The golgin PpImh1 mediates reversible cisternal stacking in the Golgi of the budding yeast Pichia pastoris

The Golgi plays a centralized role in the processing, sorting and secretion of various cargo molecules destined for various intracellular and extracellular destinations (Nakamura et al., 2012). The Golgi display variable shape in different species: from dispersed cisternae in Saccharomyces cerevisiae to stacked cisternal structures in Pichia pastoris and to laterally connected ribbon of cisternal stacks in vertebrates. However, the mechanisms that regulate such exotic organizations are still poorly understood (Lowe, 2011; Papanikou and Glick, 2009).

In the light of the cisternal maturation model, the cisternal stacking of the Golgi needs to be reversible – to be initiated and undone in a continuous cycle to keep up with the cycle of cisternal maturation. Microscopic evidence in support of such reversible nature of the cisternal stacking has been reported. Trans-Golgi network (TGN) elements have been shown to peel off from Golgi stacks in various species ranging from P. pastoris to plants to mammals (Mogelsvang et al., 2003; Mollenhauer and Morré, 1991). These well-documented results point to a potential evolutionarily conserved regulatory mechanism of reversible stacking. However, mechanistically, nothing is known about such reversible stacking to date.

The most well-known factors that have been reported to play a role in cisternal stacking are GRASP proteins. Double knockout of GRASPs in mammalian cells (GRASP55 and GRASP65, also known as GORASP2 and GORASP1, respectively) disperses the Golgi ribbon structure into individual cisternae and tubulovesicular structures (Bekier et al., 2017). GRASPs are peripheral membrane proteins on the cytoplasmic face of the Golgi cisternae that form trans-oligomers through their N-terminal GRASP domain and thereby adhere adjacent cisternae together into a stack and link Golgi stacks into a ribbon, suggesting oligomerization as a mechanism of cisternal stacking (Zhang and Wang, 2015). However, no regulatory on–off switch that regulates the cisternal stacking function of the GRASPs to accommodate the reversible stacking theory has yet been discovered.

Moreover, the primary functions of GRASPs are not at all conserved among different species. In the budding yeast P. pastoris deletion of the GRASP homolog GRH1 does not affect Golgi stacking, and plant cells apparently have no identifiable GRASP homolog (Levi et al., 2010; Ito et al., 2014). These studies suggest that GRASP homologs are not functionally conserved regarding Golgi stacking in species that evidently displays stacked Golgi. Such observation indicates that other Golgi factors could mediate regulated adhesive interaction necessary for the cisternal stacking.

The golgins are long coiled-coil domain proteins that have been shown to be necessary for the maintenance of the Golgi structure and vesicle tethering (Ramirez and Lowe, 2009). The discovery of the golgins coincided with the observation that Golgi membranes could be extracted with detergent to leave a proteinaceous matrix that retained the organization of Golgi cisternae. Golgins such as GM130/golgin-95 (also known as GOLGA2) and p115 (also known as USO1) were found to be components of this matrix, and with their elongated structure, it was suggested that golgins function in the maintenance and establishment of Golgi structure (Slusarewicz et al., 1994; Nakamura et al., 1995; Waters et al., 1992). Knockdown of Golgin-97, Golgin-245 and GCC185 (also known as GOLGA1, GOLGA4 and GCC2, respectively) have been shown to affect the Golgi structure, suggesting a role for golgins in Golgi structure maintenance (Lu et al., 2004; Derby et al., 2007; Alzhanova and Hruby, 2006). Recently it has been reported that efficient stacking occurs in the absence of GRASP65 and/or GRASP55 when either of these two golgins is overexpressed (Lee et al., 2014). This result suggests that golgins serve as potential alternative cisternal adhesive factors.

GRIP-domain-containing golgins associate and dissociate with the Golgi in a GTP-dependent manner (Setty et al., 2003; Wu et al., 2004). In this study, we have tested the role of the GRIP domain golgin PpImh1p in the Golgi structure in the budding yeast P. pastoris. P. pastoris provides an excellent model to study cisternal stacking as we can study the individual Golgi stack and adhesion between two individual Golgi cisternae. We observed that deletion of P. pastoris GRIP domain golgin PpIMH1 dramatically increases the peeling of late cisterna, causing partial unstacking from the rest of the Golgi stack. Deletion of the PpIMH1 dimerization motif, overexpression of its GRIP domain alone, deletion mutants of arl1 or arl3, and a GDP-locked arl3 mutant also show a similar phenotype. We have shown the evolutionarily conserved GRIP domain golgin PpImh1 mediates reversible stacking between medial and late cisterna, and this mechanism is regulated by the Arl3–Arl1 GTPase cascade switch.

We wanted to investigate the potential roles of GRIP-domain-containing golgins in mediating cisternal stacking in budding yeast. In the budding yeast P. pastoris, the only GRIP-domain-containing golgin is PpImh1 (Jain et al., 2018). We here test whether the deletion of PpImh1 has any effect on cisternal stacking. To study cisternal stacking, we decided to create a Pichia strain in which early Golgi compartment and late Golgi compartments are labeled with two different fluorescent protein fusions. For studies in S. cerevisiae, our lab, as well as other groups, has used fluorescently tagged versions of the early Golgi protein VRG4 and the late Golgi protein Sec7 to mark early and late Golgi compartments, respectively (Bhave et al., 2014; Iyer et al., 2018; Losev et al., 2006). For our Pichia assay system, we created a strain in which the early Golgi protein VIG4 is tagged with msGFP, and late Golgi protein SEC7 is tagged with DsRed.M1x6 (hereafter referred to as the two-color Golgi strain). Live cells were imaged by fluorescence confocal microscopy. Golgi cisternae were visualized both with 2D projections along with 3D rendering, for quantitative measurements. In wild-type cells, we observe that GFP–VIG4 and SEC7–DsRed.M1x6 show elongated green and red signal, respectively, representing early and late cisterna stacked in close proximity resulting in a ‘traffic light’ type of juxtaposed signal (Fig. 1A). The green and red signals are very close and almost located on the top of each other. However, deletion of the golgin PpImh1 results in a clear separation of these green and red signals, allowing each cisterna to be visualized distinctly with no apparent overlap (Fig. 1A). This result suggests that PpImh1 deletion causes a slight separation between Golgi cisternal stacks, which was confirmed by quantitative measurements of the distance between the center of green and red spots achieved through 3D rendering (Fig. 1B). To gain further insight into structural details, we resorted to electron microscopy of Ppimh1Δ cells along with wild-type cells, as a control. Electron microscopy (EM) data show that there is a clear increase in the inter-cisternal distance between medial and late Golgi compartments in Ppimh1Δ cells (Fig. 1C). It appears, in the majority of the cells, that the TGN is positioned at an angle to the Golgi stack with only one of its ends attached to the rest of the stack. To quantitatively characterize this mutant phenotype, we have measured several parameters from the EM micrographs (Fig. 1D). We found that the inter-cisternal angle between TGN and medial Golgi was increased in Ppimh1Δ strain as compared to the wild-type. Moreover, the total area covered by the entire Golgi stack was also increased. These experiments indicate that PpImh1 deletion affects cisternal stacking of Golgi and possibly between medial and late Golgi compartments.

PpIMH1 contains an N-terminal head domain, Golgi-localizing C-terminal GRIP domain and long central coiled-coil domains (Munro and Nichols, 1999). Through a two-hybrid analysis, we have previously provided direct evidence that coiled coil domain of PpImh1 can mediate self-dimerization (Jain et al., 2018). The long coiled-coil domains could potentially mediate dimerization of golgin molecules residing on two different Golgi cisterna and multiple such dimerized golgin pairs could bring Golgi cisternae together to form a stack. To test this hypothesis, we need to test whether the coiled-coil domain is essential for cisternal stacking or not. According to coiled-coil domain analysis, the central region of PpImh1 (residues 150–1070) is likely to form a coiled-coil structure (Fig. S1) (Jain et al., 2018). Full-length PpImh1 was fully competent to rescue the unstacking phenotype of PpImh1Δ cells, but PpImh1Δ(150-1070) was not able to rescue the unstacking phenotype (Fig. 2). PpImh1(150-1070)Δ localizes to the Golgi (Fig. S2). Cells with a deletion of the PpImh1(150-1070) coiled-coil domain from the endogenous PpIMH1 in the background of the two-color Golgi strain displayed a similar cisternal unstacking phenotype (Fig. 2C). We also found that the inter-cisternal distance, area cover by the entire Golgi stack and inter-cisternal angle increase in the PPY12-PpImh1(150-1070)Δ strain (Fig. 2D). Along with previously demonstrated direct evidence of self-dimerization capabilities of PpImh1 the coiled-coiled domain (Jain et al., 2018), these results suggest that PpImh1(150-1070), which has shown a high probability to form a coiled-coil domain, is essential for the cisternal stacking function of PpImh1.

The GRIP domain of TGN golgins acts as a Golgi-localizing signal. A GRIP domain tagged with GFP is known to localize to the late Golgi/TGN in yeast and mammalian cells. In addition, overexpression of a GRIP domain causes a dominant-negative like phenotype by competing with endogenous GRIP-domain-containing proteins for binding to Arl1 (Munro and Nichols, 1999; Setty et al., 2003; Lu et al., 2004). If PpImh1 is indeed mediating the cisternal stacking through dimerization of coiled-coil regions, then it is conceivable that the overexpression of GRIP domains would saturate all the binding sites in PpImh1. Accordingly, such overexpression in P. pastoris cells might cause cisternal unstacking as dominant-negative-like phenotype as most of the endogenous PpImh1 would not be localized to the Golgi. To test this hypothesis, we overexpressed the GRIP domain under the methanol-inducible AOX1 promoter; this indeed resulted in a cisternal unstacking phenotype (Fig. 3A). We observed that GRIP domain overexpression resulted in an increase in the inter-cisternal distance, the area cover by the entire Golgi stack and inter-cisternal angle (Fig. 3B). However, similar overexpression of the coiled-coil domain did not show any effect on cisternal stacking (Fig. 3A,B). To understand the localization of these overexpressed domains, we tagged the GRIP domain (Pp-GRIP domain) and coiled-coil (Pp-CC) domain with GFP, and expressed them under the control of AOX1 promoter. Western blot analysis confirmed that these fusions were stably expressed and were free from any proteolytic degradation (Fig. S2C). We observed that overexpressed Pp-GRIP domain showed a typical punctate Golgi pattern, while overexpressed Pp-CC, which lacks a GRIP domain, failed to localize to the Golgi and instead was accumulated in cytosol (Fig. S2B). Hence, we conclude that overexpression of the coiled-coil domain alone did not affect cisternal stacking owing to the lack of Golgi localization of the overexpressed fusion protein. This result also suggests that although the coiled-coil domain may be essential for cisternal stacking, it is itself not sufficient to confer the cisternal stacking function, because it requires the GRIP domain to localize to Golgi. It is to be noted that GRIP domain overexpression in mammalian cells causes disruption in Golgi morphology and substantially affects TGN architecture (Yoshino et al., 2003). Taken together, all these results support our hypothesis and potential conservative nature of function of the GRIP domain of golgin in cisternal stacking.

Yeast Arl1 and Arl3 are divergent members of the ARF family of GTPases, referred to as ARF-like or ARL GTPases, and they show a high level of conservation with the human ARL1 and ARF-related protein (ARP) GTPases, respectively. Arl1-GTP interacts with the GRIP domain, and this interaction regulates the Golgi recruitment of Golgin-97, the mammalian homolog of Imh1. In the budding yeast S. cerevisiae, Arl1–Arl3 works in cascade, whereby the GTPase cycle of Arl3 regulates the Golgi localization of Arl1, which in turn binds to the GRIP domain of Imh1 and recruits it to the Golgi. Arl3 and Arl1 are reported to be receptors for the GRIP domain proteins (Lu and Hong, 2003; Panic et al., 2003; Setty et al., 2003). To validate this in P. pastoris, we tagged the chromosomal locus of PpIMH1 with GFP and transformed into wild-type and arl3Δ strains separately. GFP–Imh1 was found to be localized throughout the cytoplasm in case of arl3Δ cells, whereas in the wild-type it was localized to the Golgi (Fig. S3A). A similar result was previously observed with the arl1Δ strain (Jain et al., 2018). Together, these results suggest that the roles of both ARL3 and ARL1 are functionally conserved. We also determined the localization of GFP–PpImh1 in strains with GDP-locked Arl3 (T31N). We observed that the GFP–PpImh1 signal mostly localized to the cytosol, suggesting that the GTP-GDP-association-based regulatory role of Arl3–Arl1 GTPase cascade switch for PpImh1 recruitment to Golgi is also conserved (Fig. S3B).

Since the recruitment of PpImh1 to the Golgi is dependent on the function of Arl3–Arl1, we hypothesize that deletion of either arl3 or arl1 in P. pastoris cells should result in a similar cisternal unstacking phenotype to that seen when PpImh1 is deleted. To test that, we created arl3Δ and arl1Δ strains in the background of the two-color Golgi strain. Indeed, we observed a cisternal unstacking phenotype in both the strains both through light microscopy and electron microscopy (Fig. 4A,B). There was an increase in inter-cisternal distance, area cover by the entire Golgi stack and inter-cisternal angle (Fig. 4C). These results once again strengthen our hypothesis that PpImh1 is indeed mediating the cisternal stacking of Golgi.

The cisternal maturation model suggests that early cisterna mature into late cisterna and the late compartment/TGN ‘peels’ away from the stack (Mogelsvang et al., 2003). The repeated cycle of cisternal maturation needs to be continued to maintain the Golgi cisternal stack, which suggests that cisternal stacking needs to be reversible. Arl3–Arl1 works in coordination, so that the GTPase cycle of Arl3 regulates Golgi localization of Arl1, and that, in turn, binds to the GRIP domain of PpImh1p and recruits it to the Golgi. It is expected that the Arl3–Arl1 GTPase cascade, through its GTP hydrolysis cycle, functions as an oscillatory regulatory switch for association and dissociation of PpImh1 to the Golgi, and thereby regulates its function in reversible stacking. In the absence of this regulation, or in the absence of PpImh1 itself, it would be expected that such reversible stacking would be lost, resulting in faster TGN peeling. To test this concept, we captured 4D live cell movies of the arl3Δ strain (Movie 4; Fig. 4D) along with wild-type cells (Movie 1, Fig. S4). 4D movies of arl3Δ, harboring two-color Golgi cisterna, shows that these cells have increased TGN peeling (Fig. S5). TGN peeling events occur (as red-labeled late cisterna that are be seen to peel off from the cisternal stack) and new red late cisterna can subsequently be seen maturing. We also captured live-cell 4D movies of PpimhΔ (Movie 2) and arl1Δ (Movie 3) mutant strains, as well as movies from cells expressing the GDP-locked version of Arl3 (Arl3T31N) (Movie 5). All strains showed a significant increase in the TGN peeling frequency as compared to wild type (Fig. S5). These results suggest that PpImh1 indeed mediates the reversible stacking function, which is regulated by the Arl3–Arl1 GTPase cascade switch. The GTP hydrolysis cycle of the Arl proteins regulates PpImh1 association and dissociation to Golgi cisterna and thereby functions as an on-off switch for the stacking function of PpImh1 in a periodic manner.

Knockdown of individual mammalian GRIP domain proteins has been reported to cause defects in the retrograde trafficking of some cargo proteins from endosomes to the TGN (Derby et al., 2007; Lu et al., 2004). Golgin-97, the mammalian homolog of PpImh1 has been implicated in recycling endosome trafficking (Burguete et al., 2008; Lu et al., 2004).

To test whether deletion of PpImh1 has any effect on endosome to TGN trafficking, we tagged, with GFP, the cargo protein Vps10, which shuttles between pre-vacuolar endosome and late Golgi/TGN. Vps10 functions as a sorting receptor in the Golgi compartments for transport and sorting of vacuolar proteins like CPY. Vps10 cycles between the late Golgi and pre-vacuolar endosome compartment (Cooper and Stevens, 1996; Marcusson et al., 1994). We checked the localization of Vps10–GFP in wild-type P. pastoris cells, which shows that Vps10 localized to compartments labeled by late Golgi marker Sec7 (Fig. 5A). Furthermore, we checked the localization of Vps10–GFP in Ppimh1Δ and Ppimh1(150-1070)Δ cells. In the case of Ppimh1Δ as well as Ppimh1(150-1070)Δ cells, Vps10 was not localized to the late Golgi (Fig. 5A). To further confirm the localization of Vps10 in Ppimh1Δ and Ppimh1(150-1070)Δ strains, we checked the localization of Vps10 with respect to FM-4-64, a dye previously shown to mark pre-vacuolar endosomal compartments (Bhave et al., 2014; Day et al., 2018). We also undertook a kinetics experiment to standardize the optimal exposure of endosomal compartments label FM-4-64 in P. pastoris (Fig. S6B). We observed that in wild-type cells, majority of Vps10 molecules localized to compartments marked by late Golgi marker Sec7 (Fig. 5B), but in case of Ppimh1Δ and Ppimh1(150-1070)Δ strains, localization of Vps10 increased in pre-vacuolar endosome compared to wild-type (Fig. S6A). This suggests that PpImh1 deletion affects the pre-vacuolar endosome to late Golgi/TGN trafficking.

A short well-conserved region at the N-terminus of TGN golgins has been shown to be necessary and sufficient to nucleate the capture of endosome-to-Golgi carriers in mammalian systems. (Wong et al., 2017; Cheung and Pfeffer, 2016).To validate whether the similar region of PpImh1 is functionally conserved or not, we deleted the N-terminal 100 amino acids of endogenous PpImh1. We observed that, in such strains, Vps10–GFP failed to localize in Golgi, suggesting that pre-vacuolar endosome to late Golgi vesicle capturing function is compromised (Fig. 5; Fig. S6A). These results further confirm that the deletion of only 1–100 amino acids residues of endogenous PpImh1 is sufficient to abolish the vesicle capture function of PpImh1.

Deletion of the coiled-coil domain of endogenous PpImh1 also abolished the vesicle capture function (Fig. 5). This suggests that both the N terminal domain (amino acids 1–100) and the coiled-coil domain are necessary for the vesicle capture function of PpImh1. We also tested the effect of the N-terminal deletion on the cisternal stacking phenotype by electron microscopy (Fig. 6C) and light microscopy (Fig. 6A). Surprisingly, this experiment revealed that there was no change in inter-cisternal distance and other parameters (Fig. 6B,D). This result suggests that the cisternal stacking function of PpImh1 is not dependent on its vesicle capturing function. However, the vesicle capturing function may be dependent on its stacking function, as the deletion of the coiled-coil domain (the essential domain for cisternal stacking function) also abolishes the vesicle capture function.

GRIP domain golgins are known to localize in late Golgi/TGN. The GRIP domain of Golgin-97 is sufficient for the recruitment of Golgin-97 to the TGN (Munro and Nichols, 1999; Setty et al., 2003). To test what is the exact localization of PpImh1, we tagged PpImh1 with GFP and checked its localization with respect to early and late Golgi markers. GFP–PpImh1 was found to overlap with both cis-Golgi and late-Golgi markers (Fig. 6E). By measuring the percentage of the green spot that overlapped the red spot, we confirmed that it colocalized with both early and late Golgi (Fig. 6F). These results suggested to us that PpImh1 could be localized to the medial compartment. This result also fits well with our hypothesis that PpImh1 mediates the stacking between medial and late Golgi. We also observed that GFP–PpImh1 forms a ring-like pattern (Fig. 6E), with a central clearance and higher concentration of signal at the periphery. This suggests that PpImh1 may localize at the periphery of the Golgi cisterna to mediate cisternal stacking. To further analyze this, we compared the localization pattern of the Golgi-localizing GRIP domain (GFP–GRIP) and full-length PpImh1 tagged with GFP. GFP–PpImh1 showed an elongated ring-like the pattern, but in the case of the GRIP domain, we observed a punctate pattern (Fig. S7).

The role of golgins in maintaining Golgi structure has been experimentally shown previously (Nakamura et al., 1995; Slusarewicz et al., 1994; Waters et al., 1992; Derby et al., 2007; Lu et al., 2004), but, until now, there was not definitive proof of a direct role for any golgin in maintaining reversible cisternal stacking. Our results, for the first time, suggest that golgin PpImh1 knockout affect cisternal stacking between medial and late-Golgi. Moreover, the coiled-coil domain of golgin PpImh1 was shown to be essential for cisternal stacking. Our previous studies have shown that PpImh1 forms parallel homodimers where central coiled-coil domain remains in the dimeric state (Jain et al., 2018), suggesting that coiled-coil domain of golgin PpImh1 dimerizes and holds cisternae together. The results shown here support our hypothesis that the long coiled-coil domain could mediate dimerization of golgin molecules residing on two different Golgi cisterna and that multiple such dimerized golgin pairs could bring two Golgi cisternae together to form a stack. Our data suggest that PpImh1 mediates the cisternal stacking of late and medial Golgi. We also have shown that the Arl3–Arl1 GTPase cascade switch regulates this function. Our results suggest that the GTP hydrolysis cycles of Arl proteins regulate the cycle of PpImh1 association and dissociation to Golgi cisterna and thereby function as an on-off switch for the stacking function of PpImh1 in a periodic manner; this, in turn, results in the reversible stacking.

Based on our results, we propose the following working model of PpImh1 on Golgi cisterna (Fig. 7A). First, GTP-Arl3 recruit Arl1, and as a result, PpImh1 associates to medial and late cisterna through its GRIP domain. As late cisterna matures to TGN, Arl3 goes through GTP hydrolysis and becomes GDP-Arl3. As a result, the GRIP domain anchor of PpImh1 dissociates from the maturing TGN. This, in turn, initiates the peeling of TGN. Eventually, the remaining anchor of the PpImh1 dimer dissociates from corresponding cisterna, as it matures, and the local GDP-Arl3 population increases. As a result, PpImh1 dimers dissociate, which in turn causes complete dissociation of the TGN from the Golgi stack. This separating form of TGN, also referred to as ffTGN, has been visualized experimentally to separate rapidly from the rest of the Golgi stack (Mogelsvang et al., 2003). As PpImh1 dimers dissociate from outgoing late compartments, new PpImh1 dimers reform, tethering between freshly matured medial and late cisterna with the progression of cisternal maturation. When PpImh1 is absent, functionally inactive or not recruited to the Golgi, the tethering between medial and late cisterna is lost. As a result, the late compartment, including the TGN peels off earlier and at a faster rate, as seen by Sec7-labeled compartments dynamics (Fig. 4D). The precise mechanism of the association–dissociation cycles of PpImh1 dimers to Golgi cisterna needs further investigation. It would be interesting to know whether the dissociating PpImh1 dimers are reformed or recycled.

One intriguing question is how the cell could ensure that only PpImh1 molecules residing on two neighboring cisternae can dimerize. It is possible that PpImh1 molecules residing on the same cisterna could also dimerize, but a significant number of PpImh1 molecules, enough to mediate reversible stacking, residing on two neighboring cisternae also dimerize. A mixed population of these two types of PpImh1 dimers (depending on their GRIP domain anchoring) possibly exists. Further investigations will decipher the actual ratio of these two population at a given time.

The established function of golgins is to capture vesicles coming from the different region of the cells. The GRIP domain golgins GCC185, Golgin-97 and Golgin-245 capture vesicles coming from the endosome and transfers them to the trans Golgi (Derby et al., 2007; Lieu et al., 2007; Lu et al., 2004). Our results support that golgin PpImh1 mediates transport vesicles between endosome and the TGN. The GRIP domain golgin GCC185 forms a Y-shaped structure where the N-terminal domain forms a splayed end, which has been shown to be essential for vesicle capture (Cheung et al., 2015; Wong et al., 2017). The golgin PpImh1 forms a parallel homodimer with a splayed N terminus (Jain et al., 2018). Our data further support that the N-terminal domain of PpImh1 is essential for the vesicle capture function of golgin PpImh1.

It appears that the cisternal stacking function of PpImh1 is independent of its vesicle capturing function since deletion of the vesicle capture domain PpImh1 (amino acids 1–100) had no effect on Golgi cisternal stacking. However, the deletion of the coiled-coil domain, which is essential for cisternal stacking, affects the vesicle capture function. That suggests that stacking could be indispensable for the vesicle capture function. However, it is also possible that deletion of coiled-coil domain results in reduction in the length of golgin molecule and this leads to the failure to capture the incoming vesicle. It is to be noted that the central coiled-coil domain is possibly necessary but not sufficient to confer the cisternal stacking function. Only the coiled-coil domain, along with the GRIP domain, is sufficient for cisternal stacking function (Fig. 6C). Our result also suggests that for thevesicle capture function PpImh1 needs all its three domains, that is, the N-terminal domain, coiled-coil domain and GRIP domain (Fig. 5). Since the deletion of the N-terminal domain (amino acids 1–100) compromises vesicle capture, and the same domain has been previously shown to form the splayed N-terminal Y -shaped structure suitable for vesicle capture, most likely this region is essential for vesicle capture. However, this region is dispensable for cisternal stacking (Fig. 6C). We hypothesize that cisternal stacking is independent of vesicle capturing functions, but the efficacy of the latter may be dependent on the former, which possibly enhance the robustness of the secretory function of the Golgi.

However, a natural question arises about how PpImh1 can accommodate these two separate functions simultaneously. We have accommodated this in our model (Fig. 7B). For the vesicle capturing function, the Y-shaped N-terminal regions of golgin dimers are a favored structure in which the splayed N-terminal of dimers are proposed to mediated interaction with the vesicles (Wong et al., 2017; Cheung et al., 2015). We have shown that the N-terminal domain of PpImh1 is essential for vesicle capture. The C-terminal domains of the PpImh1 dimers anchor the Golgi membrane of neighboring cisterna in a manner mediated by Arl1 interaction. The coiled-coil region of PpImh1 contains certain ‘break’ regions which possibly form hinge region that can provide flexibility to golgin molecule to allow vesicle transport to the Golgi membrane (Cheung et al., 2015). The PpImh1Y-shaped N-terminal dimer keeps on engaging in vesicle transport, while GRIP domain anchors mediate the stacking, as described above.

Our results suggest the possible role of GRIP domain golgins in Golgi stacking and vesicle capture. It seems likely that Golgi stacking involves multiple types of interactions. For the early Golgi, the mechanisms might vary between species, with GRASPs being one set of players in animal cells. By contrast, the GRIP domain-mediated reversible stacking mechanism might be conserved at the level of the late Golgi since various species have displayed a ‘TGN peeling’ phenotype (Mogelsvang et al., 2003; Mollenhauer and Morré, 1991). Moreover, GRIP domain proteins are conserved in plants, animals and fungi, so it is plausible that this mechanism of reversible Golgi stacking is evolutionarily conserved. Our studies provide a potential molecular basis for the ‘TGN peeling off’ phenomenon that has been seen by EM for decades.

Experiments with P. pastoris were carried out using the haploid wild-type strain PPY12 (his4 arg4) (Gould et al., 1992) and its derivatives (Table S2). General methods for the growth and transformation of P. pastoris were as described previously (Sears et al., 1998). Cultures were grown in rich glucose medium (YPD), synthetic glucose medium (SD), or nonfluorescent synthetic glucose medium (NSD) (Bevis et al., 2002) in baffled flasks at 30°C with shaking at 200 rpm. Selection of strains were carried out on YPD supplemented with G418 (500 µg/ml) or Hygromycin B (250 µg/ml) or SD medium as per integrating plasmids. P. pastoris transformation was performed with linearized integrating vectors using the electroporation method. P. pastoris gene sequences were obtained from the NCBI database. Molecular biology procedures were simulated and recorded using SnapGene software. All plasmids and primers used in this study are listed in Tables S1 and S3.

The P. pastoris yeast strain expressing Pp-GRIP and Pp-CC were grown to log phase [0.5 optical density at 600 nm (OD₆₀₀)] in SYG medium (0.67% yeast nitrogen base, 0.05% yeast extract, 0.4 mg/l biotin, 40 mg/l arginine hydrochloride and 1% glycerol). Then cells were washed with SYM medium (0.67% yeast nitrogen base, 0.05% yeast extract, 0.4 mg/l biotin, 40 mg/l arginine hydrochloride and 1% methanol), and resuspended in SYM medium for 8 h to induce expression from the AOX1 promoter. The cell pellet was washed with water and lysed with breaking buffer (50 mM sodium phosphate buffer, pH 7.4, 1 mM PMSF, 1 mM EDTA and 5% glycerol) and acid-washed glass beads as per the Invitrogen Pichia expression kit protocol. The cell lysate was mixed with 2× SDS gel loading dye. The protein samples were separated by SDS-PAGE and immunoblotted with anti-His antibody (1:3000; ab18184, Abcam, Cambridge, MA,).

Full-length P. pastoris VIG4 were PCR amplified. The amplified fragments were digested with EcoRI and SphI and ligated into the pIB1 (Sears et al., 1998) cut with the same enzyme. This plasmid was then mutagenized to introduce NotI and BamHI. The resulting plasmid was digested with NotI and BamHI and the fluorescent protein msGFP was inserted at NotI and BamHI site resulting in the msGFP-VIG4-pIB1 construct. This construct was linearized with StuI to integrate at his4 locus of P. pastoris.

SEC7 was epitope tagged with a 6X-DsRed.M1 (Bevis et al., 2002) cassette by pop-in gene replacement using the same general strategy as described above for VIG4. The pop-in plasmid pUC19-ARG4(-XmnI)-PpSEC7-DsRed.M1×6 was linearized with XmnI for transformation into an arg4 strain.

1-kb sequences flanking the PpIMH1 coding sequence were amplified from genomic DNA. The amplified fragments were digested with NdeI and SalI (for the upstream fragment) and XhoI and HpaI (for the downstream fragment). The upstream fragment was ligated with a pUG6 (Guldener et al., 1996) vector that had been digested with NdeI and SalI. The resulting plasmid was cut with XhoI and HpaI to ligate downstream fragment to give pUG6-PpIMH1::Kanmax. Finally, a 3.2-kb NdeI-HpaI fragment was excised from this plasmid and transformed into PPY12 cells. G418-resistant transformants were screened by PCR to confirm that IMH1 had been deleted. Full-length PpIMH1 was PCR amplified. The amplified fragment was digested with XmaI and SphI and ligated with pIB1 (Sears et al., 1998) cut with the same enzyme. The resultant plasmid was linearized with StuI to integrate at the His4 locus.

Full-length PpIMH1 was PCR amplified. The amplified fragment was digested with XmaI and SphI and ligated with the pUC19-His (Connerly et al., 2005) cut with the same enzyme. This plasmid was then mutagenized to introduce BamHI-NotI sites. The resulting plasmid was then digested with BamHI and NotI, and the GFP tag was ligated as a BamHI–NotI fragment. The GFP-PpIMH1-pUC19-His construct was linearized with PstI to integrate at PpIMH1 locus of P. pastoris.

1-kb sequences flanking the ARL3 coding sequence were amplified from genomic DNA. The amplified fragments were digested with NdeI and SalI (for the upstream fragment) and SacII and HpaI (for the downstream fragment). The upstream fragment was ligated with pUG6 (Guldener et al., 1996) vector, which had been digested with NdeI and SalI. The resulting plasmid was cut with SacII and HpaI to ligate the downstream fragment, resulting in pUG6-ARL3: Kanmax. Finally, a 3.2-kb NdeI–HpaI fragment was excised from this plasmid and transformed into PPY12 cells. G418-resistant transformants were screened by PCR to confirm that ARL3 had been deleted.

1-kb sequences flanking the ARL1 coding sequence were amplified from genomic DNA. The amplified fragments were digested with NdeI and SalI (for the upstream fragment) and EcoRV and NotI (for the downstream fragment). The upstream fragment was ligated with a pUG6 (Guldener et al., 1996) vector that had been digested with NdeI and SalI. The resulting plasmid was cut with EcoRV and NotI to ligate downstream fragment that results in pUG6-ARL1:Kanmax. Finally, a 3.2-kb NdeI–NotI fragment was excised from this plasmid and transformed into PPY12 cells. G418-resistant transformants were screened by PCR to confirm that ARL1 had been deleted.

The open reading frame of the P. pastoris IMH1 coiled-coil domain was deleted as follows. The sequence upstream of the PpIMH1 coiled-coil domain (150–1070) and PpIMH1 (1–100) coding region were amplified from genomic DNA. The amplified fragments were digested with SmaI and BamHI and ligated with a pUC19-His (Connerly et al., 2005) vector that had been digested with SmaI and BamHI. The resulting plasmid was cut with BamHI and SphI to ligate downstream fragments to result in pUC19-PpIMH1(150-1070)Δ and PpIMH1(1-100)Δ. These constructs were linearized with EcoRI and PstI to integrate at PpIMH1 locus of P. pastoris.

A 3′ portion of the P. pastoris VPS10 coding sequence plus a downstream region were amplified by PCR. The amplified fragment was digested with EcoRI and HindIII and ligated with the pUC19-Arg4 cut with the same enzyme. This plasmid was then mutagenized to introduce BamHI-NotI sites. The resulting plasmid was then digested with BamHI and NotI, and a 3×GFP tag was ligated as a BamHI–NotI fragment. The Vps10-3×GFP-pUC19Arg4 construct was linearized with NsiI to integrate at VPS10 locus of P. pastoris.

A 3′ portion of the P. pastoris VPS8 coding sequence plus a downstream region were amplified by PCR. The amplified fragment was digested with EcoRI and SphI and ligated with the pUC19-Arg cut with the same enzyme. This plasmid was then mutagenized to introduce BamHI-NotI sites. The resulting plasmid was then digested with BamHI and NotI, and the 3xGFP tag was ligated as a BamHI-NotI fragment. The Vps8-3xiGFP-pUC19Arg construct was linearized with BspEI to integrate at VPS8 locus of P. pastoris

Live-cell dual-color 4D confocal imaging was performed previously described (Day et al., 2017) using a strain expressing GFP–Vig4 as an early Golgi marker and Sec7–DsRed as a late Golgi marker, with the following modifications. Cells attached to the cover glass dish surface were washed and covered with minimal SD medium. Image capture was performed using a Zeiss 780 and Leica SP8 confocal microscope. GFP fluorescence was visualized using 488-nm laser light source excitation and 495–550 nm bandpass emission, and DsRed fluorescence was visualized using 561-nm excitation and 580–750 nm bandpass emission. The pixel size was 65 nm, the pinhole size was 1.2 Airy units, and the interval between the optical sections was 0.3 μm. Every 10 s, 20 optical sections were captured to span the entire cell thickness. The red and green fluorescence images were converted to 16-bit and average projected, then range-adjusted to the minimum and maximum pixel values in ImageJ, and merged with blue images of the cells.

4D imaging was performed using Leica SP8 with an optical z slice of 0.3 nm, zoom factor 7, and 100× objective magnification. The 4D movies were deconvoluted using Huygens Pro software, and Z projected using ImageJ. The entire event of TGN peeling was measured by analyzing the 10-min live-cell movies.

FM 4-64 labeling of PVE or endosome was performed as described previously (Vida and Emr, 1995). Cells were grown to log phase; then, 1 ml of log-phase cells were suspended in YPD medium containing 1.6 µM FM 4-64 dye (Life Technologies, Carlsbad, CA) and incubated for 10 min at 30°C. Cells were washed and resuspended in 5 ml YPD and chased for 20 min at at 30°C and immediately observed under the microscope by spotting on the coverslip. To chase FM 4-64 in Sec7- and Vps8-tagged strains, cells were grown to log phase and incubated with FM 4-64 dye for 10 min. Cells were washed with YPD and chased for 3 min, 10 min, 15 min or 20 min.

The overlap between two differently labeled punctate compartments was quantified as described previously (Levi et al., 2010). The fraction of the GFP–PpImh1 punctate signal that overlapped with the red Golgi marker punctate signal was quantified in ImageJ. The red and green channels from a merged RGB image were separated in Photoshop (Adobe), converted into grayscale, inverted to give dark spots on a light background, processed with the Despeckle filter and saved as separate TIFF files. The TIFF file for the corresponding channel was then opened in ImageJ, and a binary threshold image of black spots on a white background was created using the Dynamic threshold 1d plug-in; a similar procedure was used to generate a modified TIFF file displaying only the punctate spots from the green channel. The total signals from the red and green spots were then obtained using the Measure command in ImageJ. To determine the fraction of the punctate red signal that overlapped with the punctate green signal, the binary threshold image from the green channel was subtracted from the TIFF file for the red channel. The resulting signal was measured and was divided by the total signal previously measured for the red spots.

Thin-section EM was performed essentially as described previously (Gould et al., 1992). In brief, a 50-ml culture of yeast cells in rich glucose medium was grown to an OD₆₀₀ of ∼0.5. The culture was concentrated to a volume of <5 ml with a bottle-top vacuum filter, and 40 ml of ice-cold 50 mM potassium phosphate buffer, pH 6.8, 1 mM MgCl₂ and 2% glutaraldehyde was added rapidly with swirling. After fixation for 1 h on ice, the cells were washed repeatedly, and then resuspended in 0.75 ml 4% KMnO₄ and mixed for 1 h at room temperature. The cells were washed, and then resuspended in 0.75 ml 2% uranyl acetate and mixed for 1 h at room temperature. Finally, the cells were embedded in Spurr's resin; 50 ml of yeast culture yielded enough cells for three BEEM capsules. The resin was polymerized for 2 days at 68°C. Sections were stained with uranyl acetate and lead citrate and viewed on an electron microscope (100 CXII; JEOL Inc). All the parameters were measured using iTEM analysis software.

Three independent experiments were performed to capture fluorescence and electron microscopy images. In each experiment, 20 images were captured. The total number of images used for analysis are given as ‘n’ values in the legend. An unpaired Student's t-test was performed to assess the statistical significances of differences among datasets. Two-tailed unpaired Student's t-tests were performed using GraphPad Prism. P<0.05 was considered statistically significant.

We thank Prof. Benjamin Glick of the University of Chicago for his various inputs and his critical comments for the preparation of the manuscript. We also thank all the Bhattacharyya lab members for carefully reviewing the manuscript.