The spatial separation of processing and transport functions to the interior and periphery of the Golgi stack

It is unclear how the two principal functions of the Golgi complex, processing and transport, are spatially organized. Studying such spatial organization by optical imaging is challenging, partially due to the dense packing of stochastically oriented Golgi stacks. Using super-resolution microscopy and markers such as Giantin, we developed a method to identify en face and side views of individual nocodazole-induced Golgi mini-stacks. Our imaging uncovered that Golgi enzymes preferentially localize to the cisternal interior, appearing as a central disk or inner-ring, whereas components of the trafficking machinery reside at the periphery of the stack, including the cisternal rim. Interestingly, conventional secretory cargos appeared at the cisternal interior during their intra-Golgi trafficking and transiently localized to the cisternal rim before exiting the Golgi. In contrast, bulky cargos were found only at the rim. Our study therefore directly demonstrates the spatial separation of processing and transport functions within the Golgi complex.


Introduction
The Golgi complex is one of the most important processing and sorting stations along the secretory and endocytic pathway (Glick and Luini, 2011;Klumperman, 2011;Lu and Hong, 2014). In mammalian cells, it consists of a network of laterally linked Golgi stacks. As the structural unit, a Golgi stack comprises 4-7 flattened cisternae and can be divided into cis, medial and trans-regions. The trans-Golgi region further develops into the trans-Golgi network (TGN). It is known that the cis-Golgi receives secretory cargos from the endoplasmic reticulum (ER) exit site (ERES) and ER Golgi intermediate compartment (ERGIC), while the trans-Golgi and TGN exchange materials with endosomes and the plasma membrane (PM). At the moment, we still don't understand how the Golgi becomes organized and works at the molecular and cellular level (Glick and Luini, 2011). One of the challenges in studying the Golgi is to spatiotemporally resolve residents and transiting cargos among individual cisternae of Golgi stacks, a task currently beyond the capabilities of even super-resolution and electron microscopy (EM).
It has been hypothesized that the two principal functions of the Golgi, processing and transport, are spatially organized for optimal efficiency (Patterson et al., 2008). However, such molecular organization across the Golgi stack has not been directly demonstrated. Previously, by utilizing nocodazole-induced Golgi mini-stacks, we developed a conventional microscopy based super-resolution method, named GLIM (Golgi localization by imaging center of fluorescence mass), to quantitatively map the axial position or localization quotient (LQ) of a Golgi protein with nanometer accuracy (Tie et al., 2017;Tie et al., 2016b). To understand the molecular organization of the Golgi ministack, the lateral localization, which refers to the distribution of a protein within Golgi cisternal membrane sheets, is also required. Although more structural details of the Golgi can be resolved with the The double-punctum appearances of Giantin, Golgin84 and GPP130 indicate side views of Golgi mini-stacks. In each merge, the intensity profile is generated along a thick line, represented by a dotted box, with the direction indicated by the arrow (the same scheme is used throughout this study). The dotted box schematically marked the start, end and width of the line. The direction arrow roughly follows the cis-to-trans Golgi axis using the cismost (GM130 in this case) and trans-most markers in each panel. Dotted pink lines connecting double-punctum are almost orthogonal to the cis-totrans Golgi axis. The intensity plot is normalized and color-coded as the corresponding merge image. (D-F) En face averaged images of Giantin, fluorescence protein (FP)-Golgin84 and GPP130-GFP. The corresponding radial mean intensity profile is shown at the right with distance from the center of fluorescence mass (normalized to the radius of Giantin) as the x-axis and radial mean intensity (normalized) as the y-axis. Both GFP and mCherry-tagged Golgin84 images were used for FP-Golgin84. n, the number of averaged Golgi mini-stacks. (G) GPP130 mostly localizes to the cisternal rim (arrows) of the native Golgi by EM. NRK cells transiently expressing GPP130-APEX2-GFP were subjected to APEX2-catalyzed reaction followed by EM. Note that cells were not subjected to nocodazole treatment. The EM thin section image displays the side view of a Golgi mini-stack. The electron density indicates the localization of GPP130 (arrows). (H) The histogram showing the distribution of diameters of Giantin-rings. (I, J) Giantin N and C-terminus colocalize at the cisternal rim. In (I), cells were co-stained using Giantin antibodies raised against its N and C-terminus. In (J), Giantin N-terminus was stained by an antibody and its C-terminus was revealed by exogenously expressed mScarlet-Giantin-C129. In the en face view, dotted arrow represents the line used to generate the line intensity profile (width = 1 pixel), while in the side view, the dotted box that is in the direction of the arrow and parallel to the Golgi cisterna represents the line for intensity profile. (K) The interior localization of MGAT2 within the Giantin-ring. Line intensity profiles of the en face and side views are acquired as those in (I) and (C) respectively. Scale bar, 500 nm. Figure 1 continued on next page Identifying en face and side views of Golgi mini-stacks By assessing the super-resolution staining patterns of Giantin, GPP130 or Golgin84, we can conveniently identify en face and side view oriented Golgi mini-stacks, images of which should appear as a ring and double-punctum, respectively. It was discovered that some Golgi residents, such as MGAT2, localized to the interior of Giantin-rings ( Figure 1K). Consistent with this interpretation, side views of MGAT2 appeared as a short bar connecting the Giantin double-punctum ( Figure 1K). Under the EM, MGAT2-APEX2-GFP preferentially localized to the cisternal interior (next section). Therefore, there are at least two types of lateral localizations: rim and interior, as represented by Giantin and MGAT2.
Golgi trafficking components mainly localize to the periphery of a Golgi mini-stack We systematically examined the lateral localization of Golgi residents using their en face and side views. Two types of residents were studied in this work -components of trafficking machinery, including those involved in the structure and organization of the Golgi, and enzymes involved in the post-translational modifications, particularly glycosyltransferases. Due to the lack of reagents to detect endogenous proteins, many residents were detected by the overexpression of their tagged fusions (Table 1). Caution must be taken in the interpretation of our data as it has been documented that overexpression can change both the axial and lateral localization of Golgi residents (Cosson et al., 2005). We discovered that the lateral localization of trafficking machinery components shares common features according to their LQs.

ERES, ERGIC and cis-Golgi proteins (LQ <0)
COPII coat subunits, including Sec13 and Sec23a, COPI coat subunits, including b and g-COP, KDEL receptor, GS27, ERGIC53, Arf4 and Arf5, displayed lumps or puncta around Giantin-rings in en face views and at one side of Giantin-double-punctum in side views ( , suggesting that they mainly localize to the rim of their corresponding cisternae and are mostly absent from the cisternal interior. Arf1, whose LQ is 0.75, is an exception here. Although its en face view demonstrated that it is in the cisternal interior, side view images uncovered that there were two pools: a cis/medial and a trans-Golgi/TGN pool, with a much reduced presence in between ( Figure 3G,H). This observation is consistent with the notion that Arf1 functions in the cis-Golgi and TGN for the assembly of the COPI and clathrin coat, respectively (Gillingham and Munro, 2007).    (Ladinsky et al., 1999) and the role of clathrin coat in transporting these cargos to the endolysosome (Peden et al., 2001;Teuchert et al., 1999). Our data are also consistent with the notion that the TGN comprises domains of distinct molecular compositions (Brown et al., 2011;Derby et al., 2004).
In summary, our extensive super-resolution imaging data suggest that Golgi trafficking components mainly localize to the entire cis-cisternae, rim of medial and trans-cisternae and punctate or tubular profiles at non-stacked regions, which include the ERES, ERGIC and TGN.

Glycosylation enzymes reside at the interior of a Golgi stack
We studied components of Golgi post-translational modification machinery (Table 1; Supplementary file 1), including a GDP-fucose transporter, SLC35C1 (Lübke et al., 2001), and more than a dozen enzymes involved in N-glycosylation (Man1B1, MGAT1, ManII, MGAT2, GalT, SialT and MGAT4B), O-glycosylation (GALNT1, GALNT2 and POMGNT1), poly-N-acetyllactosamine synthesis (b4GalT3), glycosaminoglycan synthesis (b3GalT6 and b4GalT7) and sulfation (TPST1 and 2). Interestingly, their LQs were found to be in the range from 0.23 to 1.0 (Table 1), suggesting that Golgi enzymes mainly localize to the medial and trans-region of the Golgi, but not to the cis-Golgi and TGN. This observation is consistent with previous EM studies. For example, in plant cells, polysaccharides were mainly detected in the medial and trans-Golgi cisternae (Zhang and Staehelin, 1992). Similarly, in mammalian cells, the N-glycan modifying enzymes ManI, ManII and MGAT1 have been mapped to the medial and trans-region of the Golgi stack (Dunphy et al., 1985;Nilsson et al., 1993;Rabouille et al., 1995;Velasco et al., 1993). However, in contrast to our quantitative results, previous EM work has assigned GalT (Nilsson et al., 1993;Rabouille et al., 1995;Roth and Berger, 1982) and SialT (Rabouille et al., 1995;Roth et al., 1985) to the TGN in addition to the trans-Golgi. Sub-Golgi localizations are not always consistently reported, which is likely due to two reasons. First, the cis, medial, trans-region and TGN are not rigorously defined and the assignment of Golgi regions can be subjective. Second, it has been documented that the sub-Golgi localization of enzymes can be cell-type dependent (Velasco et al., 1993).
In contrast to trafficking components, our Golgi enzymes and SLC35C1 localized within Giantinrings as a central disk in en face views ( Figure 1K; Figure (Table 1), a significant amount of these proteins are expected to reside in the same cisternae. The lateral distribution pattern of MGAT2 and MGAT4B suggests that they should mainly localize to the interior of cisternae as a central disk and inner-ring, respectively, within the Giantinrim in the same cisternae ( Figure 4G). Enzymes, such as b4GalT3 and ST6Gal1, which have similar LQs (Table 1), were observed to localize to shared and distinct domains within Giantin-rings ( Figure 4H).
To substantiate our light microscopic data, we examined the localization of MGAT2-APEX2-GFP in the native Golgi by EM. 93% (n = 58) of Golgi stacks showed an enrichment of MGAT2 in the cisternal interior ( Figure 4I; Figure 4-figure supplement 3A-C), which is in contrast to the staining pattern observed for GPP130 ( Figure 1G). Noticeably, APEX2-generated electron density was also found in vesicles and budding profiles at the rim (arrow heads in Figure 4I). However, we did not find MGAT2-AcGFP1 ( Figures 1K and 4C) or MGAT2-APEX2-GFP (Figure 4-figure supplement 1U) signal outside Giantin-rings by fluorescence imaging of Golgi mini-stacks. Although the identity and destiny of these vesicles are currently unknown, our observations suggest that Golgi enzymes might be depleted from the rim either by retrieval to the interior or by sorting into membrane carriers. Together, our data demonstrate that Golgi enzymes mainly localize to the interior of medial and trans-cisternae as a concentric disk or inner-ring, while trafficking machinery components exhibit rim localization.

A quantitative molecular map of the Golgi mini-stack
To quantitatively describe the overall lateral distribution of Golgi proteins, we assume that a Golgi protein has a radial symmetry localization around the Golgi axis as a concentric disk or ring. The normalized radius of the ring or disk can be measured using the radial mean intensity profile of en face averaged images (see Materials and methods). A plot of the normalized radius versus LQ quantitatively summarizes our morphological observations of ring and disk distribution of various Golgi residents ( Figure 4J). While medial and trans-Golgi trafficking machinery components are at the cisternal rim, Golgi enzymes all localize to the interior with Man1B1, ManII, MGAT4B and TPST2 appearing as concentric inner-rings and the rest as central disks. Interestingly, it also reveals that ciscisternae have smaller diameters than medial ones, consistent with many EM thin-section or tomographic 3D images (Bykov et al., 2017;Engel et al., 2015;Staehelin and Kang, 2008), though the biological significance of which remains to be further investigated.

Imaging the organization of the native Golgi complex
Having studied in detail the organization of Golgi mini-stacks, we attempted to resolve the organization of the native Golgi complex by the super-resolution microscopy. Giantin and Golgi enzymes were used to mark the rim and interior of stacked cisternae, respectively. In the less dense region, Giantin and GPP130 staining appeared as distinctive ring-or loop-patterns, with b4GalT3 and GM130 filling the interior ( Figure 4K,L), similar to the nocodazole-induced mini-stack. b4GalT3 and GM130 positive membrane sheets likely correspond to stacked Golgi cisternae. In most cases, Giantin and GPP130 positive curvy lines did not correspond to side views or cross-sections of Golgi stacks. Instead, they corresponded to the rim of cisternae in oblique or en face views (arrows in Figure 4L). In the more densely packed region, cisternae appeared to pile on top of each other, a configuration that requires much higher z-axis resolution to be resolved. Nonetheless, we

The lateral localization of secretory cargos during their intra-Golgi trafficking
To study the lateral localization of secretory cargos during their intra-Golgi trafficking, the retention using selective hooks (RUSH) system was adopted to synchronously release secretory cargos (Boncompain et al., 2012). The RUSH reporter CD59, a GPI-anchored protein, was first detected in the interior of cis-Golgi cisternae after 10 min of chase ( Figure 5A,B). During its transition through the Golgi mini-stack, as evidenced in its LQ versus time plot ( Figure 5B), CD59 remained in the interior ( Figure 5A), although its total intensity in Golgi mini-stacks initially increased and subsequently decreased due to the export toward the PM. At the later stage of the chase, there were CD59 positive puncta and tubular profiles outside Giantin-rings, which were likely Golgi-derived exocytic transport carriers ( Figure 5A, arrows in 60 min). Similarly, in live-cell super-resolution imaging, RUSH reporter mCherry-GPI started to appear in the interior of the Golgin84-ring 6 min after chase; it remained there for >30 min before disappearing due to post-Golgi exocytic trafficking ( Figure 5C; Figure 5-video 1). Transmembrane RUSH reporters, E-cadherin, VSVG and CD8a-Furin, and a soluble secretory reporter, signal peptide fused GFP, followed similar lateral localization pattern during their intra-Golgi trafficking ( Figure 5-figure supplement 1A-D). Collectively, our data demonstrated that conventional secretory cargos partition to the interior of the cisternae during their Golgi transition.
The secretory cargo wave does not seem to grossly affect the interior distribution of Golgi enzymes, as evidenced by ST6Gal1 ( Figure 5-figure supplement 2). By image quantification, >85% of ST6Gal1-moxGFP was found to remain in the interior during the Golgi transition of synchronized VSVG, although a small fluctuation (<4%) was noticed ( Figure 5-figure supplement 2A-C). Our finding is different from a previous EM study, in which the shift of Golgi enzymes from the rim to the interior was observed under a traffic wave (Kweon et al., 2004). A more systematic investigation is required to resolve this discrepancy.

Bulky size prevents the localization of secretory cargos at the cisternal interior
Based on EM data, Rothman lab previously proposed that large secretory protein aggregates are segregated to the cisternal rim (Lavieu et al., 2013). To investigate if bulky cargos partition to the rim, we imaged the RUSH reporter GFP-collagenX, a soluble secretory protein that tends to form oligomers (Kwan et al., 1991), by Airyscan super-resolution microscopy. We observed that Golgitransiting GFP-collagenX appeared either diffuse or punctate ( Figure 5D). Assuming that Golgi- Figure 4 continued stack images. (H) b4GalT3 and ST6Gal1 can localize to shared (arrows) and distinct domains within the cisternal interior. (I) MGAT2 localizes to the cisternal interior of the native Golgi by EM. NRK cells transiently expressing MGAT2-APEX2-GFP were subjected to APEX2-catalyzed reaction followed by EM. Note that cells were not subjected to nocodazole treatment. The thin section EM image displays the side view of a Golgi stack. MGAT2-APEX2 positive cisternal interior and budding profiles are indicated by arrows and arrow heads, respectively. (J) A quantitative molecular map of the Golgi mini-stack. The normalized radius of a Golgi protein is plotted against its corresponding LQ (Table 1). Red open and closed circle denote ring and disk lateral localization pattern, respectively. n, the number of Golgi mini-stacks used to calculate normalized radius. (K,L) Identifying the rim and interior of native Golgi cisternae. Cells were not treated with nocodazole. In (K), the cisternal rim (arrows) and interior are labeled by Giantin and b4GalT3, respectively. In (L), Giantin and GPP130 positive curvy lines (arrows) represent cisternal rim and do not correspond to side views or cross sections of Golgi stacks. The boxed region in each image is enlarged in the upper right corner. Scale bars represent 500 nm unless specified otherwise. DOI: https://doi.org/10.7554/eLife.41301.011 The following figure supplements are available for figure 4: localized GFP-collagenX puncta were single multimeric aggregates, using GFP-tagged nucleoporin Nup133 as an in vivo GFP fluorescence standard, we estimated that Golgi-transiting GFP-collagenX puncta had 190 ± 20 copies (n = 77) ( Figure 5-figure supplement 3A,B). The diffused collagenX is probably in a much lower oligomeric state. Throughout its intra-Golgi trafficking, collagenX, either in punctate or diffuse appearance, was excluded from the interior of Giantin-rings, where co-expressed mCherry-GPI clearly localized ( Figure 5D-F). Instead, it always resided at the rim, either colocalizing with Giantin or surrounding Giantin-rings as discrete puncta. At later stages, the puncta outside Giantin-rings were probably exocytic carriers targeting to the PM.
We also tested soluble and transmembrane secretory cargos, FM4-moxGFP and GFP-FM4-CD8a, whose aggregation states can be controlled by the small molecule -D/D solubilizer. These two cargos are similar to the ones used previously (Lavieu et al., 2013). NRK cells expressing either cargo were treated with D/D solubilizer at 20˚C for 2 hr to accumulate and arrest the de-aggregated chimera at the Golgi mini-stack. At 20˚C, cells were subsequently subjected to 2 hr of incubation in the presence or absence of D/D solubilizer to either de-aggregate or aggregate the cargo respectively (nocodazole was in the system throughout the procedure). Our previous work has established that secretory cargos such as VSVG are mostly arrested at the medial Golgi under 20˚C treatment (Tie et al., 2016b). In some experiments, 10 min warm up at 37˚C was conducted before imaging. Using this protocol, the re-aggregated GFP-FM4-CD8a and FM4-moxGFP Golgi puncta upon D/D washout were estimated to have 830 ± 30 (n = 184) and 660 ± 50 (n = 127) copies, respectively (Figure 5-figure supplement 3C,D). We observed that, when in the de-aggregated state, both soluble and membrane FM4-chimeras localized to the interior of Giantin-rings ( Figure 5G; Figure 5-figure supplement 3E). Intriguingly, once aggregated, they partitioned to the rim as discrete puncta. Therefore, our light microscopic data indicated that large cargos preferentially partition to the cisternal rim, possibly due to their bulky sizes, while conventional or small cargos tend to locate to the interior.

Discussion
It poses a great challenge to investigate the structure and organization of the Golgi complex by the light microscopy. We established a method to identify the cisternal rim and interior by taking advantage of rim-localized Golgi markers. In addition to quantitative axial localization using the LQ (Tie et al., 2016b), we further showed the advantage of nocodazole-induced Golgi mini-stacks in elucidating the molecular organization of the Golgi complex. We analyzed dozens of Golgi residents representing diverse families of proteins for their lateral localizations. The distribution of enzymes is Figure 5 continued transition through the Golgi mini-stack. Cells transiently co-expressing RUSH reporter, SBP-mCherry-GPI, and GFP-Golgin84 were chased in biotin and imaged live under Airyscan super-resolution microscopy. The boxed region in the upper image, which was acquired before the chase, is selected to show the time series. Arrow heads indicate the interior localization. See also Figure 5-video 1. (D-F) The partition of collagenX and mCherry-GPI to the cisternal rim and interior respectively during their intra-Golgi transport. Cells transiently co-expressing RUSH cargos, SBP-GFP-collagenX and SBP-mCherry-GPI were chased as in (A). Arrows and arrow heads indicate the cisternal rim and interior localization respectively. The intra-Golgi transport of collagenX and mCherry-GPI was demonstrated by LQ vs time plots measured from the same experiments in (E) and (F). Error bar, mean ± SEM. n, the number of Golgi mini-stacks used for the calculation. (G) GFP-FM4-CD8a partitions to the cisternal rim upon aggregation. NRK cells transiently expressing GFP-FM4-CD8a were subjected to a combination of D/D solubilizer treatment and wash out at either 20˚C or 37˚C, as indicated. First set of images is the negative control showing that aggregated GFP-FM4-CD8a was not exported from the ER. Aggregated GFP-FM4-CD8a partitioned to the rim (arrows), while non-aggregated form was still interior-localized (arrow heads). Scale bars represent 500 nm unless specified otherwise. DOI: https://doi.org/10.7554/eLife.41301.015 The following video and figure supplements are available for figure 5: restricted to the interior of the medial and trans-cisternae. In contrast, trafficking machinery components appear to complement Golgi enzymes by residing at the rim of medial and trans-cisternae, entire cis-Golgi cisternae and trans-Golgi/TGN. Previous EM studies on lateral localizations of trafficking machinery components, including COPI (Orci et al., 1997), giantin (Koreishi et al., 2013), KDEL receptor (Cosson et al., 2002;Martinez-Menárguez et al., 2001;Orci et al., 1997), GS27 (Cosson et al., 2005) and GS15 (Cosson et al., 2005), Golgi enzymes, including Man1B1 (Rizzo et al., 2013), ManII (Cosson et al., 2002;Cosson et al., 2005;Martinez-Menárguez et al., 2001;Orci et al., 2000), MGAT1  and GalT (Cosson et al., 2005), and Golgi-transiting cargos including VSVG (Martinez-Menárguez et al., 2001;Mironov et al., 2001) and soluble aggregated FM4-fusion protein (Volchuk et al., 2000), which are summarized and compared in Supplementary file 1 and 2, are mostly consistent with our observations. Our qualitative and quantitative data sketch a Golgi mini-stack as spindle-shaped with medial-cisternae possessing a larger diameter than both cis-and trans-cisternae ( Figure 4J). Our morphological description of the Golgi mini-stack, such as the spindle shape of the stack and organization of the TGN, bear similarities to the plant Golgi mini-stack observed by electron tomography (Staehelin and Kang, 2008), probably due to the lack of microtubule cytoskeleton in plants, which is similar to nocodazole-treated mammalian cells. Our findings suggest the spatial partition of the processing and transport function to the interior and rim of the Golgi stack, as depicted by our model in Figure 6.
EM studies have revealed that cisternal rims are dilated with a width of~100 nm, while their stacked interiors are narrow and tightly spaced with a width of~20 nm (Bykov et al., 2017;Engel et al., 2015;Staehelin and Kang, 2008). Recently, zipper-like intracisternal and intercisternal  Table 1) are overlaid onto a simplified diagram of a Golgi mini-stack together with the ERES and ERGIC. The red circle represents the mean of the LQ with flanking black bars representing the SEM. The cisternal interior, including central disks and inner-rings, is shaded yellow while the periphery of the Golgi mini-stack, including the cisternal rim, is shaded blue. Within the plot, red circles representing Golgi enzymes (labeled orange at the x-axis) are overlaid onto the yellow-shaded interior region, while those of components of the transport machinery (labeled black at the x-axis) are outside the mini-stack to indicate their periphery localization. DOI: https://doi.org/10.7554/eLife.41301.020 protein arrays have been discovered at interior regions of medial and trans-cisternae in green alga through the cryo-electron tomography (Engel et al., 2015). It was proposed that these tightly packed protein arrays comprise Golgi enzymes. Our super-resolution and EM data from the Golgi mini-stack provide direct evidence supporting this hypothesis. These enzyme-arrays might organize as an 'enzyme matrix' to 1) stack cisternal membrane, 2) retain Golgi enzymes or accessory proteins and 3) exclude trafficking machinery components by a possible molecular crowding mechanism. Therefore, it seems that, collectively, Golgi enzymes determine and maintain the characteristic structure of the Golgi complex. Most secretory cargos in higher eukaryotes undergo glycosylation in the Golgi complex. Our finding that the cisternal interior and rim correspond to processing and transport domain, respectively, implies that secretory cargos must access interior domains of different cisternae and then reside there long enough for sequential glycosylation. This is indeed what we observed for conventional cargos, such as GPI-anchored proteins, Furin, E-cadherin, VSVG and secretory GFP. On the other hand, these cargos probably have a sufficiently short residence time in the cisternal rim, in which they are either retrieved and retained by the 'enzyme matrix' to the interior or packed into membrane carriers targeting to the PM at the trans-Golgi. It seems that the retention by the 'enzyme matrix' occurs by default and is independent of glycosylation because secretory GFP is preferentially found within the cisternal interior. However, this is not the case for bulky cargos, such as collagenX and aggregated GFP-FM4-CD8a and FM4-moxGFP, which localized only at the rim and were excluded from the interior. These observations suggest that bulky cargos might be incompatible with the crowded molecular environment of the tightly packed 'enzyme matrix' and/or the narrow luminal space at the interior, which can have a width of <10 nm (Engel et al., 2015). Rim partitioning of large secretory cargos has previously been noted by EM (Bonfanti et al., 1998;Engel et al., 2015;Lavieu et al., 2013;Volchuk et al., 2000). Here, we directly visualized by light microscopy the size-dependent lateral partitioning of secretory cargos within the Golgi stack.
This study did not attempt to resolve different intra-Golgi trafficking models and our discoveries can be explained by both cisternal progression and stable compartment models or their modified variants. Nonetheless, our findings provide important insight into the structure and organization of the Golgi complex.

Cell lines
HeLa and normal rat kidney fibroblast (NRK) cells were from American Type Culture Collection. Cell were assumed to be authenticated by the supplier. The presence of mycoplasma contamination was monitored by Hoechst 33342 staining.

Cell culture and transfection
HeLa and NRK cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 10% fetal bovine serum. Cell transfection was conducted using Orci et al., 2000 (Invitrogen) according to manufacturer's manual. In live-cell imaging, cells grown on a F35 mm glass-bottom Petri-dish (MatTek) were imaged in CO 2 -independent medium (Invitrogen) supplemented with 4 mM glutamine and 10% fetal bovine serum at 37˚C. Unless otherwise indicated, all cells used were HeLa and treated with 33 mM nocodazole to induce the formation of Golgi mini-stacks.

Production of Giantin C-terminal antibody
It was conducted as previously described (Madugula and Lu, 2016;Mahajan et al., 2013). Briefly, His-Giantin(3131-3201) was purified in urea from bacteria and used as the antigen to raise the antiserum in rabbits (Genemed Synthesis Inc). Recombinant GST-Giantin(3131-3235) was purified from bacteria and subsequently used to purify the antibody from the anti-serum.

Super-resolution fluorescence microscopy
The Airyscan super-resolution microscope system (Carl Zeiss) comprises a Zeiss LSM710 confocal microscope equipped with an oil objective lens (alpha Plan-Apochromat 100 Â, 1.46 NA), a motorized stage, a temperature control environment chamber and Airyscan module. Fluorophores were excited by three laser lines with wavelengths of 488, 561 and 640 nm and their respective emission bandwidths were 495-550 nm, 595-620 nm and long pass 645 nm. The microscope system was controlled by ZEN software (Carl Zeiss). Pixel size of images ranged from 40 to 54 nm. For 3D imaging, the z-step of image stacks was 170 nm. Image stacks were subjected to Airyscan processing and maximal intensity projection (MIP) in ZEN software. Image analysis was performed in Fiji (https:// imagej.net/Fiji). We exhausted our images for all Golgi mini-stacks that were visually identifiable as either en face or side views.
En face averaging of golgi mini-stack images and radial mean intensity profile acquisition En face view images of Giantin-labeled Golgi mini-stacks were averaged in semi-automatic software tools that were developed using macros of Fiji. Mini-stack images were first cropped to square shape and subjected to background subtraction. To quantify the size of the Giantin-ring, we adopted the concept of the gyradius from physics. For pixel i in the Giantin-ring image, assuming that I i is its intensity and r i is its distance to the center of fluorescence mass, the gyradius of the Giantin-ring can be calculated as with all pixels of the image considered. The macro 'gyradius and intensity normalization' (see Source code 1) calculates the gyradius of Giantin in a set of multi-channel images and resizes the set of images so that the gyradius of Giantin is 100 pixels. The canvas of the image set is further expanded to 701 Â 701 pixel. Using the macro 'Golgi mini-stack alignment' (see Source code 2), Golgi marker images are aligned so that their centers of fluorescence mass are at (350, 350), the center of the image. The en face averaged Golgi mini-stack image is acquired by z-projection of these aligned images. The radial mean intensity profile is acquired using the macro 'Radial mean intensity profile' (see Source code 3). The mean intensity of all pixels within a circle around the center of the fluorescence mass is plotted against its radius (ranging from 1 to 350 pixels). The radius of a Golgi marker is defined by the half maximum position of its outer slope of the intensity plot and is normalized by the corresponding radius of Giantin. Detailed steps are described in Supplementary file 4.

Measuring diameters of Giantin-rings
To measure the diameter of a Giantin-ring, a line was first drawn across its center. In the resulting line intensity profile (Fiji), the diameter of the ring was defined as the distance between the two halfmaximum-intensity points at outer slopes.

Immunofluorescence labeling and RUSH cargo trafficking assay
These were conducted as previously described (Tie et al., 2016b). By default, tagged-proteins were transiently transfected while non-tagged proteins were native and immuno-stained by their antibodies.

Fluorescence labeling of APEX2-mediated biotinylation
Nocodazole-treated HeLa cells expressing MGAT2-APEX2-GFP were incubated with 500 mM biotin phenol for 30 min at 37˚C. Cells were subsequently transferred to ice and treated with 1 mM H 2 O 2 for 1 min with brief agitation. After extensive washing with PBS containing 10 mM sodium ascorbate (Sigma-Aldrich), 5 mM Trolox (Sigma-Aldrich), and 10 mM sodium azide (Sigma-Aldrich), cells were fixed and processed for immunofluorescence. Biotinylated proteins were labeled by Alexa Fluor 594 conjugated streptavidin.

APEX2-EM
EM was performed as previously described (Ludwig et al., 2016) with minor modifications. In brief, NRK cells transiently expressing GPP130-APEX2-GFP or MGAT2-APEX2-GFP were fixed with 2% glutaraldehyde in 0.1 M cacodylate buffer pH 7.4 (CB) containing 2 mM CaCl 2 for 1 hr on ice, rinsed three times in CB, and incubated in 0.5 mg/ml 3,3'-diaminobenzidine and 0.5 mM H 2 O 2 in CB for 5 min. Cells were washed several times in CB and post-fixed in 1% osmium tetroxide in CB containing plasmid-backbones/), which is called GFP in this study, and the copy number of FM4-moxGFP in the Golgi punctum was calculated as 10.9 Â I punctum / I Nup .

Fluorescence-conjugation of Giantin antibodies
Alexa Fluor 647 and Alexa Fluor 488 were covalently conjugated onto a commercial (BioLegend) (against the N-terminus) and our homemade (against the C-terminus) rabbit pAb against Giantin, respectively, using APEX antibody labeling kit (Invitrogen) according to the manufacturer's protocol. Data availability All data generated or analysed during this study are included in the manuscript and supporting files.