Positive feedback between Golgi membranes, microtubules and ER exit sites directs de novo biogenesis of the Golgi

How organelles are formed and what determines their shape, size and intracellular position are fundamental questions in cell biology. Two alternative mechanisms can play a role in the biogenesis of a cellular organelle. The first one is to copy a template acquired early during evolution. This model is generally accepted for the endoplasmic reticulum (ER) and mitochondria, which act as templates for their own biogenesis and growth before their mitotic partitioning in daughter cells (Palade, 1983; Warren and Wickner, 1996). The second mechanism is self-organization, which is a well-established paradigm in cell biology from the molecular to the organismal level (Gerstman and Chapagain, 2005; Misteli, 2001; Wennekamp et al., 2013). Self-organization is defined as the capacity of an organelle to acquire its own shape de novo as a result of the interaction of its components. At the subcellular level, this model has been suggested to apply, for instance, to the organization and architecture of the cytoskeleton (Nédélec et al., 1997; Vignaud et al., 2012).

How the Golgi complex acquires and maintains its structure is not completely clear. This organelle is evolutionarily conserved in all eukaryotic cells (Dacks et al., 2009). Although its structure has varied significantly during evolution, its basic molecular composition has been very well conserved, together with its fundamental role in the secretory pathway. In most eukaryotic organisms, the Golgi complex is composed of stacked flattened cisternae, and in mammalian cells it has an additional complexity level due to the lateral linking of its stacked units in the perinuclear area to form a ribbon (Wei and Seemann, 2010). Despite its complex organization, the Golgi complex is a very dynamic organelle (Lippincott-Schwartz and Zaal, 2000), and it undergoes dramatic structural changes during mitosis (Wei and Seemann, 2009). During interphase, it receives material from the ER and the endosomal system and gives rise to several anterograde and retrograde transport pathways. Nonetheless, an overall stable structure is correctly generated as a result of these highly dynamic events. Moreover, the morphology and size of the Golgi complex depend on the balanced flux of membranes that pass through it (Glick, 2000; Trucco et al., 2004; Guo and Linstedt, 2006; Griffiths et al., 1989), and the input of material from the ER is crucial for its homeostasis (Zaal et al., 1999; Storrie et al., 1998). These observations make the Golgi complex a good candidate for a case of a self-organizing organelle. The self-organization model postulates that the Golgi complex should be able to form de novo as a result of the fusion of membranes derived from the ER, and many independent studies suggest that this could be the case (Lippincott-Schwartz et al., 1989; Miles et al., 2001; Ward et al., 2001; Puri and Linstedt, 2003; Kasap et al., 2004; Altan-Bonnet et al., 2006; Tängemo et al., 2011). Nonetheless, other studies have reported the possibility of a template-mediated Golgi assembly, implicating the Golgi matrix proteins in functioning as seeds or templates for biogenesis (Jesch et al., 2001; Shorter and Warren, 2002; Seemann et al., 2002; Axelsson and Warren, 2004). To what extent Golgi biogenesis can be considered a self-assembly process is therefore still not clear (Lowe and Barr, 2007; Wang and Seemann, 2011).

The ER exit sites (ERES) are the sites of production of post-ER carriers directed to the Golgi complex – ribosome-free ER subdomains that produce coat protein complex II (COPII)-coated transport vesicles. In cultured mammalian cells, ERES are distributed throughout the entire cytoplasm, with an accumulation in the juxta-nuclear area close to the Golgi complex (Hammond and Glick, 2000; Budnik and Stephens, 2009). During mitosis, coincident with the Golgi breakdown, the recruitment of COPII components is dramatically reduced and ER export is blocked (Prescott et al., 2001; Kano et al., 2004). Upon mitotic exit, when reassembly of the Golgi complex requires the restoration of an efficient ER export process, COPII proteins are again efficiently recruited to the ER membrane (Hammond and Glick, 2000; Altan-Bonnet et al., 2006; Budnik and Stephens, 2009). How they achieve their interphase distribution with an accumulation in proximity to the juxta-nuclear Golgi complex is not clear. It has been proposed that ERES distribution depends on the scaffolding function of Sec16, a peripheral membrane protein (Connerly et al., 2005; Ivan et al., 2008; Hughes et al., 2009). This view has been recently challenged, and a self-organizing process underlying the early secretory pathway has been suggested, with a putative role of Golgi-localized tethers in the localization of ERES (Glick, 2014).

Although the secretory pathway has been largely explored in the last decades and our knowledge of the molecular details is continuously progressing, how dynamic processes such as ER export, Golgi formation and microtubule network organization are integrated to give rise to and maintain the organization of this membrane system is not clear. To address this question, we made use of a recently developed technique to deplete the Golgi complex from living cells by laser nanosurgery and monitor its subsequent biogenesis (Tängemo et al., 2011; Ronchi et al., 2012; Ronchi and Pepperkok, 2013). This approach has provided direct evidence for Golgi de novo biogenesis in mammalian cells (Tängemo et al., 2011). Here, we have further characterized this process by using quantitative time-lapse imaging correlated with electron microscopy. Our data demonstrate that the mechanisms underlying Golgi de novo biogenesis and the organization of the early secretory pathway are governed by self-organizing principles.

The Golgi complex can be de novo synthesized after its depletion from BSC1 cells (a renal epithelial cell line of African green monkey origin) by laser nanosurgery (Tängemo et al., 2011). To gain a better insight into this process, we quantitatively analyzed the dynamics of the YFP-tagged Golgi enzyme GalNAc-T2–YFP (YT2; Axelsson and Warren, 2004) in karyoplasts with high temporal resolution time-lapse microscopy and correlative light and electron microscopy. Based on time-lapse analysis of at least ten cells, three distinct phases of Golgi biogenesis could be observed (Fig. 1A; supplementary material Movie 1). In phase 1, the karyoplasts synthesized the YFP-tagged Golgi marker, which increased in intensity in the ER over time (Fig. 2A), and small post-ER carriers were occasionally produced (Fig. 1A,B; Fig. 2A). These carriers moved on linear tracks, consistent with microtubule-mediated transport, but without any preferential directionality (Fig. 1B,F). These structures were not substantially accumulated over time in the cell, and they disappeared on average 16 min after their formation (Fig. 1E), most likely owing to their reabsorption into the ER and/or degradation. Quantification showed that, despite the increase in the total amount of YFP-tagged Golgi marker synthesized after laser nanosurgery, the total fluorescence of the observed small post-ER carriers remained very low in phase 1 compared with the subsequent phases (Fig. 2A; supplementary material Fig. S1A). In order to compare the time-lapse analysis curves of the karyoplasts, we calculated the parameter Δt (Fig. 2A,C), which represents a measure of the difference between the start of protein synthesis and net ER export. This parameter displayed a median delay for secretion of 354 min, demonstrating that post-ER material accumulation is inhibited after Golgi removal in most cells. As a result of this low efficiency of net ER export, YT2 accumulated in the ER (Fig. 1A; Fig. 2E). By contrast, when the Golgi complex was only partially depleted by laser nanosurgery the total fluorescence of post-ER structures increased with the same kinetics as the total amount of the marker-specific cellular fluorescence (Fig. 2B), and the median Δt was only 6 min. No accumulation of the fluorescent marker in the ER was observed under these conditions (Fig. 2E). The observed inhibition of ER exit during phase 1 did not appear to be specific for the Golgi marker YT2, as the cargo protein GPI–CFP (Keller et al., 2001) displayed the same behavior after laser nanosurgery (supplementary material Fig. S1B). However, this ER export inhibition does not represent a general inhibition of intracellular trafficking as the internalization of transferrin and EGF occurred with no changes compared with that observed in unperturbed control cells (supplementary material Fig. S2A,B).

At the transition from phase 1 to phase 2 (656±173 min after nanosurgery; ±s.d.) the small post-ER carriers occasionally collided and fused with each other. This resulted in the formation of more stable and larger structures (Fig. 1C). After their formation, these structures grew further in size and became predominantly located in the juxta-nuclear area during phase 2. Interestingly, they were able to frequently attract and absorb smaller carriers even from the cell periphery (Fig. 1D). In contrast to the small post-ER carriers observed in phase 1, phase 2 small carriers showed directional movements from the cell periphery towards the cell nucleus (Fig. 1D,F). Quantification showed that the total fluorescence of post-ER structures rapidly increased within 171±88 min in phase 2, from 20% to 60% of the maximum intensity reached during the experiment (Fig. 2A; supplementary material Fig. S1A). The onset of this burst of Golgi precursor synthesis was not dependent on the amount of fluorescent Golgi marker expressed (supplementary material Fig. S1B). To compare the kinetics of post-ER material accumulation during phase 2 in the ten karyoplasts analysed, we measured the average slope of the secretion curves in phase 2 and normalized for the slope of the protein synthesis curve in the same interval (Fig. 2D). In cells where the Golgi was not entirely removed the median ratio was 1.05, whereas after complete Golgi removal the cells displayed a median ratio value of 1.95, indicating that, in the latter case, the kinetics of net ER export in phase 2 were faster than the synthesis.

Finally, in phase 3 the larger structures that formed during phase 2 clustered in a single juxta-nuclear location (Fig. 1A). Interestingly, the position of the new Golgi complex in phase 3 did not appear to be biased towards one side of the nucleus (supplementary material Fig. S1D), suggesting that there is no memory of where the organelle had been located prior to the nanosurgery.

In order to gain ultrastructural details of Golgi precursors during the biogenesis process, we used correlative light and electron microscopy (CLEM; supplementary material Fig. S3). In agreement with our previously published results (Tängemo et al., 2011), phase 1 karyoplasts showed no evidence of stacked Golgi-like structures (Fig. 3). By contrast, we could identify ERES, which appeared to be structurally normal (Fig. 3A), and occasionally we observed electron-dense accumulations in the ER lumen (Fig. 3B). At the transition between phase 1 and 2 electron microscopy analysis showed small clusters of vesicles and tubules (Fig. 3C).

Interestingly, analysis of phase 2 cells by CLEM revealed that the larger structures forming during this phase were generally not showing a typical Golgi-like ultrastructure (Fig. 3D). They represented clusters of juxtaposed membrane-bounded structures, which appeared to be compact and convoluted (Fig. 3D–F). Some structures in the cluster started to flatten and acquired cisternae-like shapes (Fig. 3E). The regular spacing and the presence of electron-dense material in the lumen of these flattened areas was indicative of the presence of a molecular bridge between the two membranes (Fig. 3E,F). Furthermore, the close association and the constant spacing between the peripheral membranes of adjacent structures suggested the initiation of lateral stacking (Fig. 3E,F). To gain better insight into the ultrastructure of Golgi precursors during phase 2 of Golgi biogenesis, we used electron tomography. Three-dimensional (3D) reconstruction showed a convoluted network of membranes forming interconnected structures (Fig. 3G–I; supplementary material Movie 2). Most of these structures had cavities and cisterna-like features. The precursors did not show any connection with the neighboring ER, confirming their post-ER nature. Only occasionally (1/9) did phase 2 karyoplasts show ministack-like structures, suggesting that this is the crucial phase for the maturation of the Golgi into a stacked structure. Accordingly, phase 3 cells presented clusters of Golgi ministacks (Fig. 3J–L). Nonetheless, in contrast to the Golgi complex of control cells (Fig. 3M,N), phase 3 structures did not form complete Golgi ribbons within 24 h after nanosurgery. Taken together, these electron microscopy analyses suggest that the building of a stacked Golgi occurs by remodeling of the precursor interconnected membrane structures observed during phase 2.

Next, we set out to characterize the molecular composition of the Golgi precursors. We have previously shown that our nanosurgery procedure strongly depletes many Golgi-associated proteins (Tängemo et al., 2011). We performed immunolabeling experiments with karyoplasts fixed in the different phases of biogenesis in order to monitor the presence of the Golgi proteins GM130, GRASP55, GRASP65, giantin, AKAP450 and CLASP2 (Fig. 4). As expected, in phase 1, very low levels of all the tested proteins were present. Interestingly, in phase 2, the levels of GM130, GRASPs and CLASP2 in Golgi precursors increased to ∼10–15% of their amount in neighboring control cells, whereas the amount of giantin and AKAP450 remained low, suggesting that different proteins associate with Golgi precursors at different time-points and might contribute to different biogenesis steps. In phase 3 cells, Golgi elements contained 16–30% of control cell levels, proving that this amount is sufficient to build stacked structures.

Next, we set out to obtain possible explanations for the bimodal kinetics of ER export during phases 1 and 2 of Golgi biogenesis. To this end, we analyzed ERES in karyoplasts in the different phases of the process and quantified their features. Immunofluorescence experiments revealed that ERES were present throughout all phases of Golgi de novo biogenesis and the amounts of the COPII component Sec31A per cell area remained largely unchanged throughout all phases (Fig. 5A,B). By contrast, the number of ERES in early and late phase 1 cells was drastically reduced to 22±5% and 18±4% (±s.e.m.,) of the number observed in control cells, respectively (Fig. 5C). Although these data seem to be contradictory, they could be explained by the existence of a large non-membrane-associated fraction of the protein at steady state, consistent with previously reported data (Koreishi et al., 2013). The reduction in the number of ERES could provide a plausible explanation for the ER exit inhibition we observed during phase 1 of Golgi de novo biogenesis. Consistent with this, during phase 2 and phase 3, when ER exit was apparently restored, the number of ERES increased again to 63±34% and 94±40% of the number observed in control cells, respectively (Fig. 5C).

Not only was the number of ERES altered during the phases of Golgi de novo biogenesis, but their spatial distribution was also affected. In contrast to control cells, where an accumulation of Sec31A-positive ERES in proximity to the Golgi complex could be observed (Fig. 5A,D), in phase 1 cells ERES were distributed throughout the karyoplasts without any apparent accumulation in a juxta-nuclear region (Fig. 5A,D). A similar reduction in polarized distribution was observed for the early endosomal marker EEA1 (supplementary material Fig. S2C,D). A likely explanation for these observations is that ERES and endosomes that initially localize in proximity to the Golgi complex might be depleted together with the Golgi during laser nanosurgery. During phase 2, ERES started to accumulate in proximity to newly forming Golgi precursors and a full restoration of the ERES distribution compared with that of control cells was achieved during phase 3 (Fig. 5A,D), despite the lack of the centrosome (Tängemo et al., 2011). By contrast, a full restoration of the early endosome distribution in proximity to newly forming Golgi membranes could not be observed even for phase 3 cells (supplementary material Fig. S2C,D). We further tested the behavior of Sec16, an ERES component believed to be involved in the early steps of ERES formation (Ivan et al., 2008). Colocalization analysis between Sec16 and Sec31 showed a good overlap of the two proteins during all phases of Golgi biogenesis (Pearson's R values between 0.77 and 0.90, comparable to those of control cells in the same image; supplementary material Fig. S4B). We did not observe Sec16-positive puncta that were negative for Sec31 during phase 1 (supplementary material Fig. S4A), suggesting that the bimodal kinetics observed for the formation of Sec31-positive structures are followed by Sec16 as well.

Next, we investigated the role of the microtubule network in the different phases of Golgi biogenesis. After laser nanosurgery, the microtubule network showed a lack of central organization, which, by contrast, could be observed in control cells or in phase 3 karyoplasts (Fig. 6A). This result could most likely be explained by the depletion of the centrosome and, thus, the lack of centrosomal microtubule nucleation in karyoplasts. To test this hypothesis, cells were treated with 10 µM nocodazole for 30 min to depolymerize all dynamic microtubules. After this period only a few ‘stable’ microtubules remained in control cells and karyoplasts (data not shown). Interestingly, nocodazole washout, which is followed by the centrosomal nucleation of microtubules in control cells shortly after the washout (Fig. 6B,C), showed that even at 10 min after drug removal phase 1 cells did not nucleate microtubules (Fig. 6B). Consistent with these observations, immunostaining of EB1, a protein that is associated with the plus-end tips of growing microtubules (Galjart, 2010), was strongly reduced in phase 1 cells (Fig. 7A). The observation that a microtubule network exists at steady state in phase 1 karyoplasts (Fig. 6A) appears to be contradictory to our findings that microtubule growth is reduced in these cells (Fig. 6B,C; Fig. 7A). One explanation might be that during phase 1 microtubules are stabilized. In support of this hypothesis, microtubules in phase 1 cells showed an increased level of acetylation (Fig. 7B,C), which is a typical characteristic of stable microtubules (Westermann and Weber, 2003), compared with that observed in phase 3 and control cells (Fig. 7B,C). Nocodazole washout in phase 2 and phase 3 cells was followed by microtubule regrowth within 10 min after drug removal, although this was less efficient compared with regrowth in control cells (Fig. 6B). A central microtubule organization from the area of the newly synthesized juxta-nuclear Golgi could be observed in phase 3 cells (Fig. 6B). These observations suggest that Golgi precursors have the capacity to nucleate microtubules during phase 2 and 3. In agreement with this idea, analysis of cells at 3 min after nocodazole washout showed that phase 2 Golgi precursors were frequently associated with short microtubules (Fig. 6C). Microtubule distribution appeared to be significantly biased towards the association with Golgi elements (Fig. 6D). The association of microtubules with Golgi structures was even more pronounced in phase 3, although, at this stage, it was not yet comparable to the efficiency of microtubule nucleation of the centrosome in control cells (Fig. 6C). EB1 staining of karyoplasts in phases 2 and 3 after nocodazole washout further confirmed that the microtubules associated with Golgi precursors had been nucleated at these sites (Fig. 6E).

The data presented thus far are consistent with a model of Golgi de novo biogenesis that is based on positive-feedback loops involving microtubule network organization and ERES biogenesis that lead to the self-assembly of post-ER material and finally to the formation of a juxta-nuclear Golgi complex (Fig. 8A). On one hand, the inhibition of ER export during phase 1 could be attributed to the observed reduction in the number of ERES (Fig. 5). On the other hand, we propose that the disorganization of the microtubule network in phase 1 cells has a negative effect on the probability of small post-ER carrier fusion, which is a requirement for the formation of Golgi precursors. Indeed, under these conditions, the formation of Golgi precursors would only occur by stochastic fusion of small post-ER carriers. Nucleation of microtubules from Golgi precursors would facilitate the recruitment of small post-ER carriers and thus drastically enhance precursor growth. In turn, this growth could have a stimulatory effect on ERES biogenesis, which would additionally contribute to the observed burst of accumulation of post-ER material during phase 2.

In this model, a crucial step appears to be the fusion of small post-ER carriers, which is inhibited by the lack of organization of the microtubule network and can be drastically accelerated by the nucleation of microtubules by Golgi precursors. Therefore, the model would predict that experimental conditions that enhance the chance of small post-ER carrier fusion should accelerate Golgi de novo biogenesis, as well as ERES biogenesis. To test this hypothesis, we took advantage of the fact that, in nocodazole-treated cells, small post-ER carrier movement is impaired and the carriers are not efficiently moved away from their site of production (Presley et al., 1997; Scales et al., 1997), therefore enhancing the possibility that the small post-ER carriers will fuse with one another. Time-lapse analysis of nocodazole-treated karyoplasts showed that accumulation of post-ER membrane structures occurred much earlier compared with that of non-treated karyoplasts (Fig. 8B). In contrast to non-treated karyoplasts (Fig. 2A), the total fluorescence of post-ER structures increased with similar kinetics to those of the total cell fluorescence of the marker, and an inhibition of post-ER carrier formation was not observed (Fig. 8C; Fig. 2C,D). In addition, accumulation of the Golgi marker YT2 in the ER was not observed after nocodazole addition (Fig. 2E). Moreover, the amount of Sec31 associated with ERES appeared to increase earlier after laser nanosurgery in nocodazole-treated cells compared with non-treated karyoplasts (Fig. 8D). Electron microscopy analysis of nocodazole-treated karyoplasts showed that the post-ER membrane structures forming early after laser nanosurgery had Golgi-like features. Membranes appeared to be stacked and started to elongate into cisternae, although proper ministacks with flattened cisternae could be identified only at later time-points (Fig. 8E). These data demonstrate that if enough post-ER material is accumulated in a small enough area, it is able to efficiently self-organize, induce ERES biogenesis and acquire Golgi-like features.

Here, we have characterized in detail the organization and dynamics of the early secretory pathway during Golgi de novo biogenesis, by using our recently developed laser nanosurgery approach to deplete cells of their Golgi complex (Tängemo et al., 2011). In summary, quantitative time-lapse microscopy in combination with electron microscopy analyses revealed three phases of Golgi de novo biogenesis after laser nanosurgery. Phase 1 is characterized by a net inhibition of ER exit lasting for several hours before directed transport is restored and Golgi precursor growth is accelerated exponentially during phase 2. This is followed by the formation of proper Golgi ministacks that cluster in a juxta-nuclear position during phase 3.

The observed inhibition of net ER exit during phase 1 is unlikely to be the effect of laser-induced stress, as in experiments where the Golgi was only partially removed or in karyoplasts treated with nocodazole the restoration of the secretory traffic was much faster. Moreover, endocytosis of transferrin and epidermal growth factor occurred with the same efficiency as observed in unperturbed control cells under our experimental conditions. At least two observations could account for the net inhibition of ER exit in phase 1. First, the number of ERES was drastically reduced, most likely due to their depletion during laser nanosurgery. Second, small post-ER carriers occasionally formed in phase 1 and did not fuse with one another, disappearing on average 16 min after their formation as a result of their degradation or ER reabsorption.

In phase 2, which is typically much shorter than phase 1, small post-ER carriers occasionally meet and fuse with one another to form larger post-ER structures, an event that coincides with the exponential increase in the formation of further stable post-ER structures and Golgi precursors. Several of our observations indicate what factors might contribute to this exponential increase in post-ER structures. First, in contrast to smaller post-ER carriers forming in phase 1, the larger phase 2 structures are able to nucleate microtubules, as demonstrated in nocodazole washout experiments. The microtubules associated with phase 2 post-ER structures most likely provide the tracks for the small post-ER carriers to move towards and be efficiently absorbed by the larger structures, which can consequently further grow in size. The presence during phase 1 of very low levels of proteins with microtubule nucleation capability, such as AKAP450 (Rivero et al., 2009) or CLASP2 (Efimov et al., 2007), could explain the inefficiency of phase 1 structures to nucleate microtubules. Interestingly, during phase 2, the levels of Golgi-associated AKAP450 and CLASP2 are consistent with a prominent function of the latter in microtubule nucleation in this context.

Another process that could contribute to the exponential growth of post-ER material is the formation of new ERES, which increase in number about threefold during phase 2 and thus can significantly contribute to the formation of new small post-ER carriers. Interestingly, at the same time, ERES accumulate in the vicinity of the large post-ER structures. Therefore, the vesicles that they produce are able to fuse immediately and efficiently with the large post-ER structures and increase their material, without the need for long-distance microtubule-mediated transport. What are the mechanisms that induce ERES biogenesis in phase 2 karyoplasts? One possibility might be that secretory cargo and Golgi material accumulating in the ER during phase 1 (Fig. 2E) induce ERES formation, in agreement with previous findings that secretory cargo has a stimulatory activity on ER exit (Aridor et al., 1999; Forster et al., 2006; Simpson et al., 2006; Guo and Linstedt, 2006). However, we consider this possibility unlikely, as we could not observe any correlation between the amount of the YFP-tagged Golgi marker accumulating in the ER after laser nanosurgery and the onset of phase 2 (supplementary material Fig. S1C). Moreover, the number of ERES does not increase for the entire duration of phase 1 (Fig. 5C), although a continuous increase in the amount of YT2 in the ER was observed during this phase. The correlation of Golgi and ERES biogenesis suggests a coupling of Golgi and ERES formation by means of a positive-feedback loop (Fig. 8A), and therefore implicates a possible stimulating role of the Golgi precursors on ER exit and ERES biogenesis. Consistent with this hypothesis, when nocodazole accelerates Golgi biogenesis, ERES biogenesis is accelerated as well (Fig. 8D). One possibility for how this Golgi–ERES coupling could be mediated might be that Golgi precursors stabilize ERES in their vicinity and vice versa. This could be achieved by the Golgi membrane signaling to neighboring ERES through the cytosol, by retrograde traffic from Golgi precursors that could locally increase the concentration of recycling proteins in the ER (Hammond and Glick, 2000) and/or by a direct Golgi–ERES tethering, which is a generally accepted model in plant cells (daSilva et al., 2004; Staehelin and Kang, 2008; Sparkes et al., 2009; Glick, 2014). Our data propose a model of Golgi biogenesis wherein the stochastic coalescence of ER-derived carriers into larger structures triggers their capability to nucleate microtubules and stimulate ERES biogenesis, processes that, in turn, fuel the formation of the Golgi.

The ultrastructural characterization of the intermediates of Golgi biogenesis showed that the final shape of the Golgi complex is acquired in phase 3 by the flattening and stacking of a pool of convoluted interconnected membranes that had accumulated during phase 2. Molecular characterization of phase 2 Golgi precursors showed that golgins and GRASPs are present on these structures, which could provide the adhesive forces necessary for their lateral association and fusion. However, the observation that, in nocodazole-treated karyoplasts, partially stacked Golgi precursors can be observed much earlier after laser nanosurgery (as early as 3.5 h after nanosurgery) compared to those of non-treated karyoplasts (on average 14 h) suggests that the small post-ER carriers observed in phase 1 have the intrinsic capability to self-assemble into partially stacked Golgi precursors when they coalesce. These data suggest that very low amounts of these proteins are sufficient to promote membrane tethering and stacking. By contrast, the final flattening of the partially stacked Golgi precursors and morphological transformation into proper Golgi ministacks occurs with similar kinetics after nanosurgery as in non-treated control karyoplasts, suggesting the need for the synthesis of crucial factor(s) for this process to occur. Giantin and AKAP450 could be good candidates for such factors, as they become associated with the Golgi elements only in phase 3. Taken together, our data demonstrate that the processes of Golgi biogenesis, ERES biogenesis and microtubule network organization are integrated in positive-feedback loops that guarantee the efficient formation and maintenance of the structure and function of the early secretory pathway.

The antibodies used were as follows: mouse anti-GM130 (BD Biosciences), rabbit anti-AKAP450 (Sigma), rabbit anti-giantin (Abcam), mouse anti-Sec31A (BD Biosciences), rabbit anti-Sec16A (Bethyl Laboratories), mouse anti-α-tubulin (Neomarkers), rabbit anti-β-tubulin (Abcam), mouse anti-acetylated-tubulin (Sigma), mouse anti-EB1 (BD Biosciences), rat anti-EB1 (Abcam) and mouse anti-EEA1 (BD Biosciences). Sheep anti-GRASP65, sheep anti-GRASP55 and rabbit anti-CLASP2 were gifts of Martin Lowe (University of Manchester, UK), Francis Barr (University of Oxford, UK) and Irina Kaverina (Vanderbilt School of Medicine, Nashville, TN), respectively. All cell culture reagents were purchased from Gibco.

The patterning was performed as described previously (Pouthas et al., 2008; Tängemo et al., 2011) from a photolithography mask produced by GeSim (Dresden, Germany). Laser nanosurgery was performed with a pulsed 355-nm laser (DPSL-355/14, Rapp Optoelectronic, Hamburg, Germany) coupled in an Olympus CellR widefield microscope. The pulse duration of the laser source was 1 ns with a repetition rate up to 200 Hz. The laser beam was focused and moved on the sample with an integrated UGA-40 scanner (Rapp OptoElectronic). All nanosurgery experiments were performed with a 60× water-immersion objective.

Live imaging of karyoplasts after nanosurgery was performed with the multi-position time-lapse microscope Cell^∧R (Olympus). After immunofluorescence experiments, samples were imaged with LSM510 (Zeiss), SP2 or SP5 laser scanning confocal microscopes (Leica Microsystems).

High-temporal-resolution time-lapse movies (15-s interval between acquisitions) were analyzed with ImageJ software, and the movements of carriers were tracked (‘Manual tracking’ plugin). The centre of the nucleus was set as the reference point. The 2D net movement of the trajectories during the time-lapses (5 min) was determined, and the component parallel to the cell axis was calculated. Movements were defined as positive when moving away from the reference point and negative when moving towards it.

For automated object (post-ER structure) detection in fluorescence images of Golgi reformation we developed a Matlab-based image-processing software (Morphoquant; Gonzalez et al., 2004). The software facilitates automated detection of locally bright structures of both small dot-like or large arbitrary shape. The segmentation can be adapted by the following parameters: objects, diameter and threshold; noise suppression, minimal area of an object. For any choice of parameters, the resulting object detection can be visually inspected on an image of choice. Once optimal parameters have been identified, the software runs automatically on multiple data sets (for background subtraction and measurement of total cell fluorescence the user provides additional images with background and cell masks). For each cell and each movie frame, the software outputs the integrated optical density of all objects [object fluorescence intensity (FI)] as well as of the entire cell (total cell fluorescence intensity) (Fig. 2).

To determine the Sec31 and EEA1 clustering index (Fig. 5D and supplementary material Fig. S2D, respectively), images acquired and processed as described above were treated as follows: (1) from the YT2 channel, a mask around the Golgi structure(s) was manually drawn. In the case of phase 1 karyoplasts, where post-ER structures could not be identified, the mask was designed to cover a random area in the perinuclear area. (2) The mask was projected on the Sec31 or EEA1 channel, and the fluorescence intensity in this area was measured (FI-Golgi). (3) The mask was then rotated 180° around the nucleus and Sec31 and EEA1 fluorescence intensity was measured (FI-opposite_side). (4) The ratio FI-Golgi∶FI-opposite_side (clustering index) was measured for karyoplasts and control cells.

Correlative light and electron microscopy was performed as described previously (Colombelli et al., 2008; Tängemo et al., 2011). Epon serial sections were then collected and imaged with a Philips Biotwin CM120 electron microscope. For electron tomography, thicker serial sections (250–300 nm) were cut and collected on electron microscopy grids. Colloidal gold particles were then deposited on both surfaces to allow subsequent alignment of the tilted images. The tilted series of the cells of interest were then acquired with a TECNAI F30 (FEI) and the 3D reconstruction of the tomograms was carried out using the IMOD software (Kremer et al., 1996). Regions of interest were segmented using the same software.

We thank Uta Haselmann and Charlotta Funaya (Electron Microscopy Core Facility at EMBL) for technical support. We are grateful to Martin Lowe, Francis Barr and Irina Kaverina for sharing reagents, and to all members of the Pepperkok team and the Advanced Light Microscopy Facility at EMBL for discussions and comments on the manuscript.