The Arf GEF GBF1 and Arf4 synergize with the sensory receptor cargo, rhodopsin, to regulate ciliary membrane trafficking Free

The Arf family of small G-proteins constitutes a crucial component of the intracellular membrane trafficking machinery. Through the control of lipid metabolism and the recruitment of canonical coat complexes and protein adaptors that recognize and sequester the appropriate membrane cargo, Arf GTPases play a central role in key processes such as the maintenance of the Golgi architecture, progression of cargo through the Golgi complex, as well as the Golgi-to-plasma-membrane targeting that is responsible for the delivery of sensory receptors and their associated complexes to primary cilia (Deretic, 2013; Donaldson and Jackson, 2011; Ezratty et al., 2016; Hilgendorf et al., 2016; Humbert et al., 2012; Schou et al., 2015; Schwarz et al., 2012; Wang and Deretic, 2014; Wright et al., 2011). The distal ciliary membrane of vertebrate retinal rod photoreceptor cells elaborates a unique sensory organelle, the rod outer segment (ROS), which is filled with several thousand membranous disks containing as many as a billion copies of the light receptor rhodopsin (Besharse, 1986). Rhodopsin is directed to cilia through the ciliary targeting signal (CTS) VxPx, which directly binds activated Arf4 at the Golgi/TGN (Deretic et al., 2005; Mazelova et al., 2009; Wang et al., 2012). The importance of this trafficking pathway is underscored by autosomal dominant retinitis pigmentosa (ADRP), a group of blinding diseases that result from mutations in >25 genes. Mutations affecting the rhodopsin CTS VxPx are among the most severe forms of ADRP (Berson et al., 2002). On the other hand, different targeting signals and trafficking mechanisms direct other ROS membrane components to cilia. Cyclic nucleotide-gated (CNG) channel transport relies on the cytoskeletal adaptor ankyrin-G (Kizhatil et al., 2009). Guanylyl cyclase 1 (GC1) and the progressive rod-cone degeneration (PRCD) protein appear to require rhodopsin for their ciliary trafficking (Pearring et al., 2015; Spencer et al., 2016), whereas targeting of the ROS disk rim protein peripherin-2/rds (P/rds, also known as Prph2) requires its C terminus and interactions with SNARE proteins (Salinas et al., 2013; Tam et al., 2004; Zulliger et al., 2015). Defects in P/rds cause retinitis pigmentosa and macular dystrophies (Goldberg et al., 2016), while defects in trafficking of prenylated ROS proteins cause retinitis pigmentosa 2 (Zhang et al., 2014).

Broad dysfunction of ciliary trafficking causes human genetic diseases and syndromic disorders collectively known as ciliopathies (Reiter and Leroux, 2017). The transition zone and basal body multiprotein complexes NPHP-JBTS-MKS and BBS participate in ciliary morphogenesis and gating. These processes are affected by mutations causing nephronophthisis, as well as Joubert, Meckel and Bardet Biedel syndrome, which affect multiple organs, including the eyes (Craige et al., 2010; Datta et al., 2015; Garcia-Gonzalo et al., 2011; Nachury et al., 2010; Sang et al., 2011; Shimada et al., 2017; van Reeuwijk et al., 2011). Intraflagellar transport regulates the entrance and exit of regulatory components and the progression of ciliary cargo, including rhodopsin, through the transition zone (Bhowmick et al., 2009; Eguether et al., 2014; Keady et al., 2011; Krock et al., 2009; Liew et al., 2014; Zhao and Malicki, 2011). These ciliary networks are directly linked to the small GTPase Rab8 and its GEF Rabin8 (also known as Rab3ip) (Bachmann-Gagescu et al., 2011; Chiba et al., 2013; Nachury et al., 2007; Omori et al., 2008), which are crucial regulators of ciliary membrane trafficking (Deretic et al., 1995; Feng et al., 2012; Moritz et al., 2001; Wang and Deretic, 2015; Westlake et al., 2011). Dysfunction of Arf GTPases and their regulators is also a known cause of ciliopathies (Seixas et al., 2013; Wiens et al., 2010; Zhang et al., 2013).

Arf GTPases exert their regulatory function through the cycles of GTP binding and hydrolysis that are regulated by Arf guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs), which control their membrane association and signaling pathways through activation cascades and positive-feedback loops (Bui et al., 2009; Casanova, 2007; Jackson and Casanova, 2000; Lowery et al., 2013; Stalder and Antonny, 2013). One of the outstanding questions in the regulation of Arf GTPases is the role of protein cargo in their activation. It has been proposed that the cargo acts upstream of Arf activation, in a manner analogous to the activation of heterotrimeric G-proteins by G-protein-coupled receptors (GPCRs) that serve as their GEFs, upon light or ligand stimulation (Caster et al., 2013). Although multiple Arf GEFs activate Arfs in spatiotemporally restricted manners, it is not clear what signals Arf GEFs recognize in order to activate Arfs. The specific cargo has the capacity to regulate the Arf-dependent recruitment of the protein adaptors (Caster et al., 2013), which suggests that a functional complex between the cargo, the cognate Arf and an Arf GEF likely exists during membrane trafficking. However, currently the evidence for such a complex is absent.

The BIG/GBF family of Golgi-localized large Arf GEFs contains the highly conserved Sec7 domain involved in the nucleotide-exchange activity, surrounded by several conserved domains involved in functional interactions that regulate their activity and membrane association (Bui et al., 2009; Casanova, 2007; Wright et al., 2014). Golgi-localized large Arf GEFs are autoinhibited in solution. Their catalytic activity and membrane association are controlled by cooperative allosteric regulation via coincidence detection by dimerization and cyclophillin-binding (DCB) and homology downstream of Sec7 (HDS) domains, which integrate direct inputs from membranes and multiple activated Arfs and Rabs (Alvarez et al., 2003; Bouvet et al., 2013; McDonold and Fromme, 2014; Monetta et al., 2007; Nawrotek et al., 2016; Richardson and Fromme, 2012; Richardson et al., 2012; Stalder and Antonny, 2013). GBF1 and Arf4 function within the early Golgi, and at the TGN (Ben-Tekaya et al., 2010; Chun et al., 2008; Garcia-Mata et al., 2003; Kawamoto et al., 2002; Mazelova et al., 2009; Nakai et al., 2013; Szul et al., 2005, 2007; Wang et al., 2012; Zhao et al., 2006). At the TGN, GBF1 initiates an Arf activation cascade through direct interactions of Arf4 with the DCB domains of BIG1 (also known as Arfgef1) and BIG2 (also known as Arfgef2) (Lowery et al., 2013). It is thus plausible to hypothesize that GBF1 might function as the Arf4 GEF that activates Arf4 in ciliary receptor targeting.

Although the directed cargo delivery is tightly regulated in all cells, the limited quantity of a specific ciliary cargo often necessitates its overexpression to analyze ciliary transport, thus retinal rod photoreceptors provide a clear advantage for these studies (Pearring et al., 2013; Wang and Deretic, 2014; Wensel et al., 2016). Because of their extensive ROS membrane turnover, amphibian rods have consistently offered a unique model in which biochemical and morphological data can be correlated in a single experimental system for the study of otherwise basic mechanisms underlying ciliary and photoreceptor membrane biogenesis (Besharse, 1986; Hall et al., 1969; Papermaster et al., 1975, 1985; Young, 1967, 1976). Although photoreceptors are highly specialized cells, a recombinant rhodopsin–GFP fusion protein expressed in epithelial cells maintains the restricted ciliary localization, indicating that certain aspects of ciliary transport are highly conserved (Mazelova et al., 2009; Trivedi et al., 2012; Wang et al., 2012; Ward et al., 2011). In this study, we take advantage of the photoreceptor paradigm because of the abundance of rhodopsin transport carriers (RTCs) that mediate Golgi-to-cilia transport in photoreceptors (Deretic and Mazelova, 2009; Wang and Deretic, 2014), and examine the role of the ciliary cargo, rhodopsin, in the recruitment of the Arf GEF GBF1 and the activation of Arf4. We find that GBF1 interacts with rhodopsin at the Golgi/TGN, and that the activity of GBF1 is essential for their interaction and for communication with Arf4, which, in turn, regulates the formation of RTCs and the delivery of rhodopsin to sensory cilia.

To determine which Arf GEF is responsible for the activation of Arf4 in retinal photoreceptor cells, we examined the distribution of GBF1, a candidate Arf GEF that is reported to activate Arf4 in HeLa cells (Lowery et al., 2013). By confocal microscopy, GBF1 exhibited a distribution that differed from that of the cis-Golgi marker GM130 (UniProt Q08379) and closely resembled that of the trans-Golgi marker Rab6 (also known as Rab6a) (Fig. 1A–D, arrows). The pixel colocalization of Rab6 with GBF1 was significantly higher than with GM130 (P=3.42×10⁻⁵, n=5 cells) (Fig. 1D). In photoreceptors, Golgi is localized within the myoid region (M) of the rod inner segment (RIS), as schematically presented in Fig. 1E. To pinpoint the localization of GBF1 within the Golgi, we performed in situ proximity ligation assay (PLA), a molecular technique suitable for proteomic analysis, because a positive signal is possible only when the fluorescent PLA probes are <40 nm apart (Raykova et al., 2016; Söderberg et al., 2006). We employed PLA modified for studies of brain and retinal tissue (Blasic et al., 2012; Trifilieff et al., 2011; Wang and Deretic, 2015; Wang et al., 2012; Zulliger et al., 2015). GBF1-Rab6 interaction sites (Fig. 1F, red dots), aligned well with the trans-Golgi, which was identified post-PLA by staining with anti-Rab6 conjugated to Alexa Fluor 488 (Fig. 1F, green). No interaction sites were detected between GBF1 and the cis-Golgi markers GM130 (Fig. 1G) and p115 (UniProt O60763) (Fig. 1H), despite the robust Golgi labeling with the antibody to p115 (Fig. 1I). Thus, in photoreceptor cells, GBF1 does not associate with the cis-Golgi but is specifically localized at the trans-Golgi.

Next, we determined that, in the RIS, GBF1 and Arf4 interact in close proximity to the trans-Golgi (Fig. 1J, red dots), identified by Rab6, as above (Fig. 1J, green). Notably, within the same area, GBF1 also interacted with rhodopsin (Fig. 1K). The distribution of these interaction sites was comparable to the distribution of rhodopsin-Arf4 interaction sites (Fig. 1L), as noted previously (Wang et al., 2012). Unexpectedly, GBF1 also interacted with the Arf GAP ASAP1, which is known to form a complex with rhodopsin and Arf4 (Wang et al., 2012). As shown in Fig. 1M, GBF1-ASAP1 interactions were not restricted to the Golgi area, but were distributed throughout the RIS. To quantify the number of protein-protein interaction sites in these experiments, the red fluorescent signals detected by PLA in the Golgi area of the myoid region were assigned to the Golgi/TGN and those in the ellipsoid region (E) to RTCs, as described (Wang et al., 2012). The quantitative analysis revealed that interactions of rhodopsin with GBF1 and Arf4 occur nearly exclusively at the Golgi/TGN, whereas GBF1-ASAP1 interactions occur at the Golgi/TGN and on RTCs (Fig. 1N).

To further characterize the subcellular localization of GBF1, we performed retinal subcellular fractionation by a standard procedure (Deretic and Mazelova, 2009), generating ROS and retinal postnuclear supernatant (PNS), highly enriched in photoreceptor biosynthetic membranes (Deretic and Papermaster, 1991; Papermaster et al., 1975). PNS was separated into three fractions designated as the Golgi/TGN/ER-enriched, RTC-enriched (T/G/E and RTCs hereafter) and cytosol, as previously reported (Deretic, 2000; Deretic et al., 1995, 1996; Mazelova et al., 2009; Morel et al., 2000; Wang et al., 2012). In agreement with microscopy data, GBF1 was present in the T/G/E fraction, on RTCs and in the cytosol (Fig. 1O), paralleling the known distribution of the Arf GAP ASAP1 (Mazelova et al., 2009) (Fig. 1O). Arf4 was detected only in the T/G/E fraction and in the cytosol, as previously determined (Mazelova et al., 2009; Wang et al., 2012) (Fig. 1O). Subcellular fractionation corroborated the PLA data and revealed that despite the lack of Arfs (Mazelova et al., 2009; Wang et al., 2012), both the Arf GEF GBF1 and the Arf GAP ASAP1 also associate with RTCs.

To determine whether the activity of GBF1 affects its interactions with the ciliary cargo, Arf4 and ASAP1, we inactivated GBF1 in cultured eyecups with Golgicide A (GCA) for 3 h. GCA is a selective inhibitor of GBF1 that has no effect on other Arf GEFs, owing to the unique conformation of the nucleotide-binding pocket of GBF1 (Sáenz et al., 2009). In control retinas, GBF1 interactions were detected around the Golgi/TGN (Fig. 2A,C), as before. Rhodopsin-Arf4 interactions were detected at the Golgi (Fig. 2B, arrows). ASAP1 interactions with rhodopsin and GBF1 were detected at the Golgi and on RTCs (Fig. 2D,E, arrows). GCA treatment greatly diminished rhodopsin-GBF1, rhodopsin-Arf4, Arf4-GBF1 and rhodopsin-ASAP1 interactions (Fig. 2F–I), but had minimal effect on GBF1-ASAP1 interactions (Fig. 2J). In GCA-treated retinas, rhodopsin-ASAP1 interactions were diminished both at the Golgi and on RTCs, and the remaining interaction sites were observed along the myoid-ellipsoid border (Fig. 2I, arrows), at an unusual location not seen in the controls. These data suggest that GCA affected both the formation of nascent RTCs at the Golgi/TGN, and the cilia-directed trafficking of RTCs formed before the addition of GCA. The activity of GBF1 was essential for rhodopsin-GBF1-Arf4-ASAP1 communications, as GCA caused a significant decrease in their interaction sites (P<2.2×10⁻⁶) (Fig. 2K–N). By contrast, GBF1-ASAP1 interactions were unaffected (Fig. 2O). From the comparable number and distribution of rhodopsin interaction sites with GBF1 and Arf4 detected by PLA at the Golgi/TGN, we conclude that the specific complex between the cargo, the cognate Arf and the Arf GEF may be formed there.

Given the reported disassembly of the Golgi by GCA (Lowery et al., 2013; Sáenz et al., 2009), we examined the state of the Golgi complex in GCA-treated retinas. Surprisingly, in photoreceptor cells, GCA had minimal effect on the Golgi morphology. Both cis-Golgi, identified by GM130 staining, and trans-Golgi, identified by Rab6, were largely unchanged in GCA-treated retinas (Fig. 3A). By contrast, Brefeldin A (BFA), a noncompetitive inhibitor of Golgi Arf GEFs (Peyroche et al., 1999), caused substantial swelling and perturbation of the photoreceptor Golgi, as previously described (Deretic and Papermaster, 1991; Mazelova et al., 2009) (Fig. 3A). Thus, in all probability, the Golgi organization in photoreceptors is not controlled by GBF1, but by another BFA-sensitive Arf GEF, very likely to be BIG1 (Boal and Stephens, 2010), which is detected in the mouse retinal transcriptome at a similar level to GBF1 (Brooks et al., 2011).

Because the activity of GBF1 is important for its interactions with Arf4 and the ciliary cargo at the Golgi, we asked whether it is also necessary for ciliary trafficking. We performed pulse-chase experiments in isolated retinas using established methodology: following a 1 h pulse and a 2 h chase, photoreceptor ER membranes are cleared of newly synthesized proteins and radiolabeled rhodopsin localizes in the Golgi/TGN, RTCs and the ROS, with the kinetics paralleling its trafficking in vivo (Deretic and Papermaster, 1991; Mazelova et al., 2009). We followed the progression of radiolabeled proteins through the biosynthetic membranes separated on sucrose density gradients in control retinas, or in the continuous presence of GCA. At the GCA concentration tested, the majority of the newly synthesized rhodopsin was arrested in the T/G/E fraction and its delivery to the ROS was significantly inhibited (P<0.01) (Fig. 3B–D). Based on its uniform molecular weight in all fractions, we determined that, in the presence of GCA, newly synthesized rhodopsin had left the ER and reached the Golgi, where it was terminally glycosylated. This is in contrast to BFA-induced trafficking disruption, which also affects oligosaccharide trimming (Deretic and Papermaster, 1991). Furthermore, GCA did not alter the Golgi localization of GBF1 (Fig. 3E, arrows), in line with the report that in GCA-treated cells, GBF1 remains on the Golgi membranes (Lowery et al., 2013). Consistent with the preservation of interactions between GBF1 and ASAP1 detected in Fig. 2, their colocalization was unaltered by GCA (Fig. 3E, arrows). By contrast, colocalization between Arf4 and GBF1 was disrupted by GCA treatment (Fig. 3E, arrows), in accord with the absence of a PLA signal observed in Fig. 2. Finally, GCA treatment had minimal effect on the distribution of GBF1, Arf4 and ASAP1 among retinal subcellular fractions (Fig. 3F,G). Because GCA significantly slowed down the exit of rhodopsin while maintaining the Golgi structure and localization of key associated proteins, we conclude that in retinal photoreceptors the activity of GBF1 is necessary for the Golgi export of ciliary cargo.

In addition to the catalytic Sec7 domain, GBF1 contains a DCB domain, a homology upstream of Sec7 (HUS) domain and three HDS domains (Bui et al., 2009; Mouratou et al., 2005). To determine whether the binding of GBF1 to rhodopsin is direct, we employed human DCB-HUS [amino acids (AA) 1–710] and Sec7-HDS1 (AA 695–1066), fused to glutathione S-transferase (GST) (Bouvet et al., 2013) (Fig. 4A). GST fusion proteins were incubated with purified bovine rhodopsin, with or without recombinant human Arf4 preloaded with GTPγS or GDPβS. GST DCB-HUS pulled down rhodopsin significantly better than GST Sec7-HDS1 or GST alone (P<0.005), both in the presence and absence of Arf4, demonstrating a direct cargo-GBF1 interaction (Fig. 4B). GST DCB-HUS pulled down Arf4, bound to GTPγS or GDPβS, whereas Sec7-HDS1 preferably interacted with GDPβS-bound Arf4. To ascertain that GBF1 DCB-HUS is properly folded and binds rhodopsin specifically, we used Arl1 as a negative control, as its binding to the DCB domain is conserved in BIG1 and BIG2, but not in GBF1 (Christis and Munro, 2012; Galindo et al., 2016). We incubated purified bovine rhodopsin, or Arl1Q71L, with increasing amounts of GBF1 DCB-HUS bound to glutathione beads. Rhodopsin binding robustly increased with the increase of DCB-HUS beads, in contrast to the barely detectable increase in nonspecific Arl1 binding (Fig. 4C). To examine the specificity of GBF1 interaction with Arf GTPases, we compared the full-length Arf4 to Δ17Arf1, a truncated construct in which the N-terminal α-helix of Arf1 was removed to facilitate nucleotide loading in the absence of membranes (Randazzo et al., 1995). Unlike Arf1, Arf4 does not require myristoylation and membranes for activation (Chun et al., 2008; Duijsings et al., 2009); therefore, full-length Arf4 was preloaded with GTPγS or GDPβS. GBF1 is known to co-precipitate and activate both Arf1 and Arf4 in vivo (Szul et al., 2007). However, GST DCB-HUS pulled down only Arf4 (Fig. 4D), whereas Δ17Arf1 did not show above background binding (Fig. 4D). Although these data point to the specificity of Arf4 binding to the DCB-HUS domain of GBF1, they could be attributed to the absence of the N-terminal helix of Arf1, which, although different from Arf4 (Duijsings et al., 2009), might still have the ability to interact with the DCB-HUS of GBF1. Lack of strong discrimination between the nucleotide-bound states of Arf4 by GST DCB-HUS indicates that the contact surface on Arf4 does not undergo conformational changes upon nucleotide binding.

A key step in the assembly of the ciliary-targeting complex is the binding of rhodopsin to activated Arf4 at the TGN (Mazelova et al., 2009). We thus wanted to test whether the influx of rhodopsin plays a role in the activation of Arf4. For this purpose, we treated retinas with cycloheximide, which essentially abolished rhodopsin-GBF1, rhodopsin-Arf4 and Arf4-GBF1 interactions (Fig. 5A–C,E–G), but, like GCA, had minimal effect on GBF1-ASAP1 interactions (Fig. 5D,H). A significant decrease in interaction sites detected by PLA in the RIS (P<2.0×10⁻¹³) indicates that not only rhodopsin interactions, but also the Arf4-GBF1 interaction, were contingent upon the presence of ciliary-targeted cargo in biosynthetic membranes (Fig. 5I). A parallel pulse-chase experiment confirmed a near complete inhibition of protein synthesis by a 3 h treatment with cycloheximide (Fig. 5J,L). Subcellular fractionation showed that cycloheximide minimally affected the intracellular distribution of Arf4 and ASAP1 (Fig. 5K). Cycloheximide did not alter the localization of GBF1 and Arf4 (Fig. 5M), but completely depleted rhodopsin from the RIS endomembrane system (Fig. 5N,O). Although it is formally possible that the depletion of proteins other than rhodopsin might have contributed to the observed effects of cycloheximide on Arf4-GBF1 interactions, this is highly unlikely considering that rhodopsin represents as much as 90% of the ROS membrane protein, has by far the fastest turnover in the retina, and is the major protein synthesized and transported to the cilia of photoreceptor cells (Brooks et al., 2011; Deretic and Papermaster, 1991; Hall et al., 1969; Papermaster et al., 1975, 1985; Papermaster and Dreyer, 1974). The dominance of rhodopsin is also evident in Fig. 5L, where only a couple of other radiolabeled proteins are clearly detected in the autoradiogram. They most likely correspond to the subunits of the next most abundant photoreceptor protein, the heterotrimeric G-protein transducin, which is activated by rhodopsin in the ROS upon light stimulation. Fig. 5P schematically represents molecular interactions between the cargo, rhodopsin, the cognate Arf, Arf4, and the Arf GEF GBF1 during the formation of nascent RTCs from the TGN, consistent with the results of our study.

In this study, we provide evidence for the role of protein cargo in the regulation of Arf GTPases in vertebrate rod photoreceptors by establishing the existence of a functional complex between the ciliary cargo, rhodopsin, the cognate Arf, Arf4, and the Arf GEF GBF1. Although ciliary transport is highly conserved, our present study reveals a particular adaptation of photoreceptor cells that are synthesizing and transporting considerable amounts of rhodopsin-containing membranes, through apparent concentration of the Arf GEF GBF1 at the Golgi exit where it senses emergence of the cargo essential for ciliary biogenesis. Our study also broadly implicates the protein cargo in promoting its progression through the endomembrane system by providing input to a specific Arf GEF to activate a cognate Arf directing cargo transport to its correct subcellular location.

The Arf GEF GBF1 activates Arf4, which was initially identified as an essential factor for the generation of ciliary-targeted post-Golgi carriers (RTCs) via interaction with the VxPx CTS of rhodopsin (Deretic et al., 2005). Recent in vivo studies of trafficking of rhodopsin fused to the photoconvertible fluorescent protein Dendra2 in Xenopus photoreceptors showed that the VxPx motif enhances ciliary targeting ≥10-fold and accelerates trafficking of post-Golgi vesicular structures (Lodowski et al., 2013), most likely acting through Arf4. In further support of Arf4 function in ciliary trafficking, the reduction in Arf4 also caused a delay in delivery of ciliary sensory receptor fibrocystin from the Golgi to the cilium (Follit et al., 2014). In photoreceptors, the Arf4-based complex forms in sequential order at the TGN and includes the Arf GAP ASAP1, the Rab8 GEF Rabin8, Rab11 (also known as Rab11a), and the Arf/Rab11 effector FIP3 (also known as Rab11fip3) (Mazelova et al., 2009; Wang and Deretic, 2015; Wang et al., 2012). The notion that membrane-targeting modules assemble through multiple weak interactions that create high-avidity complexes was recently reinforced by crystallization and analysis of the Rab11–FIP3–Rabin8 dual effector complex (Vetter et al., 2015).

A partially redundant role for Arf1 and Arf4 in intracellular trafficking was proposed based on double knockouts, which caused tubulation and vesiculation of the Golgi and defects in HeLa cells (Volpicelli-Daley et al., 2005). However, a recent study revealed not only similarities but also many differences between Arf1 and Arf4 (Christis and Munro, 2012). Arf1 and Arf4 showed similar interactions with COP I coat proteins and GM130 from insect cell extracts, but Arf4 preferentially bound Rab6 and Rab11, the two Rabs known to be involved in trafficking of both Drosophila and vertebrate rhodopsin (Deretic and Papermaster, 1993; Mazelova et al., 2009; Satoh et al., 2005; Shetty et al., 1998). A more specific role for Arf4 in directing transport out of the Golgi complex is further substantiated by its binding to GMAP210, which is implicated in ciliary trafficking in photoreceptors (Follit et al., 2008; Keady et al., 2011), and a ninefold higher interaction with Src, a kinase that regulates Golgi exit (Pulvirenti et al., 2008) and phosphorylates the Arf4 GAP ASAP1 (Brown et al., 1998).

Our earlier study in transgenic frogs showed that the Arf4I46D mutant, deficient in GTP hydrolysis by ASAP1, caused dysfunctional rhodopsin trafficking and rapid retinal degeneration (Mazelova et al., 2009). Nevertheless, the role of Arf4 in rhodopsin trafficking in mouse retinas has been brought into question by monitoring morphology of photoreceptors in a conditional knockout mouse (Pearring et al., 2017). Using a mouse model system with low demands on membrane trafficking volumes, the authors reported that the absence of Arf4 caused no mislocalization of rhodopsin, as evidenced by the morphological appearance of the published data. However the data are difficult to interpret as no quantification of rhodopsin localization was performed, although mild mislocalization was evident. Notably, in the same mouse, in cells with high volumes of cargo and membrane transiting through the secretory pathway, such as the exocrine pancreas, the absence of Arf4 caused a major phenotype. The most likely explanation for the data showing that mouse photoreceptors lacking Arf4 appear to deliver rhodopsin to the ROS is that the compensatory mechanisms, probably involving Arf1, which interacts with many of the same proteins (Christis and Munro, 2012), allow the process to proceed, perhaps at a suboptimal level. Over time, an Arf4 knockout retina might prove to be more susceptible to light damage and other stress leading to slow retinal degeneration, as Arf4 is also implicated in the signaling pathway mediating Golgi stress response (Reiling et al., 2013).

There are four issues of significance when comparing the data from the two published Arf4 in vivo models. (1) The absence of a gene versus a dominant-negative action often has different effects ex vivo (in cellular models) and in vivo e.g. even retinas of a rhodopsin hypomorph mouse develop normally, rods elaborate ROS of normal size, and retinas look identical to controls at P41, whereas comparable expression of a dominant negative mutant affecting the VxPx CTS of rhodopsin causes retinal degeneration modeling ADRP (Concepcion and Chen, 2010; Concepcion et al., 2002; Humphries et al., 1997; Lem et al., 1999; Li et al., 1996). (2) The volume of membrane trafficking in the frog eye exceeds by an order of magnitude that of the rodent rods. Xenopus and Rana photoreceptors synthesize and transport ∼3 µm² and ∼1.5 µm² of membrane per minute, respectively, versus 0.1 µm² synthesized by rodent photoreceptors (Besharse, 1986). Additionally, owing to their larger size, light-sensing membranes in amphibians contain 6×10⁴ molecules of rhodopsin versus 2000 molecules of rhodopsin in rats. (3) Mouse models do not always recapture retinal membrane trafficking disease phenotype e.g. despite a relatively faithful manifestation of the hearing and balance disorders found in Usher syndrome, none of the Usher 1 mouse models undergo retinal degeneration (Williams, 2008). (4) Neither the frog nor mouse models are faithful representations of the human eye, but they are useful when dissecting disease-related processes. The frog, by magnifying the role of trafficking through its high volumes of membrane and cargo synthesis and vectorial transport, allows us to dissect the stages and molecular machineries involved. The mouse has its own advantages, and it would be of interest to follow up on the absence of, or only very mild morphological change as discussed above, by generating a knock-in mouse with a dominant-negative mutant to assess the role of Arf4 in this particular model.

The cognate-activating Arf GEFs principally control the membrane association of Arfs (Bui et al., 2009; Casanova, 2007; Nawrotek et al., 2016; Stalder and Antonny, 2013), but their membrane recruitment also involves protein interactions that include SNAREs and the ciliary cargo (Honda et al., 2005; Mazelova et al., 2009). Initially, Arf1 weakly associates with membranes through the N-terminal myristoyl group, but GEF activation and GTP binding cause a conformational transition, termed the ‘myristoyl switch’, which tightly couples Arf1 activation with stable membrane association (Antonny et al., 1997; Franco et al., 1996; Goldberg, 1998; Pasqualato et al., 2002; Randazzo et al., 1995). Activation of Arf1 by Arf GEF Sec7 is amplified by the DCB and HUS regulatory domains that form a single compact helical structural unit, which facilitates membrane insertion of the Arf1 amphipathic N-terminal helix (Richardson et al., 2016). By contrast, GDP-bound Arf4 and Arf5 stably associate with membranes independently of GBF1 (Chun et al., 2008). The membrane-binding properties of Arf4 and Arf5 that differ from those of Arf1 and Arf3 are mediated by the N-terminal amphipathic helix and a class-specific residue in the conserved interswitch domain (Duijsings et al., 2009). Our study suggests that unique properties of Arf4 and its N-terminal amphipathic helix could be responsible for GBF1 DCB-HUS interactions that involve both GDP and GTP-bound Arf4. If activation and membrane association are uncoupled in class II Arfs, then multiple, random activation events at the Golgi/TGN can create a small number of active Arf4 clusters. Through the operation of an autocatalytic amplification mechanism (positive feedback) (Jackson, 2014), these may serve to quickly build up levels of active Arf4 that recognizes and directly binds to the VxPx C-terminal signal of the incoming ciliary cargo, such as rhodopsin, leading to the assembly of the ciliary-targeting complex (Mazelova et al., 2009). Similar molecular interactions might be involved in epidermal differentiation, as Arf4 recognizes the VxPx motif of presenilin-2 and regulates its localization to basal bodies/cilia to modulate Notch signaling (Ezratty et al., 2016).

Further studies will be necessary to determine whether the GBF1-ASAP1 interaction results in reciprocal changes in their catalytic activity to modulate the location and the duration of Arf4 signaling. However, they will require a new paradigm: the comprehensive analysis of full-length Arf GEFs to reveal aspects of their regulation and functions that cannot be identified by using isolated domains and truncated proteins as employed thus far. Nevertheless, given the remarkably high conservation of the GEF and GAP cascades that regulate the ordered recruitment and activation of small GTPases (Deretic, 2013; Mizuno-Yamasaki et al., 2012; Stalder and Antonny, 2013), our finding that the prototypical ciliary membrane receptor rhodopsin might promote its transport from the Golgi to the primary cilium through Arf GEF-cognate-Arf interaction implies that other membrane cargo may also promote its progression through the endomembrane system through hierarchical interactions with the highly conserved functional GTPase networks.

GST DCB-HUS (AA 1–710) and GST Sec7-HDS1 (AA 695–1066) in the pGEX-4T1 vector (Bouvet et al., 2013), as well as the purified human Δ17Arf1, were kind gifts from Cathy Jackson (Institut Jacques Monod, CNRS, Paris, France). Purified bovine rhodopsin was a gift from Kris Palczewski (Case Western Reserve University, Cleveland, OH) (Palczewski et al., 2000). Purified Arl1 and anti-Arl1 were kind gifts from Antonio Galindo (MRC LMB, Cambridge, UK) (Galindo et al., 2016). Recombinant human Arf4 was expressed and purified as described previously (Wang et al., 2012). GCA and cycloheximide were from Sigma-Aldrich. Antibodies used in this study were as follows: rabbit polyclonal anti-Arf4 (Mazelova et al., 2009); anti-rhodopsin C-terminus mAb 11D5 (Deretic and Papermaster, 1991) and mAb 1D4 (ab5417, Abcam); rabbit anti-GBF1 (ab105111, Abcam); mouse monoclonal anti-GBF1 (612116, BD Biosciences), anti-ASAP1 (612073, BD Biosciences) and anti-GM130 (610823, BD Biosciences); rabbit polyclonal anti-Rab6 (sc-310, Santa Cruz Biotechnology); rabbit anti-p115 (13509-1-AP, Proteintech); rabbit anti-ASAP1 [a kind gift from Paul Randazzo (National Cancer Institute, Bethesda, MD) (Randazzo et al., 2000)]; mouse anti-Arf1 (ThermoFisher Scientific), mouse anti-GST (SAB4200237, Sigma-Aldrich); Cy3- and Cy5-conjugated secondary antibodies (Jackson ImmunoResearch) and To-Pro3 (Life Technologies). The Duolink II Rabbit/Mouse Red Kit (excitation, 598 nm; emission, 634 nm) was from Sigma-Aldrich (DUO92101). For some experiments, rabbit anti-Rab6 antibody was directly conjugated to Alexa Fluor 488 using an Antibody Labeling Kit (Invitrogen), according to the manufacturer's instructions.

These experiments were performed according to established procedures (Deretic, 2000; Deretic and Papermaster, 1991; Deretic et al., 1996; Mazelova et al., 2009; Morel et al., 2000; Wang et al., 2012). Our animal protocol is approved by the Office of Animal Care Compliance at The University of New Mexico. Briefly, following 1 h pulse labeling of seven isolated frog retinas (from Rana berlandierii; Carolina Biological Supply Company, Burlington, NC) and 2 h chase, ROSs were removed and pelleted without further purification (crude ROS), and retinal pellets were homogenized and centrifuged to generate the PNS. The PNS was centrifuged at 17,500 g for 10 min to obtain a pellet enriched in Golgi/TGN/ER membranes. The supernatant was centrifuged at 336,000 g for 30 min to separate the RTCs from the cytosol. In some experiments, retinas were incubated with 10 µm GCA or 30 µg/ml cycloheximide for 3 h, during pulse-chase labeling.

Confocal microscopy was performed on dark-adapted frog retinas as described (Mazelova et al., 2009). In some experiments isolated eyecups were incubated for 3 h with 10 µm GCA or 30 µg/ml cycloheximide. Eyecups were fixed with 4% paraformaldehyde overnight and embedded in 5% agarose. Then, 100 μm sections were cut, permeabilized in 0.3% Triton X-100, and labeled with specific antibodies as described in the figure legends. Antibodies used for confocal microscopy were as follows: rabbit polyclonal anti-Arf4 (1:400); anti-rhodopsin C-terminus mAb 11D5 (1:400); rabbit anti-GBF1 (1:200); mouse anti-GBF1 (1:200), mouse anti-ASAP1 (1:200); mouse anti-GM130 (1:200); rabbit polyclonal anti-Rab6 (1:200); rabbit anti-p115 (1:200); rabbit anti-ASAP1 (1:200). Incubation with primary antibodies was followed by incubation with Cy3- and Cy5-conjugated secondary antibodies (1:200). Nuclei were stained with To-Pro3 (1:1000). PLA was performed on fixed retinal sections using a Duolink II Rabbit/Mouse Red Kit, as described previously (Wang et al., 2012). In some experiments, retinal sections labeled with Duolink were stained overnight at 4°C with anti-Rab6 antibody conjugated to Alexa Fluor 488 (1:100) to highlight Golgi localization. Confocal optical sections were generated on a Zeiss 800 LSM (Carl Zeiss). Digital images were prepared using Adobe Photoshop CS4 (Adobe Systems). Colocalization analysis (Pearson's coefficient) was calculated using SlideBook Image Analysis software (Intelligent Imaging Innovations). To quantify interaction sites detected by the PLA in control, GCA- or cycloheximide-treated retinas, three separate experiments were conducted, each including the rhodopsin-Arf4 pair as a positive control (Wang et al., 2012), and >10 Z-stacks containing ≥10 confocal optical sections were generated for each PLA pair. From these Z-stacks, two representative confocal sections, generally from the middle of the stack, encompassing at least five photoreceptors with clearly visible RIS demarcations were selected from each experiment. Interaction sites were counted (10 photoreceptors×3 experiments) for each PLA pair. In control retinas, numerous interaction sites were detected in nearly every photoreceptor, whereas in GCA- and cycloheximide-treated retinas, occasional interaction sites were detected in <50% of the cells counted, except in the GBF1-ASAP1 pairs.

To analyze direct protein interactions, purified human proteins, Arf4 (Wang et al., 2012), Δ17Arf1 or Arl1 (5 µg each), were preincubated with 100 µM GDPβS or GTPγS in 100 µl nucleotide loading buffer (25 mM HEPES, pH 7.4, 100 mM NaCl, 0.5 mM MgCl₂, 1 mM EDTA, 1 mM ATP and 1 mM DTT) at 30°C for 1 h. Purified bovine rhodopsin [functionally equivalent to frog rhodopsin (Deretic et al., 1998)], was in a 1.6 mg/ml solution in a buffer containing 0.5 mM n-dodecyl-β-maltoside (DDM). GST and GST fusion proteins were expressed in Rosetta 2 E.coli cells and direct protein interactions were analyzed as described previously (Wang et al., 2012). Briefly, GST fusion proteins on glutathione-Sepharose 4B beads (50 µl per sample) were washed 2× with PBS. Immobilized GST fusion proteins were incubated at RT for 2 h in 500 µl reaction buffer (50 mM HEPES, pH 7.4, 150 mM NaCl, 5 mM MgCl₂ 6H₂O, 0.1% Triton X-100, 0.1% BSA and 1 mM PMSF) with 5 µg each of Arf4, Δ17Arf1, Arl1 and/or rhodopsin, as indicated. Glutathione-Sepharose 4B beads were then washed eight times with the reaction buffer. Bound proteins were eluted by 20 µl of 2× SDS-PAGE sample buffer. Protein-protein interactions were analyzed by SDS-PAGE and western blotting.

Proteins were separated by SDS-PAGE on 4–15% TGX gels (BioRad). Gels were either dried and exposed to autoradiography, or blotted onto Immobilon-P membranes (BioRad) and probed with specific antibodies, as indicated. The antibodies used for western blotting were as follows: rabbit polyclonal anti-Arf4 (1:1000); anti-rhodopsin C-terminus mAb 11D5 (1:1000), mAb 1D4 (1:500); anti-Arl1 (1:500); rabbit anti-GBF1 (1:1000); mouse monoclonal anti-GBF1 (1:1000); mouse anti-ASAP1 (1:1000); rabbit anti-ASAP1; mouse monoclonal anti-GST (1:1000) and anti-Arf1 (1:100). Bound antibodies were detected using a chemiluminescent Western Lightning immunodetection system (Perkin Elmer Life Sciences). Because of high retinal tissue requirements, and for more accurate quantification obviating the need for additional loading controls, immunoblots were cut into strips and multiple antibodies were tested on the same blot. Before its use on the strips, each antibody was tested on an entire western blot of the frog retinal PNS to ascertain its specificity. The distribution of radiolabeled rhodopsin, or antigens detected by immunoblotting, was quantified using Quantity One 1-D analysis software (BioRad) and expressed in arbitrary OD units.

We thank Drs Antonio Galindo, Cathy Jackson, Kris Palczewski and Paul Randazzo for their generous gifts of reagents.