Golgi enlargement in Arf-depleted yeast cells is due to altered dynamics of cisternal maturation

The mechanisms that control organelle size and number are attracting increasing interest (Goehring and Hyman, 2012; Levy and Heald, 2012; Marshall, 2002; Sengupta and Linstedt, 2011). Apart from being a basic cell biological problem, this topic has clinical relevance because cancer cells display altered organelle structure (Levy and Heald, 2012; Zink et al., 2004). One way to control organelle structure is with fusion and fission events. For example, mitochondrial networks are maintained by a balance between fusion and fission (Chan, 2012; Westermann, 2010). Similarly, endosomes and yeast vacuoles undergo frequent homotypic fusion (Epp et al., 2011). Another way to control organelle structure is to alter the relative rates of membrane influx and efflux (Sengupta and Linstedt, 2011). Many of these membrane remodeling events are controlled by GTPases. Improved understanding of such processes may enable the reprogramming of organelle size in diseased cells (Marshall, 2012).

An organelle that shows great plasticity and variability is the Golgi apparatus, which consists of disk-shaped cisternae that sometimes exhibit fenestrations, tubular extensions and lateral interconnections (Warren and Rothman, 2011). The mammalian Golgi undergoes changes during differentiation or cancer progression (Kellokumpu et al., 2002; Kirk et al., 2010; Lu et al., 2001; Weller et al., 2010; Wu et al., 2000). In most eukaryotes, Golgi cisternae are arranged in polarized stacks, but in some species such as the yeast Saccharomyces cerevisiae, Golgi cisternae are dispersed throughout the cell (Mowbrey and Dacks, 2009; Preuss et al., 1992). The dispersed cisternae can be tracked individually by fluorescence microscopy (Wooding and Pelham, 1998). A typical S. cerevisiae cell contains about six to ten early Golgi cisternae plus a similar number of late or trans-Golgi network (TGN) cisternae (Papanikou and Glick, 2009). Early yeast Golgi cisternae mature to generate late Golgi cisternae, a process that appears to be evolutionarily conserved (Glick and Nakano, 2009; Losev et al., 2006; Matsuura-Tokita et al., 2006).

S. cerevisiae can be used to study the regulation of Golgi structure. In a microscopy-based screen to identify thermosensitive mutants defective in Golgi inheritance, we also identified mutants that contained fewer and larger Golgi cisternae (Rossanese et al., 2001). A similar phenotype had been described for the ADP-ribosylation factor 1 (ARF1)-deleted (arf1Δ) mutant (Gaynor et al., 1998). S. cerevisiae contains the closely related genes ARF1 and ARF2, with ARF1 accounting for ∼90% of the Golgi-associated Arf (Stearns et al., 1990a). arf1Δ mutants show abnormal Golgi structure but only mild defects in secretion (Gaynor et al., 1998; Stearns et al., 1990b). We find that a thermosensitive mutant has enlarged Golgi cisternae due to impaired function of the N-myristoyl transferase Nmt1 (Johnson et al., 1994), which activates Arf (Donaldson and Jackson, 2011; Kahn, 2009). Thus, separate lines of investigation have pointed to Arf as a key regulator of Golgi size and number.

To clarify the mechanisms that generate enlarged Golgi cisternae, we used four-dimensional (4D) fluorescence microscopy. The results indicate that during Golgi maturation, early Golgi cisternae mature more slowly and less frequently in arf1Δ cells than in wild-type cells. By contrast, after the early-to-late Golgi transition, the maturation kinetics in arf1Δ cells are essentially normal. The consequence of these selective changes is a severe reduction in the number of late Golgi cisternae. Our analysis highlights the importance of kinetic parameters for regulating the size and copy number of dynamic compartments.

We used a thermosensitive yeast mutant that has larger and fewer late Golgi cisternae, as marked by Sec7–GFP (Rossanese et al., 2001). For further analysis, we also labeled the plasma membrane with mCherry–Ras2 (Tang et al., 2009). Late Golgi cisternae were visualized with 2D projections, and with 3D rendering that represented a cisterna as a closed surface (Fig. 1A). Although Golgi cisternae are topologically complex (see below), modeling a cisterna as a closed surface enabled us to use volume as a measure of cisternal size. An alternative measure was the maximal diameter of a cisterna in the X–Y plane. Quantification revealed that on average, late Golgi cisternae in mutant cells had approximately threefold greater volume and approximately twofold greater diameter than in wild-type cells (Fig. 1B,C).

To identify the mutated gene, a genomic library was screened for complementation of the thermosensitive growth defect (Rossanese et al., 2001). Two overlapping clones rescued the growth and Golgi size phenotypes (Fig. 1A–C). The overlapping region contained NMT1, which encodes N-myristoyl transferase (Johnson et al., 1994). Rescue of the mutant phenotypes was lost if mutant cells were cured of a plasmid carrying NMT1 (Fig. 1B,C). The nmt1 allele had a T400I mutation, and the growth and Golgi size phenotypes were rescued by reverting this mutation in the genome (Fig. 1B,C; supplementary material Fig. S1). Thus, a point mutation in NMT1 leads to enlarged late Golgi cisternae.

One substrate for Nmt1 is Golgi-associated Arf, a GTPase that recruits multiple effectors, including the COPI coatomer and clathrin adaptors (Donaldson and Jackson, 2011; Kahn, 2009). Because Arf activity depends on N-terminal myristoylation, we suspected that the enlarged Golgi cisternae in the nmt1 strain were due to reduced myristoylation of Arf.

This hypothesis was tested in two ways. First, we compensated for impaired Nmt1 activity in the mutant strain by overexpressing ARF1. Elevated Arf1 levels suppressed the Golgi size phenotype, and this effect was lost if the mutant cells were cured of the plasmid (Fig. 1B,C; supplementary material Fig. S1). Second, we deleted ARF1 to confirm that reduced Arf levels cause Golgi enlargement in our strain. Indeed, arf1Δ cells contained a small number of late Golgi cisternae that were often abnormally large (Fig. 2A,B). Thus, depletion or partial inactivation of Golgi-associated Arf leads to larger and fewer late Golgi cisternae.

It was reported that arf1Δ cells also had enlarged endosomes (Gaynor et al., 1998), as judged by labeling the endocytic pathway with the dye FM 4-64 (Vida and Emr, 1995). We observed abnormally large fluorescent compartments after incubating arf1Δ cells with FM 4-64 (Fig. 2C). However, FM 4-64 is a bulk membrane marker that moves from the endocytic pathway to the late Golgi/TGN (Lewis et al., 2000), and most of the Sec7–GFP-labeled structures were also labeled with recently internalized FM 4-64 in both wild-type and arf1Δ cells (Fig. 2C). We conclude that in arf1Δ cells, the enlarged compartments containing FM 4-64 are late Golgi cisternae.

Are the late Golgi structures in arf1Δ cells large individual cisternae or clusters of smaller cisternae? To address this question, we combined fluorescence microscopy with electron tomography (Kukulski et al., 2011). While optimizing the procedure, we discovered that embedding yeast cells in Lowicryl K4M resin preserved strong GFP fluorescence while yielding sufficient contrast.

For wild-type cells, we labeled late Golgi cisternae with Sec7–GFP, and prepared ∼300-nm sections from plastic-embedded samples. Labeled cisternae were visible by fluorescence microscopy (Fig. 2D). The same structures were then analyzed by electron tomography. This method examines a single thick section, so only part of each cisterna was visible, but the results indicated that wild-type late Golgi cisternae were curved and perforated disks (Fig. 2D; supplementary material Movie 1).

For arf1Δ cells (Fig. 2D; supplementary material Movies 2, 3), partial reconstructions indicated that mutant late Golgi cisternae were large fenestrated hollow structures, presumably identical to structures previously visualized in arf1Δ cells by thin-section electron microscopy (Gaynor et al., 1998). These results confirm that partial depletion of Arf generates abnormally large late Golgi cisternae.

Late Golgi cisternae are generated by maturation, and we sought to determine where in this pathway the arf1Δ mutation exerts its effects. First, we examined transitional endoplasmic reticulum (tER) sites, which produce COPII vesicles (Rossanese et al., 1999; Shindiapina and Barlowe, 2010). Some mutations cause coalescence of these structures (Castillon et al., 2009; Levi et al., 2010; Shindiapina and Barlowe, 2010), but we saw no consistent differences in the tER patterns of wild-type and arf1Δ cells (supplementary material Fig. S2), suggesting that the arf1Δ mutation acts at the level of the Golgi.

Next, we counted early and late Golgi cisternae by using GFP–Vrg4 as an early Golgi marker, and Sec7–DsRed as a late Golgi marker (Losev et al., 2006). Wild-type cells contained an average of nine early Golgi cisternae and seven late Golgi cisternae (Fig. 3A; Table 1). In the arf1Δ strain, early Golgi cisternae tended to be slightly enlarged (supplementary material Fig. S3) and their average number was reduced to five. For the late Golgi, the reduction in the arf1Δ strain was more pronounced, with an average of only two cisternae per cell (Fig. 3A and Table 1). Thus, as the Golgi matures, the arf1Δ mutation progressively reduces the number of cisternae.

The effects of the arf1Δ mutation on Golgi cisternal number might be due to altered maturation kinetics. To test this idea, we devised a simple quantitative framework in which E is the average number of early Golgi cisternae per cell, L is the average number of late Golgi cisternae per cell, m is the average maturation frequency, p_E is the average persistence time of early Golgi cisternae, and p_L is the average persistence time of late Golgi cisternae. m is defined here as the number of early-to-late Golgi transitions per minute, i.e. the frequency at which early Golgi cisternae disappear and late Golgi cisternae appear. Because the average number of Golgi cisternae is constant at steady-state, m also refers to the frequency at which early Golgi cisternae appear. The average number of early or late Golgi cisternae should be equal to the frequency of appearance multiplied by the persistence time. Thus: E = m×p_E and L = m×p_L. Combining these equations yields: E/L = p_E/p_L.

The implications are that changes in maturation frequency or persistence time could alter the number of cisternae, and that a selective change in one of the persistence times could alter the ratio of early to late Golgi cisternae.

This model was tested by measuring maturation parameters for wild-type and arf1Δ strains. An initial experiment employed arf1Δ cells expressing Sec7-GFP. In wild-type cells, Sec7–GFP is present on cisternae and then this marker is lost within ∼1.5–2 minutes (Losev et al., 2006). The same phenomenon was observed with arf1Δ cells. Because this mutant strain has so few late Golgi cisternae, we could often track every Sec7–GFP-labeled structure in a cell. Fig. 3B shows that in a representative cell over a 10-minute period, Sec7–GFP was present on each cisterna and then lost within ∼100 seconds. This result suggests that late Golgi cisternae function similarly in wild-type and arf1Δ cells (Daboussi et al., 2012).

Further analysis was performed with strains expressing GFP-Vrg4 and Sec7-DsRed (supplementary material Movie 4). Dual-color 4D movies were captured for 5 minutes, a period suitable for acquiring closely spaced Z-stacks without excessive photobleaching. To measure the maturation frequency m (Fig. 3C; Table 1), we counted the number of early-to-late Golgi transitions per minute (Losev et al., 2006). To measure the persistence times p_E and p_L (Fig. 3D; Table 1), we identified cisternae that acquired and then lost either GFP–Vrg4 or Sec7–DsRed.

Our analysis revealed a striking effect: the arf1Δ mutation selectively increased the early Golgi persistence time p_E. The p_E/p_L ratio was ∼1 in wild-type cells but ∼2 in arf1Δ cells (Table 1). This finding could explain why the number of early Golgi cisternae, E, was similar to the number of late Golgi cisternae, L, in wild-type cells, whereas E was about twice as large as L in arf1Δ cells.

Another clear difference between the two strains is the maturation frequency, m, which was ∼2.5-fold greater in wild-type cells than in arf1Δ cells (Table 1). We infer that the entire early Golgi maturation process is slowed in arf1Δ cells, with early Golgi cisternae persisting longer and maturing less frequently than in wild-type cells.

Table 1 shows a comparison of the measured values E and L, and the corresponding predicted values m×p_E and m×p_L. The number of late Golgi cisternae, L, agrees reasonably well with the predicted value m×p_L for both wild-type and arf1Δ strains. However, the number of early Golgi cisternae, E, is higher than the predicted value m×p_E for both strains.

To explore this discrepancy, we re-evaluated the rationale for using m×p_E as an estimate of E. Because the spatial and temporal resolution of the 4D movies was insufficient for counting early Golgi formation events directly, we assumed that the frequency of early Golgi formation events matched the frequency of early-to-late Golgi conversion events. This assumption would be incorrect if early Golgi cisternae undergo homotypic fusion. Indeed, such homotypic fusion events take place, as illustrated in Fig. 4A and supplementary material Movie 4. A typical wild-type or arf1Δ cell exhibited about one early Golgi homotypic fusion event per minute (Fig. 4B). As a consequence of homotypic fusion, the frequency of early Golgi formation events is higher than the maturation frequency m, so m×p_E is an underestimate of the number of early Golgi cisternae.

Despite the limitations caused by homotypic fusion, the quantitative framework yields two strong conclusions (supplementary material Fig. S4). First, the reason that arf1Δ cells contains more early than late Golgi cisternae is that in these cells, early Golgi cisternae persist longer than late Golgi cisternae. Second, the reason that arf1Δ cells have fewer late Golgi cisternae than wild-type cells is that arf1Δ cells have an abnormally low frequency of early-to-late Golgi transitions but a normal persistence time for the late Golgi.

Rates of membrane influx and efflux can regulate the overall size of the mammalian Golgi (Guo and Linstedt, 2006; Sengupta and Linstedt, 2011), and we find that in S. cerevisiae, kinetic parameters can also regulate Golgi copy number. Because Golgi cisternae are constantly turning over, the copy number of a Golgi compartment depends on how often the compartment forms and how long it persists. The total amount of Golgi material is presumably comparable in different strains, so when the number of Golgi cisternae in a strain is abnormally small, the average size of the cisternae is abnormally large. Thus, the rates of processes that transform a maturing compartment can dramatically influence compartmental size and copy number.

Slowed maturation of the early Golgi in arf1Δ cells helps to explain the observed delay in secretion (Gaynor et al., 1998), but the mechanism by which partial depletion of Arf alters Golgi maturation remains to be determined. A decrease in Golgi-associated Arf may compromise the recruitment or function of the COPI vesicle coat. In support of this idea, a thermosensitive COPI allele was reported to slow Golgi maturation at the restrictive temperature (Matsuura-Tokita et al., 2006). Another possibility is that Arf depletion perturbs lipid metabolism (Gaynor et al., 1998). Indeed, the arf1Δ mutation induces the unfolded protein response (Jonikas et al., 2009), which in turn alters phospholipid biosynthesis and other aspects of the secretory pathway (Travers et al., 2000). Regardless of the specific processes that are affected, it is notable that two previous studies of Golgi cisternal enlargement in S. cerevisiae both implicated Arf (Gaynor et al., 1998; Peyroche et al., 2001) and that our identification of the nmt1 mutant points to Arf as well. These findings support the idea that Arf is a regulator of Golgi maturation (Stalder and Antonny, 2013).

How might our results be relevant to organisms with a stacked Golgi? We propose that changes in the kinetics of Golgi maturation could alter the number of cisternae per stack. Golgi cisternal number is variable between different cell types – in mammals, the number of cisternae per stack ranges from about 3 to 10 depending on the cell type (Rambourg and Clermont, 1997). Large Golgi stacks typically contain multiple COPI-vesicle-producing Golgi cisternae but only one or two clathrin-coated vesicle-producing TGN cisternae (Mogelsvang et al., 2004; Staehelin and Kang, 2008). By analogy to the arf1Δ mutant, if the persistence time of Golgi cisternae increased while the persistence time of TGN cisternae remained unchanged, the result would be a selective increase in the number of Golgi cisternae (Fig. 5). Interestingly, when mammalian cells are depleted of the Rab6 GTPase, secretion slows and the number of Golgi cisternae per stack increases (Storrie et al., 2012). It is possible that Rab6 depletion increases the persistence time of Golgi cisternae. Future developments in live-cell microscopy should enable a test of our hypothesis that maturation kinetics regulate the size of Golgi stacks.

All experiments were performed with derivatives of the haploid S. cerevisiae strain JK9-3d, which has the genotype leu2-3,112 ura3-52 rme1 trp1 his4 (Kunz et al., 1993). A mutant with enlarged late Golgi cisternae (Rossanese et al., 2001) was backcrossed four times to the parental strain, and thermosensitive growth segregated with the enlarged Golgi phenotype as a single-gene mutation through each cross. Yeast were grown in rich glucose medium (YPD) or minimal glucose medium (SD) (Sherman, 1991), with shaking at 200 r.p.m. in baffled flasks. Growth media were obtained from HiMedia Laboratories (Mumbai, India) or Difco Laboratories (Detroit, MI, USA). Wild-type strains were grown at 30°C unless otherwise indicated. Thermosensitive mutants were grown at 25°C, and then shifted to 37°C for 30 minutes before analysis. Molecular biology procedures were simulated and recorded using SnapGene software (GSL Biotech, Chicago, IL, USA; http://www.snapgene.com).

Integrating or centromeric plasmids were transformed into yeast using the lithium acetate method (Gietz and Woods, 2002), and chromosomal gene replacements were performed using the pop-in/pop-out method (Rothstein, 1991). SEC7 was tagged by gene replacement with three or six tandem copies of monomeric EGFP (mEGFP) (Zacharias et al., 2002) as previously described (Rossanese et al., 2001). Additional integrating constructs were used to overexpress either Sec7-3xmEGFP or Sec7-6xDsRed.M1 (Losev et al., 2006). Vrg4 was tagged at its N-terminus with sGFP by gene replacement (Losev et al., 2006). Sec13 was tagged at its C-terminus with EGFP by gene replacement (Rossanese et al., 1999). Ras2 was tagged at its N-terminus with mCherry (Shaner et al., 2004) as follows. The RAS2 gene plus flanking sequences were amplified using forward primer 5′-TCGATGAATTCTAGCTCTCGGGCGAATATC-3′ and reverse primer 5′-TCTAAGGGGAAAGAGAAGCTTG-3′. The PCR product was digested with EcoRI and HindIII and subcloned at the same sites in the centromeric URA3 plasmid YCplac33 (Gietz and Sugino, 1988). A BamHI-NotI cassette was then introduced by primer-directed mutagenesis at the RAS2 start codon. mCherry was amplified with primers 5′-tggtaGGATCCATGGTGAGCAAGGGCGAGGA-3′ and 5′-tgttcGCGGCCGCaCTTGTACAGCTCGTCCATGC-3′, and the PCR product was subcloned as a BamHI–NotI fragment. A BamHI–HindIII fragment encoding mCherry–Ras2 was then subcloned into the TRP1 integrating plasmid YIplac204-T/C to drive strong constitutive expression from the TPI1 promoter (Losev et al., 2006). This construct was linearized with EcoRV for integration at the TRP1 locus.

To complement the nmt1 mutation, the wild-type NMT1 gene with its 5′ and 3′ control regions was amplified with primers 5′-AAGTATAAGCTTGCTATCGTTTACAAATTAAA-3′ and 5′-CGCTGGGTACCATTTTTTCACTTCGTCAATTGAC-3′, and subcloned as HindIII–KpnI fragment into YCplac33. To overexpress ARF1, this gene with its 5′ and 3′ control regions was amplified with primers 5′-TGCACAGTTGAATTCTCG-3′ and 5′-GATGAAAGCTTCTGCAGTTGTCAGTC-3′, and subcloned as an EcoRI–HindIII fragment into the high-copy episomal URA3 plasmid YEplac195 (Gietz and Sugino, 1988). Where indicated, strains were cured of the URA3 plasmids by growth on plates containing 1 mg/ml 5-fluoroorotic acid (United States Biological, Salem, MA, USA).

To revert the T400I mutation in the nmt1 mutant, a fragment of the wild-type NMT1 gene was amplified with primers 5′-GCCACCAAGCTTACACCGGAAGATATATCTGA-3′ and 5′-CGCTGGGTACCATTTTTTCACTTCGTCAATTGAC-3′, and subcloned as a HindIII–KpnI fragment into a modified URA3 integrating plasmid YIplac211 (Gietz and Sugino, 1988) that had been mutagenized to remove the EcoRI site. One of the two EcoRI sites in the NMT1 fragment was then removed by changing codon 293 from GAA to GAG. The final construct was linearized at the unique EcoRI site for integrative gene replacement, and the reverted NMT1 allele was confirmed by PCR amplification and sequencing of genomic DNA. To delete the ARF1 gene, the kanMX cassette was amplified with the following primers to append sequences that matched the regions flanking the ARF1 open reading frame: 5′-ATTGAAGGTATAAGAAAGAACTCAAACAGGTTTAATAGAATTAAAACGTACGCTGCAGGTCGAC-3′ and 5′-TTCATTTAGTTTATACAAGCGTATTTGATCCATATTCTAGAATTTATCGATGAATTCGAGCTCG-3′. The resulting fragment was transformed into cells, which were plated on YPD medium containing 250 µg/ml G418 (Sigma-Aldrich, St. Louis, MO, USA) to select for double-crossover replacement of the ARF1 open reading frame (Wach et al., 1994).

To capture static images of fluorescently tagged tER sites, live yeast cells were compressed beneath a coverslip and viewed using a Zeiss Axioplan 2 microscope with a 100× 1.4 NA objective. A single image plane was captured with a Hamamatsu digital camera. To quantify the number of tER sites, Z-stacks were captured with a Zeiss LSM 780 confocal microscope. Each confocal 3D data set was opened with Imaris software (Bitplane, Zurich, Switzerland). The spot detection option was used, with the threshold set to detect even the smallest fluorescent tER sites. The number of spots was then counted in randomly chosen full-size ‘mother’ cells.

To capture static images of cells labeled with FM 4-64, dye in a 1 mM DMSO stock solution was added to a log-phase culture at 23°C. After a brief incubation, non-internalized dye was quenched by adding excess SCAS (4-sulfonato calix[8]arene, sodium salt; Biotium, Hayward, CA, USA) from a 1 mM aqueous stock solution. Cells were then mixed with 3.0 µm polystyrene beads (Polysciences, Warrington, PA, USA), compressed beneath a coverslip that had been siliconized with Sigmacote (Sigma-Aldrich), and viewed with a Zeiss LSM 710. GFP fluorescence was excited with a 488-nm laser and collected between 495 and 550 nm, and simultaneously, FM 4-64 fluorescence was excited with a 561-nm laser and collected between 650 and 760 nm. A 3D image volume spanning an entire cell was captured using a pixel size of 40 nm and Z-stack spacing of 0.47 µm. Image stacks were converted to 16-bit and average projected using ImageJ (http://rsbweb.nih.gov/ij/).

Confocal imaging was performed with either a Leica SP5 or a Zeiss LSM 780 for 4D imaging, or a Zeiss LSM 510 META or LSM 710 for single time-point measurements, equipped with 100× or 63× 1.4 NA objectives. Cells grown to log phase at 25°C or 30°C in nonfluorescent or minimally fluorescent SD medium were immobilized on glass-bottomed dishes (Cell E&G, Houston, TX, USA) using concanavalin A (Sigma-Aldrich) as previously described (Bevis et al., 2002), and were imaged at room temperature. Single- or dual-color data sets were obtained using separate excitation and capture of red and green signals, with a pinhole of 1.0–1.2 AU and with line averaging of 4 to improve the signal-to-noise ratio. Transmitted light images were captured in the blue channel. The pixel size was 70–90 nm. Optical sections were 0.25–0.40 µm apart, and ∼15–20 optical sections were collected to span an entire cell. Z-stacks were collected at intervals of 2–4 seconds. To limit photodamage, laser illumination was minimized and confocal scans were carried out as quickly as possible.

Fluorescence micrographs were assembled using Adobe Photoshop, with uniform adjustments of brightness where appropriate. Projected movies of 4D confocal data sets were generated and quantified using ImageJ as follows. To remove shot noise, individual optical sections were processed with a custom plugin that implemented a 3D version of a 3×3 hybrid median filter (Hammond and Glick, 2000). The processed optical sections were then average projected and corrected for exponential photobleaching (Bevis et al., 2002), and the fluorescence and transmitted light channels were merged. Quantification of late Golgi fluorescence intensities was performed as previously described (Losev et al., 2006).

For measurements of cisternal size, a confocal 3D data set was opened with Imaris. After the image was cropped, the surface fill option was used to define a closed surface around each fluorescent structure of average diameter at least 0.5 µm, with objects being smoothed using a filter radius of 0.05 µm. The rendered 3D objects were then analyzed to obtain the desired parameters. To calculate the average volume of late Golgi cisternae, 60 cells were imaged for each strain, and all of the rendered cisternae were measured. The corresponding volume measurements of early Golgi cisternae were performed on 20 cells for each strain. To calculate the average X–Y-plane diameter, 10 cells were imaged for each strain, and the maximal X–Y-plane diameter was measured for three or four representative cisternae from each cell.

For measurements of maturation and homotypic fusion parameters, a confocal 4D data set was opened with Imaris. The image was exported to Fiji (http://fiji.sc/Fiji), and shot noise was removed by processing individual optical sections with a custom plug-in that implemented a 3D version of a 3×3 hybrid median filter with a tight mean. The processed optical sections were then corrected for exponential photobleaching, and the fluorescence and transmitted light channels were merged. This processed image was imported back into Imaris for visual inspection and manual quantification.

All measurements were performed using the same set of 20 dual-color 5-minute 4D movies, with 10 movies each for wild-type and arf1Δ strains expressing GFP-Vrg4 and Sec7-DsRed. A single full-size ‘mother’ cell from each movie was analyzed using Imaris. To count early and late Golgi cisternae in a cell, the numbers of green and red spots in each movie were recorded at 10 successive time points and averaged, with a spot being recorded only if it persisted for at least two consecutive time points. To measure the maturation frequency, the number of maturation events in each movie was counted during each of minutes 2, 3 and 4, with a cisterna being tracked for additional time if needed to verify the maturation event. To measure the persistence times of early and late Golgi cisternae, 20 cisternae that could be seen to acquire and then lose either the early or the late Golgi marker were chosen at random from the 10 movies, and the fluorescence duration for each cisterna was measured. To measure the frequency of homotypic fusion, the number of early or late Golgi fusion events was counted during a 2-minute interval for each movie. Cisternae that underwent homotypic fusion were excluded from the measurements of persistence times.

Wild-type or arf1Δ strains carried a Sec7-3xmEGFP gene replacement plus an integrating vector that drove mild (∼3×) overexpression of Sec7-3xmEGFP (Losev et al., 2006). A 100 ml culture was grown in SD medium at 30°C to mid-log phase with shaking. Cells were then subjected to high-pressure freezing, freeze substitution, and embedding in plastic as described previously (Levi et al., 2010), except that glutaraldehyde was omitted during freeze substitution and the resin was Lowicryl K4M.

Embedded samples were trimmed and cut with a Leica EM UC6 ultramicrotome to produce ∼300 nm sections, which were placed on 200 mesh London Finder Formvar/carbon-coated copper grids (Electron Microscopy Sciences, Hatfield, PA, USA). To provide fiducial markers for both light and electron microscopy, 15 nm colloidal gold was applied to both sides of the grid as described previously (Levi et al., 2010). For light microscopy, a grid was placed on a glass slide with the resin side up, and a 22×22 mm number 1.5 glass coverslip with a 10-µl drop of 500 mM Na⁺HEPES, pH 7.5 was inverted onto the grid. After capture of fluorescence and differential interference contrast micrographs on a Zeiss Axioplan 2 with a 1.4 NA objective, the grid was retrieved and blotted dry. For embedded wild-type cells, the samples were post-stained for 8 minutes with 2% uranyl acetate in 70% methanol, followed by a water rinse, followed by 5 minutes with Reynold's lead citrate. Cells that showed promising morphologies and fluorescence patterns were analyzed by electron tomography as described previously (Levi et al., 2010). Golgi membrane contours were traced in the tomograms, and the structures were modeled.