully glycosylated 160-kDa form that is resistant to Endo H accumulates after a 1 hour chase (green arrowhead). The level of newly synthesized ESL-1 remaining after the 1 hour chase is 100% of that detected at 0 hours of chase (Fig. 5I). By contrast, in GBF1-depleted cells, the core glycosylated 155-kDa form is not processed to the mature 160-kDa form (Fig. 5J, green arrowhead). Instead, a proportion of the core-glycosylated ESL-1 is chased into an intermediate form resistant to Endo H. A similar partial processing has been described for VSV-G in BFA-treated cells (Nehls et al., 2000). It is important to stress that terminal glycosylation does occur in GBF1-depleted cells, as shown by the processing of two soluble glycoproteins (see below). The level of radiolabelled ESL-1 remaining in GBF1-depleted cells after the chase is 65% of that detected at 0 hours of chase, consistent with intracellular degradation. Our immunofluorescence and biochemical studies indicate that trafficking of two transmembrane proteins, the exogenous VSV-G and the endogenous ESL-1, requires ARF-COPI events regulated by GBF1.

The inhibition of VSV-G and ESL-1 trafficking to the plasma membrane suggests that delivery of other proteins to the plasma membrane, including surface proteins involved in cell-cell and/or cell-matrix interactions, might be inhibited in GBF1-depleted cells. To explore this possibility, we assayed the motility of GBF1-depleted human D54 glioma cells on vitronectin (Lyons et al., 2002). Control untreated cells, cells transfected with scrambled siRNA or cells transfected with siRNA against GBF1 were plated on top of a trans-well filter, and the number of cells that migrated across the filter was analysed after 6 hours. An average of 52 control cells migrated across the filter in 6 hours (supplementary material Fig. S3A,D). Slightly fewer, an average of 47 cells transfected with scrambled siRNA migrated in the same time period (supplementary material Fig. S3B,D). Migration was significantly inhibited in GBF1-depleted cells, with an average of 14 cells migrating within 6 hours (supplementary material Fig. S3C,D). Western blotting confirms the efficient depletion of GBF1 in these experiments (supplementary material Fig. S3E). These results suggest that depletion of GBF1 inhibits trafficking of proteins involved in cell-cell and cell-matrix interactions and that GBF1 might regulate trafficking of numerous transmembrane proteins.

Numerous studies show that BFA treatment inhibits protein secretion from yeast and mammalian cells (De Lisle and Bansal, 1996; Fujiwara et al., 1988; Robin et al., 1996; Vogel et al., 1993), suggesting that the trafficking of soluble proteins also requires COPI recruitment to membranes. We therefore compared protein secretion from BFA-treated cells with secretion from control and GBF1-depleted cells (Fig. 6A-C). In agreement with previous reports, cells treated with BFA do not secrete proteins, whereas a set of radiolabelled endogenous proteins is released from control cells (Fig. 7B, compare lanes `+BFA' and `mock'). Unexpectedly, proteins are secreted from GBF1-depleted cells (Fig. 6B, lane `siRNA'). The difference in secretion between BFA-treated and GBF1-depleted cells is not due to a lack of protein synthesis as similar levels of radiolabelled proteins are detected in cell lysates from BFA-treated and GBF1-depleted cells (Fig. 6A). The western blot confirms GBF1 depletion in these experiments (Fig. 6C).

To characterize the kinetics of secretion from BFA-treated, control and GBF1-depleted cells, we measured total secretion in pulse-chase experiments (Fig. 6D). In agreement with previous data, BFA inhibits protein secretion. By contrast, proteins are efficiently secreted from control and GBF1-depleted cells. The secretion from GBF1-depleted cells appears slightly lower, perhaps owing to reduced incorporation of the label into secretory as opposed to cellular proteins. Again, the western blot confirms GBF1 depletion in these experiments (Fig. 6E).

The ability of GBF1-depleted cells to secrete soluble proteins was confirmed in HeLa cells (Fig. 6F-J). As in NRK cells, protein secretion in HeLa cells is inhibited by BFA (Fig. 6G). HeLa cells secrete a wider complement of proteins than NRK cells, with at least eight clearly recognizable bands secreted from control cells. Importantly, the same pattern of radiolabelled proteins, in similar relative proportions, is secreted from GBF1-depleted cells. The difference in secretion between BFA-treated and GBF1-depleted cells is not due to lack of protein synthesis as similar levels of radiolabelled proteins are detected in cell lysates from BFA-treated or GBF1-depleted cells (Fig. 6F). The western blot confirms GBF1 depletion in these experiments (Fig. 6H).

To ensure that the proteins detected in the medium reflect the release of secretory cargo, we examined the biosynthesis and release of two extracellular matrix proteins, fibronectin and laminin. The medium shown in Fig. 6G was immunoprecipitated with antibodies against fibronectin and laminin. In agreement with the block in total secretion, cells treated with BFA secrete neither fibronectin (Fig. 6I) nor laminin (Fig. 6J). By contrast, control cells synthesize and efficiently secrete both proteins. Significantly, efficient secretion of both proteins also is seen in GBF1-depleted cells.

The effects of GBF1 depletion on the trafficking of fibronectin were further characterized by examining the kinetics of secretion from control and GBF1-depleted cells (Fig. 6K-O). As shown in Fig. 6K, fibronectin was synthesized but not secreted from BFA-treated cells. By contrast, fibronectin was efficiently secreted from control cells, with the majority of fibronectin (∼80%) released within 4 hours (Fig. 6L). Significantly, fibronectin was released from GBF1-depleted cells with the same efficiency and kinetics as those of control cells (Fig. 6M). This experiment was repeated twice and the secretion kinetics were quantitated by densitometry of autoradiograms. As shown in Fig. 6O, analogous secretion rates are detected in control and GBF1-depleted cells. The western blot confirms GBF1 depletion in these experiments (Fig. 6N). The efficient transport and release of soluble proteins in GBF1-depleted cells suggests that COPI recruitment to membranes is not required for the maintenance of a secretory pathway capable of trafficking soluble proteins.

Here, we use depletion of GBF1 as a means of assessing the function of COPI within the secretory pathway. We show that depletion of GBF1 causes COPI dispersal and alters the morphology of early secretory compartments. The most dramatic disruption occurs at the cis-Golgi and results in extensive tubulation of that compartment. This is consistent with the concentration of GBF1 on the cis-Golgi (Kawamoto et al., 2002; Zhao et al., 2002). The phenotype of Golgi tubulation seen in GBF1-depleted cells differs significantly from the complete Golgi collapse seen in cells in which COPI recruitment is prevented by BFA treatment, expression of the dominant-negative E794K mutant of GBF1 and expression of the GDP-restricted T31N mutant of ARF1 (ARF1-T31N). How can we explain the relatively mild Golgi disruption observed in GBF1-depleted cells?

BFA, the E794K mutant of GBF1 (GBF1-E794K) and the ARF1-T31N mutant share a common molecular mechanism of action: all stabilize a complex of GBF1 and ARF-GDP on the membrane (Beraud-Dufour et al., 1998; Goldberg, 1998; Mossessova et al., 2003; Mossessova et al., 1998; Niu et al., 2005; Peyroche et al., 1999; Szul et al., 2005). We and others have shown that the ARF-GBF1 interaction is disrupted by GTP binding to the ARF (Niu et al., 2005; Szul et al., 2005). Importantly, GTP binding does not occur in BFA-treated cells (BFA inserts into the catalytic interface between the ARF and the GEF and prevents GDP displacement), in cells expressing GBF1-E794K (GBF1-E794K does not catalyse GDP displacement) or in cells expressing ARF1-T31N (ARF1-T31N has low affinity for GTP). All these situations result in a prolonged residency of the ARF-GBF1 complex on the membrane (Niu et al., 2005; Szul et al., 2005). It is possible that the protracted stay of the ARF and GBF1 on the membrane causes the phenotypic collapse of the Golgi. ARF regulates the activity of phospholipase D (PLD) and phosphatidylinositol 4-phosphate 5-kinase (PI4PK), and its prolonged association with the membrane might influence PLD- and PI4PK-mediated membrane remodelling (reviewed in Freyberg et al., 2003; Jenkins and Frohman, 2005; LaLonde et al., 2005). Similarly, the increased residency of GBF1 on the membrane increases the probability of GBF1 interacting with its partner proteins hGMH and the human homologue of yeast Drs2p (Chantalat et al., 2003; Chantalat et al., 2004). While the function of hGMH is unclear, yeast Drs2p is an aminophospholipid translocase known to influence lipid dynamics (Alder-Baerens et al., 2006; Hua et al., 2002; Natarajan et al., 2004; Pomorski et al., 2003). Hence, BFA treatment, expression of GBF1-E794K or ARF1-T31N all might cause cellular effects that do not reflect COPI function. Therefore, depletion of GBF1 allows the analysis of the effects of COPI dissociation while avoiding the confounding effects of the prolonged membrane association of the ARF-GBF1 complex. This study represents the first such analysis of GBF1-depleted cells.

In GBF1-depleted cells, the cis-Golgi tubules connect to ERGIC elements, suggesting that tubular continuities might form normally between the ERGIC and the Golgi and that COPI might regulate the frequency of such connections. The tubular pathway amplified in GBF1-depleted cells has been observed in normal cells: GM130 tubules linking peripheral ERGIC structures and moving in an anterograde and retrograde manner between the ERGIC and the Golgi have been reported (Marra et al., 2001; Trucco et al., 2004).

Interestingly, GBF1-depleted cells contain distinct ERGIC and Golgi compartments (albeit with tubulated cis-Golgi elements), suggesting that COPI recruitment is not essential for at least the partial differentiation of post-ER compartments of the secretory pathway. That protein sorting occurs in GBF1-depleted cells is shown by the dynamic behavior of ERGIC58 and the cis-Golgi marker GRASP65. Live imaging of ERGIC58 and GRASP65 shows that GRASP65 is preferentially sorted into tubules that form from ERGIC58-positive punctate elements. The retention of ERGIC58 in the punctate structures might be mediated by interactions with matrix or cytoskeletal components. The sorting of GRASP65 away from ERGIC58 confirms that differentiation of membranes of distinct composition occurs in GBF1-depleted cells. This unexpected finding suggests that compartment differentiation might occur in the absence of COPI-mediated sorting. This agrees with recent studies of Golgi reassembly following a BFA wash-out in cells expressing the dominant activating mutant Q71L of ARF1 (which mimics the ARF-GTP form) in which Golgi subcompartments differentiate in the absence of repeated cycles of COPI recruitment (Bannykh et al., 2005).

We cannot exclude the possibility that a fraction of COPI remains associated with membranes in GBF1-depleted cells that is below the level of detection of our immunofluorescence analysis. Theoretically, such COPI might correlate with the relatively mild disruption of the secretory pathway in GBF1-depleted cells. We do not think that such COPI is recruited by GBF1 owing to the extremely high silencing (>90%) of GBF1.

The COPI dispersion observed in GBF1-depleted cells has a dramatic effect on trafficking of select cargos. Specifically, while soluble secretory proteins appear to traffic efficiently in GBF1-depleted cells, transmembrane proteins are arrested in transit. The efficient secretion of multiple soluble cargoes suggests that a functional secretory pathway is present in GBF1-depleted cells. However, transmembrane proteins appear unable to enter and navigate through the pathway. We observe arrest of VSV-G within the ER, suggesting that GBF1-mediated COPI events are required for exit from that compartment. Interestingly, it appears that another transmembrane protein, ESL-1, exits the ER in GBF1-depleted cells but arrests while trafficking within the disrupted cis-Golgi elements. This raises the possibility that GBF1-mediated COPI recruitment facilitates the trafficking of different cargoes at different stages of the secretory pathway. We cannot exclude the possibility that the observed inhibition of the trafficking of transmembrane proteins arises from the perturbation of the cis-Golgi network/cis-Golgi function, in addition to disturbances in COPI function. Whether GBF1 depletion affects the trafficking of all or only a subset of transmembrane proteins is unknown. It is possible that GBF1 facilitates trafficking of many cargo proteins as multiple alternatively spliced GBF1 isoforms, perhaps regulating trafficking of different cargos, have been identified (Claude et al., 2003).

Our findings support a model in which a functional secretory pathway is maintained in GBF1-depleted cells, despite the dispersal of COPI (Fig. 7A). The pathway includes the largely unaffected ER, ER exit sites and ERGIC, the extensively tubulated cis-Golgi elements and a relatively unaltered medial- and trans-Golgi. We postulate that soluble cargos are first delivered from the ER to the ERGIC through a process that does not require COPI. To our knowledge, no requirement for COPI during trafficking of soluble cargo from the ER to the ERGIC has been reported. Once in the ERGIC, soluble proteins enter the tubular connections between the ERGIC and the Golgi and traffic to the Golgi. Soluble proteins appear to transit through the Golgi unimpaired, as shown by the efficient secretion of glycosylated proteins from GBF1-depleted cells. The kinetics of secretion of multiple proteins from GBF1-depleted cells appear normal, suggesting that COPI dispersal does not fundamentally affect their trafficking.

Instead, the primary function of COPI appears to be in trafficking of transmembrane proteins along a pre-existing secretory pathway. In GBF1-depleted cells, transmembrane proteins do not enter the pathway. We observed arrest of trafficking during exit from the ER and during exit from the ERGIC. The inhibition occurs despite the presence of functional transport routes between the ER and the plasma membrane, as shown by the trafficking of soluble proteins. This suggests that COPI-mediated sorting mechanisms are required for gating transmembrane proteins into the secretory pathway.

We postulate that, in normal cells, soluble proteins also utilize transient tubular connections to traffic between the ERGIC and the Golgi in a COPI-independent manner (Fig. 7B). In support of this, an extensive network of tubules has been shown to mediate the trafficking of soluble cargo (GFP targeted to the ER lumen by an N-terminal signal peptide) from the ERGIC to the Golgi in control cells (Blum et al., 2000). We suggest that the tubules also constitute the conduits for trafficking of transmembrane cargoes. However, transmembrane proteins enter the tubules after being sequestered into membrane subdomains by COPI-mediated sorting. The formation of such `patches' of transmembrane proteins requires GBF1. Our studies for the first time dissect the limited role of COPI in generating a functional secretory pathway from its essential function in trafficking of transmembrane cargo.

Rabbit polyclonal antibodies against GBF1 have been described previously (Garcia-Mata et al., 2003). Monoclonal antibody G1/93 against ERGIC53 (Schindler et al., 1993) was obtained from H. P. Hauri (University of Basel, Basel, Switzerland), rabbit polyclonal antibodies against ESL-1 were obtained from M. K. Wilde (University of Muenster, Muenster, Germany), polyclonal antibodies against laminin and fibronectin were obtained from A. Woods (University of Alabama at Birmingham, AL). The following commercially available antibodies were used: monoclonal anti-β-tubulin from Upstate (Lake Placid, NY), polyclonal anti-Myc from Santa-Cruz Biotechnology (Santa Cruz, CA), polyclonal anti-GFP from Abcam (Cambridge, MA), polyclonal anti-β-COP, monoclonal anti-GM130, polyclonal anti-COPII and polyclonal anti-calreticulin from Affinity Bioreagents (Golden, CO), monoclonal anti-fibronectin from BD Transduction Laboratories (San Jose, CA), and polyclonal anti-actin from Sigma-Aldrich (St Louis, MO). Rabbit polyclonal serum against ARF1 was obtained from S. Cockcroft (University College London, UK), rabbit polyclonal serum against ARF3 (R1023), rabbit polyclonal serum against ARF4 (R891) and rabbit polyclonal serum against ARF5 (R1525) were obtained from R. A. Kahn (Emory University School of Medicine, XX). Secondary antibodies conjugated with HRP, Alexa 488 or Alexa 594 were from Molecular Probes (Eugene, OR). Brefeldin A (BFA) was from Sigma-Aldrich (St Louis, MO). Endo-H_f enzyme was from New England BioLabs (Ipswich, MA). Autoradiography enhancer was from PerkinElmer Life and Analytical Sciences (Boston, MA); ³⁵S-Met/Cys was from MP Biomedicals (Irvine, CA). siLentFect lipid transfection reagent was from BioRad Laboratories (Hercules, CA). Trichloroacetic acid was acquired from Fisher Scientific (Fair Lawn, NJ). Construct VSV-G-GFP and ERGIC58-YFP were from J. Lippincott-Schwarz (NIH, Bethesda, MD). GRASP65-GFP was acquired from K. E. Howell (University of Colorado, Denver, CO).

HeLa cells and NRK cells were grown in minimum essential medium (MEM) and Dulbecco's modified Eagle's medium (DMEM), supplemented with glucose and glutamine (Mediatech, Comprehensive Cancer Center of the University of Alabama at Birmingham, AL), respectively. Media were supplemented with 10% (for HeLa) or 5% (for NRK) foetal bovine serum (FBS; Life Technologies, Grand Island, NY), 100 units/ml of penicillin and streptomycin (Invitrogen Corporation, Grand Island, NY) and 1 mM sodium pyruvate. Cells were grown at 37°C in 5% CO₂ in 6-well dishes until ∼70% confluent and were transfected using Mirus TransIT-LT1 transfection reagent from Mirus Bio Corporation (Madison, WI), according to the manufacturer's instructions.

#2siRNA against human GBF1 (5′-CGAAAUGCCCGAUGGAGCAtt-3′) and nontargeting siRNA were designed and synthesized as annealed primers by Ambion (Austin, TX); siRNA against rat GBF1 (5′-AGUGGAGGGUGGUUGUCAAtt-3′) and #1siRNA against human GBF1 (5′-AGGUGGAGGGCGGCUGCCACtt-3′) were synthesized by IDT Integrated DNA Technology (Coralville, IA). HeLa or NRK cells were transfected with siRNA using siLentFect lipid from BioRad Laboratories (Hercules, CA) according to the manufacturer's instructions.

Immunofluorescence was performed as described previously (Szul et al., 2005). Live imaging was performed on a Leica DMRXE upright, epifluorescence/Nomarski microscope outfitted with Leica TCS SP2 laser confocal optics (Leica; Exton, PA). The system was equipped with argon ion, solid-state, and helium-neon lasers for the imaging of a wide range of green and yellow fluorochromes. Precise control of fluorochrome excitation and emission is afforded, respectively, by an acousto-optical tunable filter and a TCS SP2 prism spectrophotometer. Optical sections through the Z axis were generated using a computer-controlled focus step motor. Flattened projections of image stacks and 3D renderings were prepared using proprietary confocal imaging software from Leica. The 100× oil (NA 1.4) objective was used in all experiments. Exact settings to separate YFP and GFP were: GFP settings: laser 488, filter 496-515 nm, pinhole 370; YFP settings: laser 514, filter 565-612 nm, pinhole 370.

Control HeLa cells or cells silenced for 72 hours with siRNA against GBF1 were starved for 1 hour with medium lacking methionine (Met) and cysteine (Cys). Cells were then labelled with ³⁵S-Met/Cys for the indicated times. Media were collected and cells were solubilized in RIPA buffer (150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS and 50 mM Tris, pH 8.0) supplemented with complete protease inhibitors (Roche). Protein concentrations of lysates were analysed using the BioRad protein assay and normalized. Aliquots were analysed by SDS-PAGE. Aliquots were subjected to immunoprecipitation as follows: samples were centrifuged at 4180 g for 15 minutes; supernatants were incubated at 4°C with 2 μg of polyclonal antibodies against fibronectin or laminin for 2 hours, followed by a 2 hour incubation with 20 ml of 50% (v/v) protein-A sepharose 4FF. Beads were recovered by centrifugation and washed four times with RIPA buffer containing protease inhibitors. Precipitates were analysed by SDS-PAGE. Following SDS-PAGE, gels were incubated in autoradiography enhancer, dried and exposed to X-ray film.

To analyse the glycosylation of ESL-1, HeLa cells were transfected with scrambled siRNA or siRNA against GBF1 for 72 hours. Cells were washed with PBS and starved for 1 hour with medium lacking Met and Cys, followed by labelling with ³⁵S-Met/Cys for the indicated times. Cells were lysed in RIPA and centrifuged at 4180 g for 15 minutes. Aliquots of the supernatants were incubated with Endo-H_f at 37°C for 45 minutes. Untreated and treated samples were incubated at 4°C with 2 μg of polyclonal antibodies against ESL-1 for 2 hours, followed by a 2 hour incubation with 20 ml of 50% (v/v) protein-A sepharose 4FF. Beads were recovered by centrifugation, washed four times with RIPA buffer containing protease inhibitors, and the precipitates were analysed by SDS-PAGE. The gels were incubated in autoradiography enhancer, dried and exposed to X-ray film.

HeLa cells grown in a 10 cm dish were treated with 5 μg/ml BFA for 45 minutes. Cells were washed twice with PBS containing 5 μg/ml BFA and lysed in HKMT buffer [20 mM HEPES pH 7.4, 0.1 M KCl, 1 mM MgCl₂, 0.5 % (w/v) Triton X-100] containing 5 μg/ml BFA. Lysates were then clarified by centrifugation and incubated with 1 μg mouse monoclonal antibody against GBF1 or Rab8 for 1 hour at 4°C. Immune complexes were collected by incubating with 10 μl of protein G sepharose (Zymed) for 1 hour at 4°C and eluted from the beads by boiling in SDS sample buffer. Bound fractions were analysed by SDS-PAGE and western blotted with polyclonal antibodies specific for each ARF isoform [antibodies against ARF1 were sourced from S. Cockcroft; and against ARF3 (R1023), ARF4 (R891) and ARF5 (R1525) were from R. A. Kahn].

HeLa cells 48 hours after transfection with scrambled siRNA or siRNA against GBF1 were transfected with VSV-G-GFP protein and incubated at the non-permissive temperature of 42°C for ∼12 hours. Cells were then moved to the permissive temperature of 32°C, incubated for 1, 2 or 12 hours and processed for immunofluorescence. Alternatively, HeLa cells were grown at the permissive temperature of 32°C. Cells were transfected with scrambled siRNA (`src') or with siRNA against GBF1 (`siRNA') for 48 hours and then were transfected with ts045VSV-G-GFP for an additional 24 hours. Cells were pulsed for 30 minutes with ³⁵S-Met/Cys and then chased for the indicated times. Cell lysates were mock treated or treated with Endo-H_f. VSV-G was immunoprecipitated and analysed by SDS-PAGE and fluorography.

NRK cells 72 hours after depletion with siRNA were labelled with ³⁵S-Met/Cys for the indicated times. Media were collected and cells were solubilized in RIPA buffer. Aliquots of cell lysates and media were precipitated for 1 hour on ice with 20% trichloroacetic acid, the pellets were washed twice in 10% trichloroacetic acid, diluted in 5× SDS-PAGE loading buffer, mixed with scintillation fluid and counted in a scintillation counter. The amount of secreted protein at each time is represented as the percentage of radioactivity in the medium and the lysate.

We thank J. Lippincott-Schwartz and R. Kahn for reagents and comments, and A. de Matteis, J. Casanova, R. Garcia-Mata and Melanie Styers for helpful discussions.