The Golgi ribbon in mammalian cells negatively regulates autophagy by modulating mTOR activity

In contrast to protists, plants and invertebrates, vertebrates have evolved mechanisms for joining the individual Golgi stacks into a higher-order ribbon structure (De Matteis et al., 2008), typically found in a juxtanuclear location in interphase cells. Multiple Golgi stacks are linked together to form a reticular, twisted ribbon structure that is actively maintained around the centrosome by interactions with microtubules (Rios and Bornens, 2003). In addition to microtubules, actin and actin-binding proteins participate in the structural organization of the Golgi (Egea et al., 2013). However, the classic functions of the Golgi, namely membrane transport and glycosylation, do not require the ribbon structure per se, as individual Golgi stacks can perform these functions (Wei and Seemann, 2010). Surprisingly, and despite our knowledge of Golgi dynamics, the fundamental biological relevance of the ribbon structure of the Golgi in vertebrates remains a mystery. Alterations in Golgi morphology have been reported in a variety of diseases; for example, fragmentation of the Golgi ribbon is observed in a range of neurodegenerative diseases (Sundaramoorthy et al., 2015). Hence, identifying functional roles associated with the Golgi ribbon structure could also be important in understanding the molecular basis of various diseases.

The recent discovery that 180 signaling genes have an impact on the organization of mammalian Golgi (Chia et al., 2012) strongly suggests a complex network of signaling pathways that regulate the Golgi structure and could also be regulated by it. For example, it is now clearly established that constitutive transport processes are regulated by signaling pathways, a finding that represents a paradigm shift in the field (Cancino et al., 2013; Giannotta et al., 2012). Given the plethora and range of signaling genes that are wired to the Golgi complex, many cell processes could be influenced by these pathways, including mitosis, metabolic regulation, autophagy, cell polarity and migration.

Membrane tethers are a class of molecules that can regulate the dynamic organization of the Golgi. Membrane tethers of the Golgi complex include the golgins and GRASPs (Golgi reassembly stacking proteins), which are families of structural molecules defined by their location at the Golgi complex and their abundance of extensive coiled coil regions (Ramirez and Lowe, 2009; Toh and Gleeson, 2016). Most membrane tethers are peripheral membrane proteins associated with the cytoplasmic side of Golgi membranes and are recruited to Golgi membranes in a highly regulated manner (Gillingham and Munro, 2003). Given the extensive coiled coil regions of golgins, they can project into the cytoplasm as long filamentous molecules. Many membrane tethers influence the structure of the Golgi complex, and our work (and that of others) has shown that golgins of the trans-Golgi network (TGN) contribute to the dynamic organization of the Golgi (Chia and Gleeson, 2011; Derby et al., 2004; Goud and Gleeson, 2010; Luke et al., 2003; Yoshino et al., 2003). TGN golgins have the capacity to interact with microtubule and actin cytoskeletons, via linker proteins, and with membranes via interaction with Rab proteins (Goud and Gleeson, 2010; Sinka et al., 2008).

Many of the approaches previously used to study the role of the Golgi ribbon have been very blunt, such as treatment with the microtubule depolymerizing drug nocodazole (Wei and Seemann, 2010). Here, we have established a novel approach for manipulating the balance between the Golgi ribbon and Golgi mini-stacks by modulating the dose of a membrane tether or golgin located at the TGN. This strategy has enabled establishment of stable cell lines that lack a Golgi ribbon and their analysis during interphase. We report the unexpected finding that the organization of the Golgi as a compact ribbon is essential for regulating the mechanistic target of rapamycin (mTOR) pathway and autophagy. mTOR is one of the major signaling pathways of eukaryotic cells and known to be a negative regulator of autophagy (Wullschleger et al., 2006). We show that loss of the Golgi ribbon results in reduced mTOR activity and an associated increase in autophagosome biogenesis. Moreover, we have extended these findings to a model of neurodegeneration that is associated with loss of the Golgi ribbon.

We previously demonstrated that high-level overexpression of the TGN golgin GCC88 (encoded by GCC1) leads to a major perturbation of the Golgi complex (Luke et al., 2003). Here, we show that low-level overexpression of either GFP-tagged GCC88 (Fig. S1A) or Myc-tagged GCC88 (not shown) results in dispersal of Golgi fragments throughout the cytoplasm. These Golgi fragments stain for both cis-Golgi and trans-Golgi markers, indicating that the fragments represent Golgi mini-stacks (Fig. S1A). Fragmentation of the Golgi was independent of the GFP or Myc tags, as untagged GCC88 also resulted in dispersal of the Golgi ribbon into stacks that stained for both cis- and trans-Golgi markers (Fig. S1B). Given these observations, we established HeLa cell clones stably expressing GFP–GCC88. Three independent HeLa cell clones were isolated and all three showed similar characteristics. Two of these clones, namely HeLa-B6 and HeLa-C4, are presented here. Clones HeLa-B6 and HeLa-C4 showed GFP–GCC88 located in punctate structures scattered throughout the cytoplasm (Fig. 1A) and super-resolution three-dimensional structured illumination microscopy (3D-SIM) showed the extensive fragmentation of the Golgi into Golgi mini-stacks (Fig. 1B). We also applied pulse-width analysis (PulSA), which measures the width of the fluorescent signal of individual cells in the population by flow cytometry, an analysis we have previously shown can discriminate different intracellular locations (Chia et al., 2014). The HeLa-B6 cell population had a broader pulse width of fluorescently marked Golgi structures than parental HeLa cells, confirming fragmentation of the Golgi ribbon throughout the entire clonal population of HeLa-B6 cells (Fig. 1C). Immunoblotting of HeLa-B6 and HeLa-C4 cell extracts with anti-GCC88 antibodies detected both endogenous GCC88 and GFP–GCC88 (Fig. 1D) and quantitation revealed a 1.5- to 2.0-fold increase in total GCC88 in these cells, representing only a modest increase in GCC88 levels. RNA interference (RNAi) of GCC88 in HeLa-B6 cells resulted in reversal of the Golgi phenotype. Following knockdown of GCC88, the dispersed Golgi staining pattern of HeLa-B6 cells was relocated to a tight juxtanuclear staining pattern similar to that of parental HeLa cells (Fig. 1E) with >80% of cells displaying a compact Golgi (Fig. 1F). These findings indicate that the Golgi phenotype of the HeLa clones overexpressing GCC88 is a result of elevated levels of GCC88 and can provide a system for exploring the role of ribbon structure in higher-order functions.

We assessed whether membrane transport was affected in HeLa-B6 cells. We transfected parental HeLa and HeLa-B6 cells with E-cadherin, a membrane cargo that is transported from the Golgi to the cell surface (Lock et al., 2005). Assessment of the relative intracellular and cell surface levels of E-cadherin showed no difference between parental HeLa and HeLa-B6 cells (Fig. S2), indicating that the secretory transport pathway of HeLa-B6 cells was efficient in delivery of newly synthesized E-cadherin to the cell surface. In addition, the steady state distribution of the mannose-6-phosphate receptor (M6PR) and the kinetics of endosome-to-Golgi-transport of the cargoes CD8-M6PR (a CD8 chimera of M6PR) and Shiga toxin from the cell surface to the Golgi (not shown) were not altered by GCC88 overexpression in HeLa-B6 cells, indicating that intracellular membrane transport was unaffected in HeLa-B cells.

To characterize further the morphology of the Golgi in HeLa-B6 cells, electron microscopy (EM) analyses were performed on 100 nm thick sections of embedded cells. Whereas parental HeLa cells had the typical long Golgi ribbon structures in close proximity to the nucleus, HeLa-B6 cells showed Golgi mini-stacks dispersed through the cytoplasm, with little evidence of the existence of a compact ribbon structure (Fig. 2A). Three-dimensional scanning transmission electron microscopy tomography (STEM-T) of thick sections also clearly revealed the individual Golgi mini-stacks scattered throughout the cytoplasm of HeLa-B6 cells (Fig. 2B). Close inspection of the tomogram movies revealed that many of the dispersed stacks in HeLa-B6 cells were connected by very long tubules (Movie 1), whereas some stacks were physically separated from each other in the limits of the section (Movie 2). The extended tubular connections between the dispersed Golgi mini-stacks in HeLa-B6 cells are atypical of the compact organization of Golgi stacks within a classical ribbon structure (Movie 1). Quantitation of >100 Golgi profiles from >45 sections of different cells revealed a significant increase (2.5-fold) in the number of Golgi profiles and a significant reduction in length of the Golgi cisterna (1.8-fold) in HeLa-B6 cells, indicative of the loss of the long Golgi ribbon and the formation of individual mini-stacks (Fig. 2C,D). In contrast, HeLa-B6 cells depleted of GCC88 by siRNA showed a similar number of Golgi profiles as parental HeLa cells, with the length of the Golgi cisternae even longer than those of the parental cells (Fig. 2A,B,D, Movie 3), indicating enhanced Golgi ribbon profiles in the absence of GCC88. No significant difference was observed in the number of Golgi cisternae per stack nor in the average Golgi stack width, indicating that changes in Golgi morphology were directly related to loss of the Golgi ribbon (Fig. 2E,F).

EM micrographs of HeLa-B6 cells indicated the presence of an increased number of electron-dense organelles (Fig. 2A, Fig. S3A) containing double membrane electron-dense structures resembling autophagosomes (Fig. 3A). To detect autophagosomes, we stained cells for LC3, a marker of autophagosomes. Parental HeLa cells had no detectable LC3 (Fig. 3B) unless treated with rapamycin to induce LC3 expression and autophagy (not shown). Strikingly, untreated HeLa-B6 (Fig. 3B) and HeLa-C4 cells (not shown) showed strong punctate staining for LC3. Immunoblotting showed increases in the lipdated form (LC3-II) and the acylated form of LC3 (which is membrane associated) and a decrease in levels of p62, a cargo receptor for delivery to autophagosomes, in both HeLa-B6 and HeLa-C4 cells (Fig. 3C,D), findings consistent with the identity of the LC3-positive structures as autophagosomes. Overexpression of untagged GCC88 also resulted in strong staining for LC3 (Fig. S3B). Therefore, modest overexpression of GCC88 results in loss of the Golgi ribbon and induction of autophagy. On the other hand, HeLa-B6 cells depleted of GCC88 by siRNA had a greatly reduced level of LC3-positive structures compared with HeLa-B6 cells treated with control siRNA (Fig. 3E), indicating that the increase in autophagy was directly associated with the loss of the Golgi ribbon in HeLa-B6 cells. Overexpression of a number of other Golgi proteins, for example Arl5b (Houghton et al., 2012), did not result in fragmentation of the Golgi ribbon nor an increase in autophagy (not shown); hence, the induction of autophagy in HeLa-B6 cells is not due to the overexpression of a Golgi protein per se.

We also assessed whether other methods of fragmenting the Golgi ribbon induced autophagy, to confirm that the effects we observed were not specific for GCC88 alone. Depletion of a number of transport machinery components have been reported to result in changes in Golgi morphology (Chia et al., 2011; Climer et al., 2015; Derby et al., 2007; Seaman, 2004). Silencing the Vps26 subunit of the endosomal coat complex retromer and the TGN tether GCC185 both resulted in dispersal of Golgi structures and an increase in LC3-positive structures (Fig. S4A–D), demonstrating that integrity of the Golgi ribbon is important in regulating the levels of autophagy rather than selectively regulating the membrane tether GCC88.

We also considered whether an ER stress response might arise following fragmentation of the Golgi, as ER stress could account for the induction of autophagy. However, using a luciferase reporter system driven by ER stress elements (ERSE), there was no evidence of an ER stress response in HeLa-B6 (Fig. S5) or HeLa-C4 cells (not shown).

The increase in autophagosomes in HeLa-B6 cells could be due to an increase in the de novo biogenesis of autophagosomes or, alternatively, to a block in the flux of the autophagy pathway resulting in accumulation of autophagosomes prior to fusion with lysosomes. To distinguish between these two possibilities, we treated cells with bafilomyin A1 for 4 h to block fusion of autophagosomes with lysosomes. If autophagosome flux is blocked in HeLa-B6 cells, then bafilomycin A1 will have no effect on levels of LC3-positive autophagosomes. However, balifomycin A1 treatment resulted in a significant increase in LC3-II staining in both parental HeLa and HeLa-B6 cells as assessed by flow cytometry (Fig. 3F), demonstrating that the autophagosome flux is not blocked in B6 cells.

Given the role of mTOR as a negative regulator of autophagy (Lamb et al., 2013; Rubinsztein et al., 2012; Wullschleger et al., 2006), we then investigated the impact of loss of the Golgi ribbon on mTOR activity. Initially, we assessed the phosphorylation status of a downstream substrate of mTOR, ribosomal S6. Total S6 protein levels were similar in HeLa and HeLa-B6 cells (Fig. 4A,B); however, there was a dramatic reduction in phosphorylated-S6 (p-S6) in HeLa-B6 compared with parental HeLa cells as assessed by flow cytometry (Fig. 4A) and immunoblotting (Fig. 4B), indicating reduced mTORC1 activity in HeLa-B6 cells.

Next, we assessed the impact of loss of ribbon on the intracellular location of mTOR. In untreated parental HeLa cells, mTOR showed substantial colocalization with the TGN marker p230 (Fig. 4C), as well as dispersed punctate staining throughout the cytoplasm that was colocalized with the late endosomal/lysosomal marker CD63 (Fig. 4E). Although the Golgi and late endosomes are both in the perinuclear region of HeLa cells, there was no detectable overlap between the TGN and late endosome markers p230 and CD63 (Pearson's correlation coefficient was 0.08, n=66 cells), as indicated by the 3D reconstructions of cells stained for both p230 and CD63 (Fig. S6). Both Golgi and endosomal staining was abolished following silencing of mTOR with siRNA (Fig. S7), demonstrating that the staining was specific for mTOR. The anti-mTOR antibody used in these experiments has been widely used in the mTOR field, with some reports of predominantly lysosomal staining (Sancak et al., 2010) and other reports indicating some Golgi staining (Thomas et al., 2014). A range of different fixation and permeabilization procedures were used in these different reports. We also observed differences in the relative staining of late endosomes/lysosomes and the Golgi in HeLa cells based on the processing conditions used (not shown). Finally, immunogold staining of ultrathin cryosections analyzed by EM also identified mTOR associated with both tubulovesicular structures and cisternae of the Golgi in parental HeLa cells (Fig. 4G).

In contrast to parental HeLa cells, mTOR in HeLa-B6 cells was no longer concentrated on the Golgi complex as assessed by confocal microscopy (Fig. 4C); mTOR was dispersedly distributed throughout the cytoplasm. Localization of mTOR to the lysosomes in HeLa-B6 cells appeared to be unaffected by the status of the Golgi structure (Fig. 4E). In HeLa-B6 cells depleted of GCC88 by siRNA, mTOR was again associated with the Golgi complex (Fig. 5A,B). Hence, mTOR localized to the Golgi ribbon but not to fragmented Golgi mini-stacks. To assess further the importance of Golgi-localized mTOR, we localized active phosphorylated-mTOR (p-mTOR) by immunofluorescence. In parental HeLa cells, a signal for p-mTOR was detected in the juxtanuclear region and was colocalized with the TGN marker p230 (Fig. 4D). In contrast, HeLa-B6 cells showed minimal Golgi staining for p-mTOR (Fig. 4D), indicating that the association of mTOR at the Golgi promotes mTOR activation. Quantitation demonstrated a significant reduction in both mTOR and p-mTOR in the Golgi region of HeLa-B6 cells compared with parental HeLa cells (Fig. 4F). In contrast, there was no change in the level of mTOR and only a modest decrease in the level of p-mTOR on CD63-positive endosomes of HeLa-B6 cells (Fig. 4E,F).

mTOR and p-mTOR were also Golgi localized in other cell lines, for example in the neuroblastoma cell SK-N-SH, where a substantial amount of mTOR colocalized with the TGN marker and p-mTOR was also concentrated within the Golgi region (Fig. 5C). The Golgi complex and CD63-positive late endosomes are both concentrated within the perinuclear region of HeLa cells, whereas in SK-N-SH cells these two compartments are spatially more distinct. The location of mTOR on the Golgi of SK-N-SH cells confirms that a significant pool of total mTOR is associated with this organelle.

Treatment of SK-N-SH cells with rapamycin eliminated p-mTOR staining; however, the pool of inactive mTOR remained associated with the Golgi/TGN (Fig. 5D). These findings demonstrate that the Golgi presents a very significant pool of rapamycin-sensitive mTORC1 and, moreover, that this organelle is an intracellular location for mTOR activation.

To explore the temporal sequence between the loss of the Golgi ribbon and loss of mTOR activation, we used the microtubule depolymerizing drug nocodazole to induce Golgi fragmentation because it provides the capacity to monitor events over a short 120 min period (Fig. 6). Golgi fragmentation was detected within 20 min of treatment with nocodazole and with similar kinetics to the loss of Golgi-localized mTOR (Fig. 6A,D), as revealed by quantitation of colocalization between mTOR and the TGN marker p230. There were also reduced levels of p-mTOR and a decrease in colocalization of p-mTOR and the Golgi marker p230 after 20 min of nocodazole treatment (Fig. 6B). Finally, and subsequent to the loss of Golgi-localized mTOR, LC3-positive autophagosomes were detected at the later time point of 120 min (Fig. 6C). Hence, these findings demonstrate an ordered sequence of events whereby loss of the Golgi ribbon is accompanied by dissociation of mTOR from Golgi membranes and is followed by induction of autophagy at a later time point. These findings demonstrate that mTOR is dissociated from the Golgi as a consequence of fragmentation of the Golgi ribbon.

Lysosomes are considered the major site for activation of mTOR by amino acids (Efeyan et al., 2014; Kim et al., 2008; Sancak et al., 2010, 2008; Zoncu et al., 2011). To investigate the relationship between Golgi and lysosome mTOR pools, we used parental SK-N-SH cells because the Golgi mTOR levels are particularly abundant and the two compartments readily distinguished. We asked whether mTOR could be recruited to Golgi membranes in the absence of a lysosomal mTOR pool. Bafilomyin A1 inhibits acidification of lysosomes and the inhibition of lysosomal acidification interferes with recruitment of mTOR to lysosomes (Zoncu et al., 2011). SK-N-SH cells were first treated with nocodazole to dissociate mTOR from the Golgi prior to balifomyin A1 treatment; then, nocodazole was withdrawn to allow the Golgi ribbon to re-form in the presence of balifomyin A1. Following nocodazole wash out, mTOR was detected on the re-formed perinuclear Golgi structures very rapidly (Fig. 7A). On the other hand, very little mTOR was detected on CD63-positive structures (Fig. 7A). As expected, treatment with nocodazole alone resulted in mTOR located to CD63-positive structures, but not to the Golgi fragments (Fig. 7A). Collectively, these data demonstrate that the recruitment of mTOR to the Golgi represents a pathway distinct from recruitment of the lysosomal mTOR pool.

To further investigate the relationship between the Golgi and lysosome mTOR pools, we treated HeLa-B6 cells with cycloheximide, which is known to activate mTOR via inhibition of protein synthesis and subsequent increase in the intracellular pool of amino acids (Watanabe-Asano et al., 2014). HeLa-B6 cells treated with cycloheximide showed an increase in p-mTOR associated with CD63-positive endosomes (Fig. 7B,C). In contrast, the Golgi mini stacks in HeLa-B6 cells were devoid of mTOR or p-mTOR following cycloheximide treatment (Fig. 7B,C). Therefore, mTOR failed to be either activated or recruited to the Golgi mini-stacks in HeLa-B6 cells following cycloheximide treatment, indicating that the Golgi pool of mTOR is functionally distinct from the lysosomal mTOR pool.

Loss of the Golgi ribbon has been reported in a range of neurodegenerative diseases (Sundaramoorthy et al., 2015). However, it is not known whether fragmentation of the Golgi contributes to neuronal degeneration or is a consequence of other pathogenic mechanisms (Gosavi and Gleeson, 2017). Based on the findings above, we determined the impact of overexpression of tau, a model for neurodegeneration induction (Liazoghli et al., 2005), on the Golgi ribbon, mTOR signaling and induction of autophagosomes in SK-N-SH cells in a time course study (Fig. 8). Transfection of neuronal SH-N-SK cells with GFP-tagged tau resulted in fragmentation of the Golgi in >50% of cells by 4 h post-transfection, whereas induction of LC3-positive punctate structures occurred at later time points (Fig. 8A,B). Moreover, tau-transfected cells showed a significant reduction in the Golgi localized active mTOR (p-mTOR) pool, compared with untransfected cells (Fig. 8C,D). In addition, there was a dramatic reduction in cytoplasmic p-S6 staining in cells positive for GPF–tau, indicating reduced mTORC1 activity (Fig. 8E). Therefore, compromised mTOR signaling and induction of autophagy associated with loss of the Golgi ribbon is relevant to understanding pathways of neurodegeneration.

It is well established that the ‘classical’ functions of the Golgi do not require the Golgi stacks to be fused into a ribbon structure, implying that there are additional, unidentified, functions associated with this architectural feature of the Golgi complex. Using a cell-based assay to probe the function of the Golgi ribbon structure, our findings demonstrate that loss of the Golgi ribbon results in the induction of autophagy by compromising mTOR signaling. Moreover, our findings reveal that the Golgi is a location for the regulation of mTOR signaling. Because the mTOR signaling pathway has a major role in regulating cell growth and metabolism, our findings reveal that the Golgi ribbon contributes to cell homeostasis.

We demonstrated that a modest overexpression of the TGN golgin GCC88 results in loss of the compact Golgi ribbon and dispersal of the organelle as mini-stacks throughout the cytoplasm. Evidence that the Golgi complex exists predominantly as individual mini-stacks scattered throughout the cytoplasm in these stable clones was revealed by confocal and super-resolution microscopy, PulSA flow cytometry, EM and tomography. Quantitative analysis of EM micrographs showed that the number of cisternae per stack was unaffected, but that the characteristic membrane continuities between the stacks of typical ribbons (Rambourg and Clermont, 1997) were absent; Golgi stacks were separated from each other spatially. Analysis of 3D tomograms showed that many of the dispersed stacks were connected by very long tubules, whereas other stacks were physically separated from each other. Arrangements of lengthy tubular connections between dispersed Golgi stacks represent a very different Golgi structure and morphology from the organization that is considered typical of the compact Golgi ribbon structure (Rambourg and Clermont, 1997). Therefore, it is clear that overexpression of GCC88 results in conversion of a typical compact Golgi ribbon to a much more dispersed array of Golgi mini-stacks, with maintained cis and trans polarity. Loss of the compact ribbon in HeLa-B6 cells is a result of intrinsic characteristics of the GCC88 golgin, rather than the GFP tag, as a similar phenotype was found with either Myc-tagged GCC88 or untagged GCC88. Moreover, the finding that knock down of GCC88 in HeLa-B6 or HeLa-C4 clones resulted in a longer ribbon structure than in parental cells demonstrated that the dose of GCC88 golgin regulates ribbon morphology. The generation of stable cells lacking a classical ribbon structure provides a system for directly analyzing the function of the Golgi ribbon.

The regulation of Golgi structure is mediated by interactions with both microtubules and actin. Notably, there are many actin-binding proteins that interact with the Golgi membranes and influence Golgi ribbon morphology (Egea et al., 2013; Gosavi and Gleeson, 2017). The mechanism by which GCC88 regulates the balance between Golgi ribbons and mini-stacks is currently under investigation and probably involves an actin-dependent process (Gosavi, Makhoul and Gleeson, unpublished observations).

Loss of the Golgi ribbon in stable HeLa cell clones was directly associated with the induction of autophagosomes. Autophagy induction was determined by an increase in LC3 staining punctate in HeLa-B6 and HeLa-C4 clones; an increase in the membrane-associated acylated form of LC3; a reduction in levels of the autophagosome cargo receptor p62; and the presence of structures with double membranes as identified by EM, features that collectively define the structure of autophagosomes. As the increase in autophagy in these clones was not the result of a block in the flux of autophagosomes, it seems likely that the loss of the Golgi ribbon promoted induction of autophagosomes biogenesis. In addition, there was no evidence of ER stress in the HeLa-B6 and HeLa C4 clones, nor major perturbation in anterograde or retrograde transport; nonetheless, there remains the possibility of other, unidentified, stress pathways that contribute to the phenotype in HeLa-B6 cells. Given the role of mTOR as a negative regulator of autophagy (Lamb et al., 2013; Rubinsztein et al., 2012; Wullschleger et al., 2006), this pathway was the obvious candidate. Total S6 protein levels were similar in HeLa and HeLa-B6 cells; however, there was a dramatic reduction in levels of p-S6, a well-defined downstream target of mTORC1, in HeLa-B6 lacking a Golgi ribbon compared with parental HeLa cells. Hence, autophagy is likely to be induced as a consequence of decreased mTOR activity. Stable HeLa-B6 cells do have a residual activity of mTORC1, which is likely to be the reason for the normal growth rates of HeLa-B6 and C4 clones. The origin of the membranes for the biogenesis of autophagosomes in this system in not known. The Golgi might contribute membranes to the biogenesis of autophagosomes; however, as the total mass of Golgi membranes does not appear to be significantly depleted in HeLa-B6 cells it is likely that membranes are also contributed from other organelles.

From comparison of the mTOR levels at the Golgi of parental HeLa and HeLa-B6 cells, we conclude that there is a considerable pool of mTOR at the Golgi that is dependent on the ribbon structure for recruitment. Golgi-localized mTOR was demonstrated using confocal microscopy and EM. In parental HeLa cells, mTOR showed substantial colocalization with Golgi markers (both TGN and cis-Golgi markers), as well as punctate staining throughout the cytoplasm; as expected, this staining was colocalized with the late endosome/lysosomal marker CD63. A signal for active p-mTOR was detected on the Golgi and Golgi-localized p-mTOR was sensitive to rapamycin, indicating that mTOR is activated at this site. In contrast, minimum mTOR and p-mTOR levels were detected on the scattered Golgi mini-stacks of HeLa-B6 cells; mTOR in these cells was restricted to endosomal structures. There was no change in the level of mTOR on CD63-positive late endosomes/lysosomes of HeLa-B6 cells compared with parental HeLa cells, and this lysosomal pool of mTOR is probably responsible for the residual mTOR activity detected in HeLa-B6 cells. mTOR and p-mTOR are also Golgi localized in other cell lines, including the neuroblastoma cell line SK-N-SH, where a significant proportion of the total pool of mTOR colocalized with the TGN marker. Late endosomes/lysosomes and the Golgi complex are spatially close together in the perinuclear region of the cell, which has the potential to confound these localization analyses. However, TGN marker p230 and the late endosome/lysosomal marker CD63 were readily distinguished, with minimum overlap in HeLa cells. Moreover, in SK-N-SH cells, the Golgi and late endosomes/lysosomes were spatially further separated than in parental HeLa cells, and mTOR was detected readily on the Golgi in these neuroblastoma cells. There have been previous reports of mTOR localized at the Golgi (Fan et al., 2016; Thomas et al., 2014; Drenan et al., 2004; Liu and Zheng, 2007), suggesting that the Golgi has a role in mTORC1 signaling. However, many of these earlier studies were limited by the insufficient organelle markers used and, in addition, the importance of the Golgi-localized mTOR to mTOR signaling was not clearly defined. Our findings confirm and extend these earlier reports and identify the Golgi ribbon as an important location for the functional regulation of mTOR activity.

An important question that arises from our observations on the localization of mTOR and p-mTOR is whether the Golgi pool of mTOR represents an independent site for the activation of mTOR or whether the lysosomal and Golgi locations act to co-ordinate mTOR activation. Following dissociation of mTOR from both the lysosomal and Golgi locations, by combined treatment with bafilomycin A1 and nocodazole, we observed that mTOR was recruited to the Golgi following nocodazole washout and re-formation of the ribbon structure, in the absence of a lysosomal mTOR pool. Hence, our analyses suggest that Golgi mTOR is independent from lysosomal mTOR activation and that both pathways are likely to be distinct. Nonetheless, we cannot rule out the possibility of coordinated mTOR activation between the two sites, possibly via membrane contact sites between the Golgi and lysosomes (Starling et al., 2016). The impact of loss of the Golgi ribbon on the integrity of these contact sites and mTOR signaling is worth exploring.

Although the late endosomes/lysosomes are generally considered the major mTOR hub (Saxton and Sabatini, 2017), the evidence that the lysosomal hub controls the full repertoire of outputs has not been examined. Indeed, mTORC1 has been reported at several cellular locations, although the relevance of many of these locations is not clear (Betz and Hall, 2013). Our findings strongly suggest that there are at least two distinct functional hubs that regulate mTORC1 signaling, lysosomes and the Golgi. Indeed, evidence has emerged over the last two years for a role of the secretory pathway, namely the ER and Golgi complex, in mTOR regulation. An RNAi screen of GTPases identified Rab1A, a small GTPase known to regulate ER to Golgi trafficking, as a regulator of mTORC1 (Thomas et al., 2014). Rab1A overexpression promotes mTOR signaling and growth, and Rab1A is highly expressed in human colorectal cancers (Thomas et al., 2014). The authors of this study propose that Rab1A coordinates the activation of mTOR on lysosomes by an unknown mechanism. However, it is also possible that the activation of mTOR occurs on the ER/Golgi in this system. A recent report by Fan et al. (2016) also suggested that mTORC1 is located at the Golgi; moreover, these authors showed that PAT4 (an amino acid transporter) influences the activation of mTOR, which suggests that differentially localized amino acid transporters at the lysosomes and Golgi contribute to the activation of mTORC1. mTOR signaling has also been shown to be modulated by membrane and cytoskeletal interactions mediated by the Golgi membrane tether GOLPH3 (Scott et al., 2009). However, whether the modulation of mTOR signaling by GOLPH3 is a direct or indirect effect of GOLPH3 is not clear from this previous study. Our work extends these earlier findings by identifying a major pool of mTOR associated with the Golgi ribbon that is activated at this site.

What are the implications of a relationship between Golgi fragmentation and autophagy, as revealed in our cell-based system? Of potential relevance are the findings from the Dennis group on metabolic regulation via hexoamine biosynthetic and Golgi N-glycan branching pathways (Boscher et al., 2011; Lau et al., 2007). Intriguingly, metabolism is linked with a pathway involving synthesis of branched glycans in the Golgi and galectin lattice formation at the plasma membrane, which in turn controls the cell surface lifespan of nutrient transporters (Lau et al., 2007). Furthermore, there is evidence that the Golgi architecture contributes to a sub-set of glycosylation events: disruption of the continuous Golgi ribbon, by depletion of the cis-Golgi matrix protein GM130, has been shown to result in a perturbation in the distribution of glycosyltransferases involved in O-glycosylation and alterations in glycosylation of glycoproteins expressed at the cell surface (Puthenveedu et al., 2006; Petrosyan et al., 2014). Defects associated with galectin-mediated responses that regulate cell growth have been reported for tumor cells with fragmented Golgi ribbons (Petrosyan et al., 2014). Also of potential relevance is the finding that cell starvation is associated with fragmentation of the Golgi ribbon (Takahashi et al., 2011). Hence, the events associated with formation of the Golgi ribbon might be integrated with the metabolic status of the cell and mTOR signaling.

Our findings also demonstrate that Golgi fragmentation can, in principle, directly contribute to pathophysiological conditions. Fragmentation of the Golgi ribbon has been reported in a variety of neurodegenerative diseases (Gonatas et al., 2006; Rabouille and Haase, 2015; Sundaramoorthy et al., 2015), including Alzheimer's disease, Huntington disease, amyotrophic lateral sclerosis and Parkinson's disease, in both animal models and human disease. Here, we used overexpression of tau in the neuroblastoma cells SK-N-SH as a cell-based system used to mimic neurodegeneration. We demonstrated that an early event following tau expression was Golgi fragmentation, which preceded compromised mTOR activity and subsequent induction of autophagy. The formation of tau aggregates occurs at later time points after transfection (Guo et al., 2016) and might also induce autophagy by other pathways. Major perturbations in Golgi morphology have also been reported to be associated with cancer (Farber-Katz et al., 2014; McKinnon and Mellor, 2017) and it is possible that perturbation of the Golgi ribbon also contributes to tumorigenesis.

Fig. 8F outlines a model for the activation of Golgi-localized mTOR, based on the findings reported in this paper. We propose that the Golgi complex acts as an mTOR signaling hub and that the molecular machinery associated with the compact Golgi ribbon recruits inactive mTOR from the cytosol. Rheb is a well-defined upstream activator of mTOR and has been previously reported to be located on both lysosomes and the Golgi complex (Buerger et al., 2006). Furthermore, Rheb is located on the Golgi ribbon in parental HeLa cells (Gosavi and Gleeson, unpublished observations); hence, the machinery for activation of mTOR is present on Golgi ribbons. We propose that the intact Golgi ribbon provides additional machinery, such as Rab1A (Thomas et al., 2014), to recruit inactive mTOR from the cytosol and that this recruitment machinery does not function after ribbon fragmentation. Significantly, conversion of the Golgi ribbon into Golgi mini-stacks results in a reduction in mTOR activity, thereby removing a negative regulation on autophagy. Our finding that a variety of different pathways that promote Golgi fragmentation lead to the induction of autophagy demonstrates that the integrity of the Golgi ribbon is crucial in regulation of the autophagy pathway.

Overall, our findings highlight the importance of the Golgi complex in higher-order functions and the relationship between autophagy, the mTOR signaling network and formation of the Golgi ribbon. The data presented herein are significant for an understanding of cell homeostasis and of various pathological conditions.

Mycoplasma-free authentic HeLa cells (Curie Institute, Paris) and SK-N-SH cells (ATCC, Manassas, VA) were maintained as semi-confluent monolayers in Dulbecco's modified Eagle's medium (DMEM) (Thermo Fisher Scientific, Australia) supplemented with 10% (v/v) fetal bovine serum (FBS) (Gibco-Life Technologies, Australia), 2 mM L-glutamine, 100 units/μl penicillin and 0.1% (w/v) streptomycin (complete DMEM) in a humidified 10% CO₂ atmosphere at 37°C. HeLa cell clones, B6 and C4, were maintained in the presence of 1 mg/ml G418 (Thermo Fisher Scientific, Australia). Transient transfections of fusion protein constructs were performed using FuGENE 6 transfection reagent (Promega, USA) or Lipofactamine 2000 (Thermo Fisher Scientific, Australia) according to the manufacturer's protocol.

A HeLa cell line stably expressing GFP–GCC88 was generated by transfecting parental HeLa cells with the GFP–GCC88 construct using FuGENE 6 (Promega, USA). Stably expressing cells were selected in complete DMEM with 1 mg/ml G418 until they formed single colonies. Isolated colonies were treated with trypsin and transferred to different wells. The resultant population was screened by immunofluorescence and GFP-positive clones further sorted. For cell sorting, cells were harvested in 0.05% (w/v) trypsin/0.02% (w/v) EDTA/PBS and resuspended in PBS; GFP-positive cells were purified using a BD FACSAria cell sorter. Three independent stable cell clones were maintained in complete DMEM with 1 mg/ml G418.

Constructs encoding GFP-full-length GCC88, Myc-tagged GCC88 and untagged-full-length GCC88 have been described previously (Luke et al., 2003). pRK5-EGFP–tau was a gift from Karen Ashe (Addgene plasmid #46904) (Hoover et al., 2010). E-cadherin-RFP encodes the full length human E-cadherin fused in frame at the C-terminus with red fluorescent protein (RFP) (Lock et al., 2005).

Rabbit polyclonal antibodies to human GCC88 and GCC185 have been described (Luke et al., 2003). Mouse monoclonal antibodies to human golgin-97 (#CDF4 A-21270,1:600) and GM130 (#610882, 1:600) were purchased from BD Biosciences (NSW, Australia). Human autoantibodies to p230 have been described (Kooy et al., 1992). Mouse monoclonal anti-GAPDH was purchased from Calbiochem (#MAP374, 1:5000). p-mTOR (phosphorylated S2448) antibody (#2971, 1:1000; Cell Signaling Technology, Australia) was a kind gift from Rick Pearson, Australia. Rabbit polyclonal antibodies to S6 ribosomal protein (5G10) (#2217, 1:1000), p-S6 ribosomal protein (phosphorylated Ser240/244) (D68F8, 1:1000), LC3B (D11XP) (#3868, 1:500) and rabbit monoclonal antibody to mTOR (7C10, #2983, 1:500) were purchased from Cell Signaling Technology, Australia. Mouse monoclonal antibodies to Rheb (ab#97896; 1:300), p62 (ab#56416, 1:1000), rabbit polyclonal antibodies to M6PR (cation independent) (ab#32815, 1:200) and Vps26 (ab#23892, 1:200) were purchased from Abcam, UK. Mouse monoclonal antibody to α-tubulin (clone DM1A, 1:5000) was obtained from Sigma-Aldrich. Mouse monoclonal antibodies to CD63 (MX-19.129.5, 1:1000) were obtained from Santa Cruz Biotechnology (sc-5275).

Secondary antibodies used for immunofluorescence were goat anti-rabbit IgG-Alexa Fluor 568 nm, goat anti-mouse IgG-Alexa Fluor 568 nm, goat anti-rabbit IgG-Alexa Fluor 488 nm, goat anti-mouse IgG-Alexa Fluor 488 nm, goat anti-mouse IgG-Alexa Fluor 647 nm, goat anti-rabbit IgG-Alexa Fluor 647 nm and goat anti-human Alexa Fluor 568 nm or 647 nm were from Life Technologies (Grand Island, NY, USA). All secondary antibodies were used at the dilution 1:500. Horseradish peroxidase-conjugated rabbit anti-goat Ig, horseradish peroxidase-conjugated sheep anti-rabbit Ig and anti-mouse Ig were from DAKO Corporation (Carpentaria, CA, USA).

Transfections with siRNA were performed using DharmaFECT1 siRNA transfection reagent (GE Lifesciences/Millennium Science, Victoria, Australia), according to manufacturer's instruction, for 72 h prior to analysis. Human GCC185 was targeted with siRNA (5′-GCUACUGUAACCUCUGAAU-3′) as described (Derby et al., 2007), human GCC88 was targeted with the specific siRNA (5′-GUCAGCAAUCUCAGGUAGA-3′) (Lieu et al., 2007) and human Vps26 siRNA (5′-CTCTATTAAGATGGAAGTG-3′) has been previously described (Arighi et al., 2004). All duplex siRNAs, including a control siRNA, were synthesized by Sigma-Aldrich (Australia). mTOR siRNA (catalogue number SASI_Hs01_00203145) was also purchased from Sigma-Aldrich.

Monolayers on coverslips were fixed with 4% (v/v) paraformaldehyde (PFA) for 15 min, followed by quenching in 50 mM NH₄Cl/PBS for 10 min. Cells were permeabilized in 0.1% TritonX-100 in PBS for 4 min and incubated in 5% FBS in PBS for 20 min to reduce nonspecific binding. Monolayers were incubated with primary and secondary conjugates as described (Kjer-Nielsen et al., 1999). For LC3B antibody, cells were fixed in 100% cold methanol for 8 min at −20°C and stained according to the manufacturer's protocol.

Images were acquired using a laser confocal scanning microscope (Leica LCS SP2 or SP8 confocal imaging system) using a 63×1.4 NA HCX PL APO CS oil immersion objective. GFP and Alexa Fluor 488 were excited with the 488-nm line of an argon laser, Alexa Fluor 568 with a 543-nm HeNe laser, Alexa Fluor 647 with a 633-nm HeNe laser and DAPI with a 405-nm UV laser. Images were collected sequentially for multicolor imaging. Fluorescence images for each experiment were collected using identical settings. Volocity 6.3 (Perkin-Elmer) imaging software was used for Pearson's correlation coefficient analysis in the entire cell with manual thresholding. Quantitation was carried out for the indicated number of cells at each time point.

Super-resolution 3D-SIM images were acquired using a Deltavision OMX Blaze V4 microscope (GE Healthcare) equipped with sCMOS camera (PCO EDGE) and a 60×1.42 NA PLAN APO N, UIS 2 BFP1 oil immersion objective. Lasers used included 100 mW output with 405, 488 or 568 nm wavelength. Emission filters used were as follows: DAPI-BP436/31, GFP-BP 528/48, Alexa 568-BP 609/37. SIM reconstructions were performed in SoftWoRx 6.1.1.

For analyses, cells were harvested in 5 mM (w/v) EDTA in PBS and single cell suspensions analyzed using a BD LSR Fortessa. Data were analyzed with FlowJo V10.

Cell monolayers were detached by incubation with 5 mM EDTA in PBS for 15 min at room temperature. For PulSA, cells were fixed with 4% PFA, quenched with 50 mM NH₄Cl, permeabilized with 0.1% Triton X-100 for 4 min in suspension and blocked in blocking solution (5% FBS in PBS). Cells were incubated with primary antibodies followed by addition of fluorophore-conjugated secondary antibodies, suspended in FACS buffer (2 mM EDTA in PBS) and analyzed at a medium flow rate in a LSRFortessa flow cytometer, equipped with 405, 488, 561 and 640 nm lasers (BD Biosciences). Approximately 10,000 events were collected, using a forward scatter threshold of 5000. For PulSA analysis, data were collected for pulse height, area and width parameters for each channel, as previously described (Chia et al., 2014; Toh et al., 2015).

Cells were lysed in RIPA buffer (1 mM Tris/Cl pH 7.5, 15 mM NaCl, 0.5 mM EDTA, 0.01% SDS, 0.1% Triton X-100 and 0.1% deoxycholate) containing protease inhibitors (Sigma-Aldrich, Australia). Extracts were resolved on SDS-PAGE using 4–12% or 12% NuPAGE gels. Proteins were transferred to PVDF membrane and probed with the indicated antibodies diluted in 10% (w/v) milk in TBST buffer (137 mM NaCl₂, 27 mM KCl, 20 mM Tris, 0.05% Tween-20, pH 7.6), either at room temperature for 1 h or overnight at 4°C, and then washed three times, each for 10 min, in TBST buffer. The PVDF membrane was then incubated with horseradish peroxidase-conjugated secondary antibodies, diluted in 10% (w/v) milk in TBST buffer for 1 h and washed three times in TBST buffer. Bound antibodies were detected by enhanced chemiluminescence (Amersham, GE Healthcare, NSW, Australia).

Cells were fixed by adding freshly prepared 2% w/v PFA and 1.5% glutaraldehyde (Electron Microscopy Sciences) in 0.1 M sodium cacodylate buffer (pH 7.4) to an equal volume of culture medium for 5 min, followed by post-fixation in the same buffer at 4°C overnight. Ultrathin sections of 100 nm were cut using a Leica Ultramicrotome UC7 and stained as described (Deerinck et al., 2010). 3D STEM tomography was performed on 600 and 900 nm thick resin sections on a FEI Tecnai F30 at 300 kV with a HAADF detector. Alignment of the tilt series, tomogram reconstruction and segmentation were performed using the IMOD package (Kremer et al., 1996).

Monolayers of HeLa cells were fixed with freshly prepared 4% (w/v) PFA in 0.1 M phosphate buffer (pH 7.4) and an equal volume of culture medium for 5 min, followed by post-fixation in 4% (w/v PFA) at 4°C overnight. Ultrathin cryosectioning and immunogold labeling were performed as previously described (Slot and Geuze, 2007). Briefly, ultrathin cryosections were labeled with anti-mTOR antibody followed by incubation with Protein A gold (5 nm size; CMC, Utrecht Medical Centre, The Netherlands).

Cells were treated with drugs diluted in complete DMEM for the indicated time periods and fixed in 4% PFA before proceeding for immunostaining. DMSO (0.05–0.2%) was used as a carrier control for all drug treatments. Cells were treated with 10 µM nocodazole (Sigma) at 37°C for up to 2 h, 200 nM bafilomycin A1 (Calbiochem) for 4 h at 37°C, 5 µM rapamycin (Sigma) for 4 h at 37°C, 50 µg/ml cycloheximide (Sigma) for 2 h at 37°C or 10 µg/ml tunicamycin (Sigma) for 4 h at 37°C.

The ERSE reporter is a mixture of a luciferase construct under the control of an ER stress response element and a constitutively expressing renilla luciferase construct (40:1). The ERSE-responsive luciferase construct encodes a firefly luciferase reporter gene under the control of a minimal CMV promoter and tandem repeats of the ERSE transcriptional response element (Cignal ERSE reporter kit, Qiagen). Cells were transfected with the dual luciferase ERSE reporter construct for 20 h followed by incubation with DMSO carrier control or 10 µg/ml tunicamycin (Sigma) for 4 h to induce ER stress. Cells were then lysed in passive lysis buffer provided in the kit and luciferase activity measured according to the manufacturer's protocol (Dual-Luciferase Reporter Assay System, Promega). Normalized luciferase activity was calculated by dividing firefly luciferase activity by renilla luciferase activity, before determination of the fold change relative to values obtained for HeLa and HeLa-B6 cells in three independent experiments.

HeLa cells were transfected with E-cadherin-RFP for 24 h. Cells were harvested with 5 mM (w/v) EDTA in PBS at room temperature for 15 min, washed in C-DMEM and resuspended in FACS buffer (2 mM EDTA in PBS) supplemented with 5% FBS for 15 min. For detection of cell surface E-cadherin, cells were then stained with mouse anti-E-cadherin antibodies diluted in FACS buffer containing 5% FBS at 4°C for 40 min and incubated with anti-mouse Alexa Fluor 647 (Invitrogen) diluted in FACS buffer containing 5% FBS for 30 min. Cells were then washed twice in FACS buffer, fixed in 4% (w/v) PFA at room temperature for 15 min. Cells were then analyzed using a BD LSRFortessa flow cytometer as described above.

Quantitation of the colocalization was performed using Volocity imaging software (Perkin Elmer, UK). Quantitation was carried out for the indicated number of cells at each time point. All analyses included samples from two or more independent experiments. Data were expressed as the mean±s.e.m. or mean±s.d., as indicated, and analyzed by an unpaired, two-tailed, Student's t-test. P<0.05 (*) was considered significant, P<0.01 (**) was highly significant and P<0.001 (***) and P<0.0001 (****) were very highly significant. Absence of a P-value indicates that the differences were not significant (ns).

Confocal and 3D-SIM microscopy was performed at the Biological Optical Microscopy Platform (BOMP) at the University of Melbourne. We acknowledge Danny Hatters and Greg Moseley (Bio21 Institute) for reagents and Diana Stojanovski and Malcolm McConville (Bio21 Institute) and Rob Parton (IMB, University of Queensland) for invaluable suggestions.