LG186: An Inhibitor of GBF1 Function that Causes Golgi Disassembly in Human and Canine Cells


David J. Stephens, david.stephens@bristol.ac.uk


Brefeldin A-mediated inhibition of ADP ribosylation factor (Arf) GTPases and their guanine nucleotide exchange factors, Arf-GEFs, has been a cornerstone of membrane trafficking research for many years. Brefeldin A (BFA) is relatively non-selective inhibiting at least three targets in human cells, Golgi brefeldin A resistance factor 1 (GBF1), brefeldin A inhibited guanine nucleotide exchange factor 1 (BIG1) and brefeldin A inhibited guanine nucleotide exchange factor 2 (BIG2). Here, we show that the previously described compound Exo2 acts through inhibition of Arf-GEF function, but causes other phenotypic changes that are not GBF1 related. We describe the engineering of Exo2 to produce LG186, a more selective, reversible inhibitor of Arf-GEF function. Using multiple-cell-based assays and GBF1 mutants, our data are most consistent with LG186 acting by selective inhibition of GBF1. Unlike other Arf-GEF and reported GBF1 inhibitors including BFA, Exo2 and Golgicide A, LG186 induces disassembly of the Golgi stack in both human and canine cells.

Much of our knowledge of the secretory pathway has come from genetic, biochemical and imaging experiments (1). In each of these contexts, chemical intervention using small molecule inhibitors has already proved extremely useful (2). For example, much of what we know about the function of ADP ribosylation factor 1 (Arf1) in membrane trafficking comes from studies using brefeldin A (BFA) (for examples, see 3–6). BFA inhibits activation of Arf1 by targeting guanine nucleotide exchange factors (GEFs) (7,8) and results in multiple morphological and functional changes to animal cells inducing very rapid and dramatic tubulation of early endosomes and a redistribution of the trans-Golgi network (TGN) and endosomes (4), and the microtubule-dependent fusion of these two compartments (9). Endoplasmic reticulum (ER)-to-Golgi transport is also inhibited and the Golgi apparatus merges with the ER as a result of continuing retrograde transport in the absence of ongoing anterograde traffic (3). The effects of BFA are caused by its inhibition of GEFs that act on members of the Arf family of small GTPases (7), notably Arfs 1, 3, 4 and 5 (Arf2 has been lost in humans during evolution). We now know that BFA is not a competitive inhibitor of either Arf or Arf–GEF function but instead stabilizes an abortive Arf–GDP/Arf–GEF complex by binding at the interface between Arf-GDP and the GEF (10). This mode of uncompetitive inhibition of biomolecular interactions has been generalized as interfacial inhibition (10,11). The diversity of BFA-induced effects is caused by its action on multiple Arf/Arf–GEF complexes and consequently on multiple transport pathways. Recent data have shown that activation of Arf1 at the Golgi is mediated by Golgi brefeldin A resistance factor 1 (GBF1), a Golgi-specific GEF for Arfs (12). In contrast to its original characterization, GBF1 is indeed BFA sensitive (12) and is recruited to Golgi and pre-Golgi membranes, but not to endosomes (13). Brefeldin A inhibited guanine nucleotide exchange factor 1 (BIG1) and Brefeldin A inhibited guanine nucleotide exchange factor 2 (BIG2) have been described as TGN/endosomal Arf–GEFs (13), not present on early Golgi membranes, and both are also sensitive to BFA (14). While it remains formally possible that the BFA-induced tubulation of endosomes results from an effect on a target other than BIG1 or BIG2 (such as perhaps BARS), indistinguishable tubulation of recycling endosomes occurs on siRNA-mediated suppression of BIG2 or overexpression of dominant negative BIG2 mutants (for examples, see 15–17). BFA is ineffective in causing Golgi collapse in canine cells (18). This has been determined to be because of a single substitution within the Arf binding site of canine GBF1 (19–21). Canine cells are widely used, notably in the field of cell polarization and epithelial differentiation (22) as well as to study kidney function (23). The inability to use BFA to study processes within these cells clearly limits our understanding of early secretory pathway function during these important processes.

Recent years have seen the application of chemical genetic screens that have identified inhibitors of secretory pathway function. Notable among these is the identification of Exo2 as an inhibitor of early secretory pathway function (24) that perturbs trafficking of protein toxins (24,25), and Golgicide A (GCA), which is proposed to be a selective GBF1 inhibitor whose binding site appears to overlap with that of BFA (21). Our own characterization of Exo2 showed it to have some selectivity for Arf/Arf–GEF function above that of BFA (25). Incubation of cells with Exo2 induces a rapid collapse of the Golgi but does not cause a tubulation of recycling endosomes (25) which is seen with BFA (2). The precise target of Exo2 has not previously been identified. We sought to define the target(s) of Exo2 in cells and also to use this small molecule as a template to develop selective inhibitor of GBF1 function.


Exo2 targets Arf–GEF activity in cells

In a specific Arf-activation assay we found that Exo2, like BFA, caused a significant decrease in Arf activation without altering the pool of total cytosolic Arf (Figure 1A). Based on our previous work, we suspected that Exo2 might be a selective inhibitor of GBF1 owing to its effects on Golgi function but not on endosome tubulation (25). Arf–GEFs activate nucleotide exchange by their Sec7 domain, which is highly conserved in sequence and structure. Comparison of the sequence of the Sec7 domains of human GBF1 and BIG Arf–GEFs shows that GBF1 contains a tripeptide insertion near the Arf- and BFA-binding site (Figure 1B). The Sec7 domains of GBF1 and BIG2 were modeled from crystal structures of related Arf–GEFs (10,26). Figure 1C (enlarged in Figure 1D) shows a model of GBF1 in which the three-residue insertion results in an extra turn of helix 8 and an alternative conformation of the subsequent helix 8–helix 9 loop (in orange) as compared to BIG1 (in red; BIG2 is identical to BIG1 in this region). As a consequence, the Arf- and BFA-binding site is somewhat enlarged in GBF1. BFA and Exo2 share similarities in their overall shape and hydrogen bonding potential (Figure S1A,B). From these homology models, we propose a mode of binding of Exo2 to GBF1 that partly mimics BFA (Figure 1C). Detailed interactions of Exo2 with GBF1 in this model are shown in Figure 1E. The phenoxyl group of Exo2 is positioned to make a hydrogen bond with Y828, analogous to that between this residue and the hydroxyl group of BFA. The M832 side chain is juxtaposed over one face of the Exo2 phenol ring. The hydrazine linker and part of the tricyclic moiety of Exo2 pass through the BFA site, where they make less close-packed interactions with the Sec7 domain compared to BFA. The tricycle of Exo2 forms numerous van der Waals interactions with T835, N839, H840 and K844. The cyclohexenyl part occupies the space made available by the longer helix 8 resulting from the extra tripeptide. In BIG1 and BIG2, the different conformation of the helix 8–helix 9 loop results in potential clashes with Exo2 if modeled with the same orientation (Figure 1C, loop in red). We propose that this is consistent with potential selective inhibition of GBF1 by Exo2 [(25) and see below]. In an attempt to enhance the selectivity for GBF1 we decided to derive from Exo2 a molecule (LG186) with an enlarged cyclooctenyl ring compared to the cyclohexenyl ring of Exo2 (Figure S1C). The rationale for this was based on the enlarged binding cavity seen in GBF1 compared to BIG1 and BIG2 models (Figure 1C). The route to synthesis of LG186 is shown in Figure S1D. A model of LG186 bound to GBF1 and BIG2 is shown in Figure 1C (in purple) and (to GBF1 only) in space-filling form in Figure 1F. Note the enlarged tricyclic ring of LG186 in closer contact with N839, H840 and K844 of GBF1.

Figure 1.

Exo2 inhibits activation of Arf1. A) Alignment of sequences of the three human large Arf–GEFs GBF1, BIG1 and BIG2 showing the additional tripeptide loop within the Sec7 domain of GBF1 (residues NAP). * Indicates residues common to the BFA and LG186 binding sites; # indicates residues unique to the predicted LG186 binding site; / indicates the three-residue GBF1 insert (NAP). Amino acids 826–852 of human GBF1 are shown. B, enlarged in C) Homology model showing the Sec7 domain of GBF1 in complex with BFA (khaki), Exo2 (pink) and LG186 (purple). The GBF1 helix 8–helix 9 loop is shown in orange, the equivalent loop from the model of BIG2 is shown in red. D, E) Docking of Exo2 (D) or LG186 (E) into the modeled Sec7 domain of human GBF1. Side chains in contact with the compounds are shown as space-filling representations with transparent surfaces for the two compounds.

Exo2 and LG186 inhibit GBF1 function

To validate GBF1 as a target of these molecules, we analyzed the intracellular distribution of green fluorescent protein (GFP)-GBF1 in the presence of each inhibitor. BFA, Exo2 and LG186 all induce a recruitment of GFP-GBF1 to juxtanuclear membranes (Figure 2A) as has been shown previously for BFA (27). Fluorescence recovery after photobleaching (FRAP) showed that GFP-GBF1 was less dynamic compared to controls (Figure 2B), again consistent with an inhibition of GBF1 function (28). Consistent with the inhibition of GEF function, Figure 2C–E shows that LG186 results in a decrease in Arf activation at 50 (Figure 2C) and 10 µm (Figure 2D, quantified in Figure 2E). Ten micromoles of LG186 (but not 10 µm Exo2) shows statistically significant (p < 0.05) inhibition of Arf activation compared to controls. An important point here is that the antibody used (1D9) is a pan-Arf antibody that recognizes all isoforms (29). To validate the activity of LG186 in more detail, we undertook in vitro assays of GEF activity, using a protocol successfully used to characterize other cell-active Arf–GEF inhibitors (30,31). LG186 was used at a concentration of 10 µm for these assays as this was found not to result in aggregation of LG186 under the conditions of these assays (Figure S1E), yet is also an effective concentration in cell-based assays (Figure S2A,B). Recombinant GBF1 showed no activity in any assay, precluding us from direct analysis of this GEF (M. Z. and J. C., data not shown). Figure 3 shows the observed rate constants (kobs) for the Sec7 domain of the cytohesin family Arf–GEF ARNO (Figure 3A), the Sec7-PH domains of brefeldin-resistant Arf–GEF 2 (BRAG2) (GEP100) (Figure 3B), the isolated Sec7 domain of BIG1 (Figure 3C) and the combined DCB-HUS-Sec7 domains of BIG1 (Figure 3D) in nucleotide exchange factor assays in the presence of dimethyl sulphoxide (DMSO) or inhibitors (10 µm) as indicated. We also included a recently published compound reported as a selective GBF1 inhibitor, GCA (21), in these assays. LG186, Exo2, GCA and BFA have no significant activity against ARNO. BRAG2 is insensitive to BFA, and is very weakly inhibited by the other compounds. A slight inhibition of activity of both BIG1 constructs is observed with LG186, Exo2 and GCA, which is considerably less than that seen with BFA. Although we did not observe a strong inhibition by any of the compounds except for BIG1/BFA, the moderate effects are consistent with the hypothesis that LG186, Exo2 and GCA can act as inhibitors of Sec7 domains. We think that these results also reinforce the interpretation that LG186 binds to a site on the Sec7 domain that overlaps with the BFA-binding site. Indeed, the very weak ability of BIG1 to bind and/or to be inhibited by LG186 is consistent with the fact that its predicted binding site is partially conserved between GBF1 and BIG1 (see alignment in Figure 1B), with the conserved residues overlapping the BFA-binding site (not shown). The replacement of H840 in GBF1 by a proline in BIG1, as well as the overlap of the helix 8–helix 9 loop with the inhibitor in BIG2 (modeled in Figure 1C), would explain why Exo2 and LG186 bind less favorably to BIG1 or BIG2.

Figure 2.

LG186, a potent and selective inhibitor of GBF1. A) HeLa cells expressing GFP-GBF1 were imaged over time (as indicated) at 37°C after addition of DMSO, BFA (10 µm), Exo2 (50 µm) and LG186 (50 µm). Note that all three inhibitors induce a juxtanuclear accumulation of GFP-GBF1. B) Quantification of fluorescence recovery of GFP-GBF1 into the bleached area over time in cells incubated with DMSO (black), BFA (red), Exo2 (green) or LG186 (blue). Bars show SD of at least three independent experiments. C, D) Pull-down of Arf-GTP using the GGA-GAT domain from cells incubated with (C) DMSO, Exo2 (50 µm), BFA (10 µm or LG186 (50 µm) and (D) DMSO, Exo2 (10 µm), BFA (10 µm) and LG186 (10 µm). E) Quantification of Arf activity remaining; the Y -axis shows arbitrary units of the ration of the Arf pull-down: total Arf ratio from three experiments, one example of which is shown in (D). Asterisk indicates p < 0.05. Note that the antibody used for Arf detection recognizes all human Arf isoforms (29).

Figure 3.

Measurement of exchange factor activity of the human Arf–GEFs in vitro in the presence of 10μmof indicated chemicals or DMSO alone. A) Against the Sec7 domain of ARNO. B) Against the Sec7-PH domains of BRAG2. C) Against the Sec7 domain of BIG1. D) Against the DCH-HUS-Sec7 domains of BIG1. Observed rate constants are shown (kobs).

Exo2 and LG186 cause a reversible redistribution of the Golgi in human cells

Consistent with the known effects of BFA and the use of GBF1 mutants or GBF1 depletion by RNAi (5,12), either 50 µm Exo2 or LG186 result in a loss of coat protein I (COPI) from Golgi membranes and collapse of the Golgi apparatus (Figure 4A,B). LG186 is also effective at 10 µm (Figure S2). Phenotypic effects on Golgi organization are seen at 5 µm (note reduced intensity of GalT labeling within the Golgi in Figure S2B). Figure S2B shows that BFA causes a loss of COPI from the Golgi within 2 min, and GCA acts within 5 min with LG186 taking up to 15 min to redistribute COPI from the Golgi. Similar observations are seen using a transmembrane marker of the Golgi, NA-GFP (Figure S2C). The delayed time–course for LG186 could be because of its lower solubility compared to BFA or GCA or to a different mechanism of inhibition. As an inhibitor of GBF1 function, one would expect LG186 to provide a robust block of secretory traffic as is seen with both BFA and GCA. Figure S3 shows that transport of both transmembrane cargo [tsO45-G-YFP (yellow fluorescent protein), Figure S3A] and small soluble cargo (neuropeptide Y-Venus, Figure S3B,C) is blocked in the presence of LG186, both markers remaining within the ER. In both cases the block is absolute, no plasma membrane tsO45-G-YFP was detected and no fusion events of NPY-Venus at the plasma membrane were observed using TIRF (total internal reflection fluorescence microscopy) imaging. Again consistent with the known cellular functions of GBF1 and effects of BFA, coat protein II (COPII)-coated ER exit sites (Sec31A labeling) remain present with some enlargement and a bias toward a more peripheral distribution indistinguishable from those seen with BFA (10 µm) (Figure 4B).

Figure 4.

LG186 reversibly inhibits GBF1 function in human (HeLa) cells. One-hour incubation with (A) BFA (10 µm), Exo2 (50 µm) and LG186 (50 µm) results in the loss of COPI from membranes (green in merge) and collapse of the Golgi apparatus (GM130, red in merge). (B) BFA, Exo2 and LG186 have no significant effects on the COPII coat (Sec31A, green in merge) despite inducing collapse of the Golgi (58k, red in merge). (C) BFA, but not Exo2 or LG186, causes a tubulation of the TGN (TGN46, green in merge). Giantin was used to mark the Golgi (red in merge). (D) Early and recycling endosomes containing the TfnR (green in merge) are tubulated in the presence of BFA, become enlarged but not tubulated in the presence of Exo2, and are dispersed but not tubulated or enlarged by LG186. Collapse of the Golgi was verified using giantin labeling (red in merge). (E) BFA, Exo2 and LG186 are all reversible. The effects of each inhibitor on TGN (TGN46, green in merge) and Golgi (GM130, red in merge) can be reversed on washout of the compounds (washout in drug-free medium for 3 h). Three examples are shown for LG186. Bars (all panels) = 10 µm.

Consistent with our previous work (25), Exo2 induces a redistribution of the TGN marker, TGN46 (Figure 4C) but not a tubulation as is seen with BFA (2). LG186 also induces redistribution of the TGN, but in contrast to Exo2-treated cells, TGN46 is not retained in punctate intracellular compartments (Figure 4C). BFA (2) but not Exo2 (25) or LG186 induces a tubulation of recycling endosomes (Figure 4D). We also noted an extensive enlargement of endosomes labeled with the transferrin receptor in the presence of Exo2 (Figure 4D). Neither BFA nor LG186 caused enlargement of transferrin-receptor-positive endosomes. In other work we (32) and others have shown that expression of a dominant negative mutant (17) or suppression of BIG2 expression using siRNA (15,32,33) results in a tubulation of recycling endosomes. We find that LG186 does not induce tubulation of early endosomal membranes, and has no significant effect on the recycling of transferrin (Figure S3D,E). Exo2 causes a slight delay in the recycling of labeled transferrin and this could be coupled to the observed vacuolation of early endosomes (Figure 4D). With the information that neither Exo2 nor LG186 induces tubulation of endosomes (Figure 4D), coupled with our modeling data (Figure 1C), we conclude that while BFA inhibits BIG2 function, Exo2 and LG186 do not. Further evidence of specificity comes from the analysis of the localization of the AP1 clathrin adaptor. AP1 is recruited to TGN membranes in a BIG1-dependent manner (15) but its recruitment is independent of GBF1 (21). Consistent with this, we find that LG186, and indeed GCA, resulted in clear collapse of the Golgi apparatus without concomitant loss of AP1 from TGN46-positive membranes (Figure S4; retinal pigment epithelial cells were used here because these have a more defined TGN morphology than HeLa allowing unequivocal identification of the localization of AP1 to TGN membranes). In contrast, BFA causes complete loss of both Golgi structure and AP1 localization. This is only evident at short times of incubation (10-min incubation with LG186 or Exo2, 5 min with BFA or GCA) because ultimately the TGN also becomes completely dispersed.

A key feature of BFA is its reversibility. We found that both Exo2 and LG186 were also reversible in their effects on Golgi and TGN membranes following washout (3 h) from treated cells (Figure 4E). Notably, cells recovered less well from LG186 washout showing a more variable recovery of Golgi and TGN localization than seen following washout of BFA or Exo2. Complete washout is possible but is less readily achievable than with BFA.

The additional loop within GBF1 confers sensitivity to LG186

Our modeling of the Sec7 domain of GBF1 suggests that the structural basis for selective inhibition of GBF1 by LG186 (and possibly Exo2) lies in the extra tripeptide comprising residues 846–848. The tripeptide is not directly part of the inhibitor-binding site, but it imposes an extra turn to helix 8 which changes the shape of the active site. Therefore, we generated a mutant of GBF1, in which these three additional amino acids (‘NAP’ using single letter amino acid code) were deleted and determined the effect of expressing this mutant on inhibitor-induced redistribution of GalT and giantin. Expression of GFP-GBF1 ΔNAP in HeLa cells conferred resistance to LG186 (Figures 5A and 6D). A similar loop deletion (GBF1 ΔQNA) was shown to protect cells against GCA (21); GFP-GBF1 ΔNAP similarly protects the Golgi (Figure 6E) and TGN (Figure 6G) against collapse in the presence of GCA. Expression of wild-type GFP-GBF1 (Figure 5B) results in a partial protection as has been described previously for GCA [(21) and see Figure 6F]. This was indeed the basis for the initial characterization of GBF1 as a BFA-resistance factor (34). Of note, there is also a direct correlation between expression level of these constructs and the level of protection against these inhibitors. Any detectable expression of GFP-GBF1 ΔNAP protects cells against the effects of LG186 or GCA, while only very high-level expression GFP-GBF1 WT provides any protection against Golgi disruption (data not shown but see for example Figure 6D). In contrast to the effects of LG186, cells remained susceptible to BFA whether expressing GFP-GBF1 WT or GFP-GBF1 ΔNAP (Figure 5C,D, Figure S6B). This was predicted from the modeled complexes of BFA with GBF1 which show that BFA does not contact the ‘NAP’ loop of GBF1 (Figure 1C). These data validate GBF1 as the primary target of LG186 and confirm the role of amino acids ‘NAP’ in terms of specificity of binding, probably by imposing an alternative conformation to helix 8. Expression of GFP-BIG1 was unable to prevent LG186-induced Golgi disassembly (Figure 5E). GFP-BIG1 localizes to a juxtanuclear compartment. BFA, GCA and LG186 all induce a redistribution of GFP-BIG1 and concomitant disassembly of the Golgi.

Figure 5.

Specificity for GBF1 is conferred by the additional loop within the Sec7 domain. Sensitivity of GBF1 to LG186 (50 µm, 1-h incubation, HeLa cells) is conferred by the additional tripeptide loop ‘NAP.’ (A) Deletion of this loop (GFP-GBF1 ΔNAP) confers resistance to LG186. (B) GFP-GBF1 WT confers only slight resistance to LG186. Neither GFP-GBF1 ΔNAP (C) nor GFP-GBF1 WT (D) confer resistance to BFA (10 µm), as is predicted from homology modeling. Cells were labeled for giantin to mark the Golgi matrix component and GalT, a transmembrane marker of Golgi content. Merge panes show images of GFP-GBF1 in green (ΔNAP or WT as indicated) and GalT in red. Bars = 10 µm. (E) Overexpression of GFP-BIG1 (green in merge) does not protect the cells from BFA, GCA or LG186. The Golgi content was labeled using GalT (red in merge).

Figure 6.

Deletion of the tripeptide loop ‘NAP’ from GBF1 protects the TGN from disruption by LG186 and GCA. A–D) Deletion of the tripeptide loop ‘NAP’ from GBF1 protects the TGN from disruption by incubation for 1 h with (C) Exo2 (50 µm) and (D) LG186 (50 µm) but not (B) BFA (10 µm), restoring a normal juxtanuclear pattern of TGN46 labeling [compare to (A)]. Bar = 10 µm. Sensitivity of GBF1 to GCA is conferred by the additional tripeptide loop ‘NAP.’ E) Deletion of this loop (GFP-GBF1 ΔNAP) confers resistance to GCA (50 µm). F) GFP-GBF1 WT does not confer resistance to GCA (50 µm) as shown previously using a similar mutant (21). Cells were labeled for giantin to mark the Golgi matrix component and GalT, a transmembrane marker of Golgi content. Merge panels show images of GFP-GBF1 ΔNAP in green and GalT in red. (G) GFP-GBF1 ΔNAP also protects the TGN against dispersion induced by GCA (50 µm) as judged by TGN46 labeling. HeLa cells used for all panels. Bar = 10 µm.

LG186 inhibits canine GBF1 function

BFA (18,35,36) and GCA (21) are ineffective against GBF1 in some species including canine. We found that Exo2 is also ineffective in perturbing the Golgi within Madin-Darby canine kidney (MDCK) cells (Figure 7A). In contrast, we find that LG186 is highly effective at removing COPI from membranes in MDCK cells and inducing a collapse of the Golgi (Figure 7A). Consistent with a selective inhibition of Arf-GBF1 function, neither BFA nor Exo2 nor LG186 detectably affected the recruitment of COPII to ER exit sites in MDCK cells but, as seen in HeLa cells, did cause enlargement of COPII-labeled ER exit sites and redistribution to a more peripheral distribution (images for LG186 shown in Figure 7B). Furthermore, the TGN localization of golgin-97 is lost on incubation with LG186 and parallels the collapse of the Golgi structure (Figure 7C). The antibody against TGN46 used here does not recognize the protein in MDCK cells, and the TGN46-GFP localization in an equivalent experiment is shown in Figure S5. Canine GBF1, unlike canine BIG1 or BIG2, includes the same additional three amino acids as human GBF1 and thus we predicted the same selectivity of LG186 for GBF1 over BIG1 and BIG2. Consistent with this, we find that LG186 does not induce a tubulation of endosomes in MDCK cells (Figure 7D).

Figure 7.

LG186, a potent and selective inhibitor of canine GBF1. A) LG186 (50 µm) but not BFA (10 µm) or Exo2 (50 µm) results in loss of COPI (green in merge) from membranes and dispersal of the Golgi (GM130, red in merge) in MDCK cells after 1 h of treatment. B) COPII labeling (Sec31A, green in merge) is unaffected by BFA, Exo2 or LG186 and only LG186 induces collapse of the Golgi (58k labeling, red in merge). C) The TGN in MDCK cells (golgin-97, green in merge) is dispersed by LG186 but not BFA or Exo2; this correlates with Golgi collapse (giantin, red in merge). D) As seen with HeLa cells, BFA, but not Exo2 or LG186, induces tubulation of early and recycling endosomes (TfnR-GFP, green in merge). Bars = 10 µm.


Our data describe the development and characterization of LG186, a reversible and apparently selective inhibitor of human and canine GBF1. While such similar inhibitors have been developed through high-content screening approaches (21,24), we describe a rational design strategy exploiting homology modeling and directed chemical synthesis. It is through this approach that we have been able to develop a compound that, in contrast to others described previously, is active against canine as well as human cells. In addition, LG186 does not result in enlarged endosomes. This phenotype is not evident following RNAi suppression of any of the large Arf–GEFs alone or in combination (15,33,37) and therefore we believe this to be an ‘off-target' effect of Exo2 that is not apparent with LG186.

We observed that Exo2 shares chemical features with BFA that it could use to bind to GBF1 in the same binding site. Modeling of GBF1 Sec7 domain, followed by manual docking of Exo2 based on this similarity, provided a rationale for the improvement in its specificity. From this, a new compound LG186 was designed and synthesized. In vitro inhibition assays showed residual (weak) inhibition of the Sec7 domain of the related Arf–GEF BIG1, which supports the prediction that the compounds can inhibit the exchange activity of Sec7 domain directly. Similar residual inhibition of BFA-insensitive Arf–GEFs by BFA was indeed reported in previous studies (31). In cells, this compound affected Golgi functions in a way that is consistent with a specific inhibition of GBF1. Loss of the integrity of the TGN likely results from an inhibition of transport through the Golgi; these findings are consistent with the integrity of the TGN being dependent on ongoing transport from the Golgi. Mutations in the vicinity of the predicted binding site in GBF1 impaired the inhibitory effect, which supports the predicted model of the binding site of these inhibitors. Specifically, the fact that deletion of the tripeptide ‘NAP’ within the Sec7 domain blocks the efficacy of LG186 is highly indicative that LG186 does indeed target this domain of GBF1.

Canine GBF1 is resistant to BFA owing to a single amino acid substitution within the Sec7 domain M832L (19–21). As Exo2 is also ineffective in MDCK cells, it too must bind more weakly if at all to canine GBF1. Steric blocking of the binding site to Exo2 by the M832L change appears unlikely because Exo2 is a looser fit in the binding pocket in this position, compared with BFA. We suggest that a major factor in weakening of binding could be caused by loss of the favorable aryl–sulfur interaction between methionine and the aromatic phenolic moiety of Exo2 (38). However, Leu-aromatic interactions are also favorable and are seen in many structures, notably interactions between proteins and nucleotides. Contacts around the cyclohexenyl ring of Exo2 are not close-packed. As stated above, LG186 is identical to Exo2 apart from a cyclooctenyl ring within the tricyclic structure instead of a cyclohexenyl ring. This bulkier component in LG186 is likely to fit more tightly within this binding pocket, showing more substantial contacts around the cyclooctenyl ring. The close interaction of LG186 with GBF1-Arf resulting from the cyclooctenyl ring could compensate for the loss of CH-π interaction caused by the substitution of M832L in canine GBF1 such that, unlike Exo2, binding of LG186 could occur to canine GBF1. While these homology models suggest a mechanistic explanation, our experimental data show clearly that LG186 is active in canine cells and we propose that this activity is directed against canine GBF1.

Together, these data strongly suggest that (1) LG186 and Exo2 are specific inhibitors of the Sec7 domain of GBF1, and (2) support our predicted model of binding for these compounds to a site that overlaps with that of BFA. As this site also overlaps the binding site of Arf proteins in subsequent stages of the exchange reaction (10,39), two main mechanisms of inhibition can be envisioned. In the simplest one, LG186 acts by competitive inhibition by blocking access of Arf to the active site of the Sec7 domain of GBF1. We do not rule out, however, that the compounds may represent a novel class of interfacial inhibitors that block GBF1 functions by trapping an abortive Arf–GDP/Sec7 complex. Replacing BFA in the Arf–GDP/BFA/Sec7 complexes by LG186 and Exo2 according to our predicted binding model suggests that the volume of the cavity is sufficient to accommodate the compounds without steric clash (not shown). Demonstrating the actual mechanism of inhibition is however a complex endeavor (reviewed in 40), which is beyond the scope of this study. Our data, notably the use of the GBF1 NAP deletion mutant, are most consistent with LG186, like GCA, acting through inhibition of GBF1. It, however, remains possible that LG186 acts through another mechanism, possibly activation of GBF1. Time-lapse imaging experiments show that LG186 causes rapid accumulation of GBF1 on membranes, while our FRAP data show a significantly enhanced immobile fraction on the membrane. This would be consistent with an effect on the rate of dissociation from the membrane.

In summary, we describe here the development of LG186, which acts to induce disassembly of the Golgi in a phenotypically near-identical manner to another recently described selective GBF1 inhibitor GCA (21). It has, to our knowledge, unique advantages over BFA, Exo2 and GCA in that it is as effective in canine cells as it is against human cells. Our characterization of Exo2 and the development of LG186 by directed modification of Exo2, in combination with docking of these compounds into homology models of GBF1, offer potential insight into CH-π packing and how close interactions of small molecule inhibitors within a binding site contribute to the selectivity and efficacy of the compound.

An important point to note is that GCA has some advantages over LG186 in that it is slightly more soluble and exerts its effects more quickly than LG186 does. LG186 is structurally distinct from GCA and therefore provides the opportunity for important confirmation of GBF1 dependence. The most significant advantage of LG186 is that, unlike Exo2 and GCA, LG186 is effective against canine cells in addition to other species opening up the possibility to dissect the function of GBF1 in widely used cell models such as MDCK cells. As MDCK cells are a common model to study cell polarization, epithelial differentiation and the onset and maintenance of multicellularity, as well as the function of the kidney epithelium, LG186 provides a unique tool to define the function of GBF1 in these processes. Together, the data presented here characterize the utility of LG186 as an effective yet reversible inhibitor of GBF1 and provide a framework to study further aspects of Golgi function and intracellular organization in many systems.

Materials and Methods

Synthesis of LG186

Compounds referred below are shown in the schematic in Figure S1D. Compounds were dissolved in DMSO and stored at −20°C.

Compound 1

To a solution of cyclooctanone (10 mmol) in ethanol (10 mL) were added sulfur (320 mg, 10 mmol), ethyl cyanoacetate (1.07 mL, 10 mmol) and morpholine (875 µL, 40 mmol). The reaction mixture was stirred at 60°C for 5 h. Eight hundred and fifty-five milligrams of 1 was obtained (yield: 34%) after purification by chromatography using dichloromethane.

1H NMR (400 MHz, CDCl3)δ: 1.28 (m, 5H, CH3 + CH2), 1.39 (m, 2H, CH2), 1.50 (m, 2H, CH2), 1.56 (m, 2H, CH2), 2.54 (m, 2H, CH2), 2.75 (m, 2H, CH2), 4.21 (q, 2H, J = 7.5 Hz, CH2), 5.84 (brs, 2H, NH2). 13C NMR (100 MHz, CDCl3)δ: 14.5 (CH3), 24.6 (CH2), 25.0 (CH2), 25.9 (CH2), 26.1 (CH2), 29.2 (CH2), 31.5 (CH2), 58.8 (CH2), 105.6 (C), 119.5 (C), 134.3 (C), 160.8 (C), 164.2 (C). ES-MS m/z 254.1 (MH+).

Compound 2

Compound 1 was heated at 150°C in 5 mL formamide for 5 h. Upon cooling overnight, the product crystallized as slightly brownish crystals. The resulting crystals were collected and washed with a mixture of cold ethanol/water (1/1) to give the corresponding thienopyrimidone ring 2 in quantitative yield.

1H NMR (400 MHz, DMSO) δ: 1.27 (m, 2H, CH2), 1.42 (m, 2H, CH2), 1.62 (m, 4H, 2 CH2), 2.87 (m, 2H, CH2), 3.06 (m, 2H, CH2), 8.01 (s, 1H, CH), 12.28 (brs, 1H, NH). 13C NMR (100 MHz, DMSO) δ: 24.4 (CH2), 25.3 (CH2), 25.4 (CH2), 26.0 (CH2), 29.9 (CH2), 31.5 (CH2), 133.7 (C), 135.0 (C), 144.6 (C), 147.8 (C), 150.0 (CH), 157.7(C). ES-MS m/z 235.1 (MH+).

Compound 3

Six hundred and fifty milligrams of 2 was dissolved in hot DMF (dimethylformamide) and ice-cooled prior to the addition of 2 equivalent of POCl3. Upon stirring overnight, the product precipitated out. The white powder was collected and washed with cold water. Further addition of cold water into the mother liquor gave additional precipitate which is used straight away in the next step.

1H NMR (400 MHz, CDCl3)δ: 1.25 (m, 2H, CH2), 1.46 (m, 2H, CH2), 1.70 (m, 4H, 2 CH2), 2.92 (m, 2H, CH2), 3.12 (m, 2H, CH2), 8.67 (s, 1H, CH). 13C NMR (100 MHz, CDCl3)δ: 25.0 (CH2), 25.4 (CH2), 26.3 (CH2), 28.2 (CH2), 30.3 (CH2), 31.6 (CH2), 128.6 (C), 129.6 (C), 142.7 (C), 151.3 (CH), 156.4 (C), 158.7 (C). ES-MS (electrospray mass spectrometry) m/z 252.1 (MH+, 35Cl), 254.1 (MH+, 37Cl).

Compound 4

To a solution of chloride dissolved in methanol was added 10 equivalent of hydrazine monohydrate. The mixture was stirred for 2 h and water was added. The resulting precipitate was filtered off and washed with cold water to afford 255 mg of 4, with 37% yield starting from 2.

1H NMR (400 MHz, CDCl3)δ: 1.27 (m, 2H, CH2), 1.44 (m, 2H, CH2), 1.65 (m, 4H, 2 CH2), 2.50 (brs, 2H, NH2), 2.83 (m, 4H, 2 CH2), 6.54 (brs, 1H, NH), 8.41 (s, 1H, CH). 13C NMR (100 MHz, CDCl3)δ: 25.3 (CH2), 26.0 (CH2), 26.1 (CH2), 27.7 (CH2), 30.1 (CH2), 31.6 (CH2), 115.7 (C), 127.7 (C), 137.3 (C), 152.3 (CH), 158.7 (C), 164.8 (C). ES-MS m/z 249.0 (MH+), 271.0 (MNa+).


To a solution of 250 mg of 4 in methanol was added 1.2 equivalent of vanillin. The mixture was stirred for 2 h, diluted with water and extracted with dichloromethane. The organic layer was dried with MgSO4, filtered off and concentrated in vacuo. Crystallization from diethyl ether gave 100 mg of yellow crystals (yield: 26%).

1H NMR (400 MHz, DMSO) δ: 1.27 (m, 2H, CH2), 1.46 (m, 2H, CH2), 1.62 (m, 2H, CH2), 1.68 (m, 2H, CH2), 2.85 (m, 2H, CH2), 3.19 (m, 2H, CH2), 3.88 (s, 3H, CH3), 6.84 (d, 1H, J = 10. 7 Hz, CH), 7.60 (d, 1H, J = 10.7 Hz, CH), 7.79 (s, 1H, CH), 8.30 (s, 1H, CH), 9.45 (brs, 1H, OH) 11.70 (brs, 1H, NH). 13C NMR (100 MHz, DMSO) δ: 24.8 (CH2), 25.8 (CH2), 26.6 (CH2), 29.7 (CH2), 31.7 (CH2), 51.8 (CH3), 111.0 (CH), 115.4 (CH), 118.8 (C), 122.2 (CH), 126.8 (C), 126.9 (C), 133.5 (C), 135.1 (C), 143.6 (CH), 144.5 (C), 147.8 (C), 148.7 (C), 153.4 (CH). ES-MS m/z383.1 (MH+). HRMS (high resolution mass spectrometry) 393.1536, found 383.1530. Anal. (C20H22N4O2S.0.2MeOH) C, H, N.

Homology models

Homology models of the Sec7 domains of the three human GEFs of interest [GBF1 (GenBank accession number NP_004184), BIG1 (NP_006412) and BIG2 (NP_006411)] were constructed based on the Arf1-GDP/BFA/ARNO [PDB entry 1RQ8, (10)] and Arf1-GDP/BFA/Gea1p [PDB entry 1RE0, (26)] complexes. Pairwise alignment of GBF1 and Gea1p gives a sequence identity of 43% and alignment of BIG1, BIG2 and ARNO gives a sequence identity of 48%. Hence, GBF1 was built on the template 1RE0, and BIG1 and BIG2 were built on the template 1RQ8.

For each model, the template structure was altered to the target sequence, loop insertions or deletions (in-dels) were rebuilt as follows. The long loop insertion in the GEA1 sequence between helices 7 and 8 (which is not resolved in the crystal structure) was rebuilt using the corresponding loop from ARNO and altered to the GBF1 sequence. Other loop in-dels were constructed using the loop-building function in InsightII. The only in-dels in the BIG1 and BIG2 models is a two-residue insertion in this region (with respect to ARNO). Of particular interest is the three-residue insertion in GBF1, with respect to the other GEFs (guanine nucleotide exchange factors), between helices 8 and 9, which is located close to both ARF1- and the BFA-binding site. This insertion was modeled as an extra turn in helix 8 and a different turn conformation. The canine GBF1 has three residue changes to the GEF domain with respect to human GBF1. Two are surface residues remote from the ARF binding interface, the third is the change M832L which prevents BFA binding. The leucine side chain is required to adopt the χ1 t, χ2 g-conformation because of steric constraints imposed by the GEF structure.

The conformations of residue side chains in the complete models were adjusted by inspection to remove bad clashes. Hydrogen atoms were added consistent with pH 7 and the model complexes soaked with a 10 Å layer of water molecules. The models were relaxed by 2000 steps of conjugate-gradient energy minimization, constraining the backbones to their original positions with harmonic restraint potential. This potential was reduced from 1000 to 0.5 kcal/Å during the minimization. The geometric quality of the models was examined using PROCHECK and found to be of similar quality to the original templates in each case.

Ligand complexes

The initial models of GBF1, BIG1 and BIG2 included BFA and Arf1 which are already present in the template structures. Ligand Exo2 was docked into the BFA site of the GBF1 complex by superimposing the Exo2 phenol group over the hydroxyl group of BFA. This maintains a hydrogen bond between the ligand and side chain of Tyr 828. Rotatable torsions in Exo2 were manipulated to allow the rest of the ligand to occupy the rest of the BFA site and causing the rest of Exo2 to project toward the extra turn in helix 8. The complex with LG186 was built in the same fashion, likewise the corresponding complexes with canine GBF1. These complexes were energy minimized using the protocol above. Models of Exo2 and LG186 with BIG1 and BIG2 were built by superimposing the protein backbones onto the corresponding minimized GBF1 complexes and transferring the ligand to the BIG structures. These non-minimized models illustrate the overlap of Exo2 and LG186 ligands with the shorter helix 8–helix 9 turn of BIG1 and BIG2.

Activated Arf pull-down

Analysis of Arf activation was done essentially as described before (41). HeLa cells were grown in 10-cm dishes and treated with the indicated chemical (at 50 µm or DMSO only) in culture medium for 1 h. After two quick washes in ice-cold PBS, cells were scraped in Buffer [200 mm NaCl; 50 mm Tris pH 7.4; 10 mm MgCl2; 1% Triton X-100; 0.1% SDS; 0.5% sodium deoxycholate; 5% glycerol; Proteases inhibitors Cocktails V (Calbiochem), pH 7.4] supplemented with the appropriate chemical. Insoluble material was pelleted by centrifugation for 10 min with 10 000 ×g at 4°C, and supernatant (about 1 mg of protein) was incubated with 50 µg of purified GST or GST-GGA2-GAT prebound on Gluthatione Sepharose for 30 min at 4°C on a rotating platform. Beads were then washed thoroughly and bound proteins were eluted in SDS–PAGE sample buffer. Samples were immunoblotted with the anti-Arf antibody 1D9 (AbCam) which recognizes all human Arf isoforms (29). The plasmid encoding for GST-GGA2-GAT (residues 157–331 of GGA2) was kindly provided by Jennifer Hirst and has been described previously (42). For Figure 2E, gels were probed using an IRDye-680CW-conjugated anti-mouse antibody (Licor) and blots scanned and quantified using a Licor Odyssey infrared imaging system (Licor).

In vitro analysis of the inhibition of Arf–GEF function

We measured the inhibitory activity of the four compounds (BFA, Exo2, GCA and LG186) toward several human Arf–GEF constructs representing three Arf–GEF families: ARNOSec7, BIG1Sec7 and BIG1DCB-HUS-Sec7, Brag2Sec7-PH, all of which were purified to high homogeneity and are efficient Arf–GEFs toward Δ17Arf1 in exchange assays in solution [(31,43); Brag2Sec7-PH to be published elsewhere]. The concentration of the inhibitors has been chosen as the highest LG186 concentration (10 µm) with no apparent aggregation. LG186 aggregation was measured at 384 nm after 1-h incubation in buffer WB (50 mm Tris pH 8, 50 mm NaCl, 2 mm MgCl2, 2 mmβ-mercaptoethanol) containing 0.2–0.4% DMSO before and after centrifugation (15 min, 20 000 ×g). The exchange activity without and with the inhibitors has been measured by tryptophan fluorescence (λEx = 292 nm, λEm = 340 nm), which monitors the conformational changes of Arf1 as it is converted from the GDP to the GTP-bound conformation. kobs (second−1) were determined from single-exponential fit of the fluorescence change. Exchange reactions were performed at 30°C, with highly purified protein in WB using a Flexstation (Molecular Devices) equipped with an eight-channel pipettor. Reaction component concentrations: Δ17Arf11 µm (soluble N-terminal truncated form), GEF 0.1 µm, compound 10 µm, DMSO 0.5%; incubation for 30 min at room temperature then for 5 min at 30°C. Reactions were started with 100 µm GTP.

Immunofluorescence and live-cell imaging

For immunofluorescence cells grown on glass coverslips were fixed either with methanol for 4 min at −20°C or with 3.5% paraformaldehyde for 15 min followed by permeabilization with 0.1% Triton X-100 in PBS for 5 min, blocked with PBS containing 3% BSA and probed with primary antibodies, as indicated and detailed below, and Alexa-Fluor™ conjugated secondary antibodies (Invitrogen) or Cy-dye conjugated secondary antibodies (Jackson Immunoresearch). The antibodies used are as follows: polyclonal rabbit anti-COPII (Sec31A) was as previously described (44); polyclonal anti-COPI (BSTR) (45); rabbit polyclonal anti-giantin (Covance); monoclonal mouse anti-GM130 (BD Transduction Laboratories); polyclonal sheep anti-TGN46 (AbD Serotec); mouse monoclonal anti-Golgin-97 (CDF4) (Invitrogen); mouse monoclonal anti-58K Golgi protein antibody (58K-9) (AbCam); Nuclei were counterstained using DAPI (4',6-diamidino-2-phenylindole) (Invitrogen) and coverslips were mounted in Mowiol.

A plasmid encoding for TGN46-GFP was kindly provided by Vas Ponnambalam (University of Leeds, UK). The transferrin receptor in fusion with GFP (TfnR-GFP) was kindly provided by Gary Banker (Oregon Health and Science University, Oregon, USA). The TfnR-GFP cassette was excised from a pJPA5-TfnR-GFP vector using EcoRI and XbaI restriction sites and cloned into a pLVX-Puro (Clontech) using the same sites. The resulting vector (pLVX-Puro-TfnR-GFP) was verified by sequencing and used in transient overexpression.

For FRAP experiments, HeLa cells were grown on live-cell dishes and a plasmid encoding for GFP-GBF1 [kindly provided by Elizabeth Sztul (28)] transfected using Fugene6 or Lipofectamine2000 according to the manufacturer's instructions. Twenty-four to Forty-eight hours after transfection, cells were treated with chemicals in complete culture medium for 1 h. The culture medium was then supplemented by 30 mm HEPES pH 7.4 and the cells were live imaged in a 37°C heated chamber using a Perkin Elmer spinning disk confocal microscope (Ultraview ERS) with Photokinesis add-on controlled by Volocity software. Images were recorded at a rate of 1 frame per second, cells were subjected to 10 pre-bleach frames, and the recovery after bleaching was followed for 1 min. Data are presented as mean ± SD from 5 to 21 different regions of interest, from at least 3 independent transfections.

Molecular cloning and site-directed mutagenesis

To obtain the full-length human BIG1 in fusion with eGFP, a pCMV-HA-BIG1 construct (described in 46) was used as a template for PCR. A two-step procedure was used to amplify an N-terminal and a C-terminal half of BIG1 with the following primers: plus strand 5′-GGGGCTCGAGCATGTATGAGGGGAAG-3′ (XhoI, restriction sites are underlined and bold nucleotides indicates mutations introducing a single SalI restriction site) and minus 5′-CCTGAAAAGTCATGTTGGTCGACATATGCATACATGAC-3′ (SalI) for BIG1-N half; and plus strand 5′-GTCATGTATGCATATGTCGACCAACATGACTTT TCAGG-3′ (SalI) and minus 5′-GGGGGGATCCTCATTGCTTGTTTATT CCAAG-3′ (BamHI) for BIG1-C half. PCR products were cloned into pGEM-T (Promega) and BIG1-N half was cloned into pEGFP-C2 (Clontech) using XhoI and SalI sites. Then the BIG1-C half was subsequently cloned into the obtained vector using SalI and BamHI sites.

The obtained construct yields to a full-length BIG1 in fusion with an N-terminal GFP and was fully sequenced to ensure the success of the cloning. The obtained sequence corresponds to the deposited GenBank Accession number NM_006421 (BIG1).

To obtain the mutant GBF1 deleted for the NAP extra loop (residues 846–848 according to human GBF1, accession number: NP_004184), a GFP-tagged version kindly provided by E. Sztul (28) was used as a template for site-directed mutagenesis using the following sense primer: 5′-CAATGTTCGTAAACAGATGACCCT(G/C)GAGGAGTTTC GCAAAAATCTG-3′ in combination with its antisense primer. Nucleotide in brackets indicates the base substitution G2553C [according to the human open reading frame (ORF) for GBF1] producing a XhoI restriction site for simple screening of the colonies. Position of the deletion is indicated by a minus sign in brackets. The resulting construct was fully sequenced to ensure the fidelity of the PCR.


We would like to thank the past and present members of the Stephens lab, and to Pete Cullen, Jon Lane and Harry Mellor, for the input on this project and helpful discussions as well as critical reading of the manuscript. We are grateful to Atsushi Miyawki for providing the NPY-Venus construct, Martha Vaughan for the pCMV-HA-BIG1 construct used as a template to generate GFP-BIG1, Gary Banker for transferrin-receptor-GFP, Vas Ponnambalam for TGN46-GFP, Jon Hanley and Dan Rocca for providing some additional recombinant protein, and Elizabeth Sztul for GFP-GBF1 and helpful discussions. This work is funded by the BBSRC through research grants under the ‘Selective Chemical Inhibition of Biological Systems' scheme and the MRC through a non-clinical senior fellowship to D. J. S. J. C. and M. Z. are supported by grants from the Assocation pour la Recherche contre le Cancer and the ANR Physique-Chimie du Vivant. Author contributions: F. B. designed, performed and analyzed experiments and assisted in the writing of the manuscript. G. J. C. and L. G. designed and synthesized Exo2 and LG186. R.B.S. performed all homology modeling. M. Z. performed and analyzed in vitro GEF assays. L. M. R., R. A. S. R. B. S., J. C. and J. M. L. assisted in the development of the project through experimental design and analysis and co-wrote the manuscript. D. J. S. designed, performed and analyzed experiments and wrote the manuscript.