Membrane adhesion dictates Golgi stacking andcisternal morphologyIntaek Lee1,2, Neeraj Tiwari1, Myun Hwa Dunlop1, Morven Graham, Xinran Liu, and James E. Rothman2

Department of Cell Biology, Yale University School of Medicine, New Haven, CT 06520

Contributed by James E. Rothman, December 27, 2013 (sent for review November 26, 2013)

Two classes of proteins that bind to each other and to Golgimembranes have been implicated in the adhesion of Golgi cisternaeto each other to form their characteristic stacks: Golgi reassemblyand stacking proteins 55 and 65 (GRASP55 and GRASP65) andGolgin of 45 kDa and Golgi matrix protein of 130 kDa. We reporthere that efficient stacking occurs in the absence of GRASP65/55when either Golgin is overexpressed, as judged by quantitativeelectron microscopy. The Golgi stacks in these GRASP-deficientHeLa cells were normal both in morphology and in anterogradecargo transport. This suggests the simple hypothesis that the totalamount of adhesive energy gluing cisternae dictates Golgi cister-nal stacking, irrespective of which molecules mediate the adhesiveprocess. In support of this hypothesis, we show that addingartificial adhesive energy between cisternae and mitochondriaby dimerizing rapamycin-binding domain and FK506-binding pro-tein domains that are attached to cisternal adhesive proteinsallows mitochondria to invade the stack and even replace Golgicisternae within a few hours. These results indicate that al-though Golgi stacking is a highly complicated process involvinga large number of adhesive and regulatory proteins, the overrid-ing principle of a Golgi stack assembly is likely to be quite simple.From this simplified perspective, we propose a model, based oncisternal adhesion and cisternal maturation as the two core prin-ciples, illustrating how the most ancient form of Golgi stackingmight have occurred using only weak cisternal adhesive processesbecause of the differential between the rate of influx and outfluxof membrane transport through the Golgi.

tethers | GRASPs

The Golgi apparatus plays a central role in the processing,sorting, and secretion of various cargo molecules destined for

various intracellular and extracellular destinations (1). In animaland plant cells, its unique structure of four to six stacked, roughlyplanar cisternae serves, among other things, as a platform toorganize Golgi resident glycosyltransferases into distinct mem-brane-bound subcompartments (the cis-, medial-, and trans-Golgicisternae) for proper and sequential posttranslational maturationof the transiting cargo proteins (2, 3).Although these characteristic features of Golgi morphology

have drawn the attention of many researchers for many decades,the molecular mechanisms underlying them are still unclear (4).Pioneering functional reconstitution studies using a cell-free sys-tem in which Golgi stacks (but not ribbons) reassemble from mi-totic extracts (5–8) yielded two classes of purified proteins, eachclearly contributing to stacking: globular Golgi reassembly andstacking proteins (GRASPs; the homologous proteins GRASP65and GRASP55) (7, 9) and the helical rod-like and partially ho-mologous proteins Golgi matrix protein of 130 kDa (GM130) andGolgin of 45 kDa (Golgin45) (10, 11). One member of each family(GRASP65 and GM130) is located in the cis-most cisterna (7, 12),and another member of each family (GRASP55 and Golgin45) islocated in this (and more so in later cisternae) (9, 10). TheGRASP proteins bind to the Golgins (note that we use the term“Golgin” in a limited sense in this article to refer either to GM130or Golgin45) (10, 11). An appealing mechanism for intercisternaladhesion has been proposed for the GRASP proteins based on

X-ray crystallography and biochemistry that involves PSD-95,Dlg1, Zo-1 domain-dependent homo-oligomerization in trans (13–15). In recent years, with the advent of RNAi-based technologies,knock-down studies have broadly confirmed a role for GRASPproteins and Golgins in controlling Golgi morphology but havenot agreed with each other on many notable details, leaving thefield in a somewhat confused and conflictory state (10, 15–20). Inthe simpler case of Drosophila, dsRNA-mediated depletion ofdGRASP results in ∼30% loss of Golgi stacks, whereas doubledepletion of dGRASP and dGM130 was shown to unstack theGolgi, as examined by EM (21). In Schizosaccharomyces pombe,depletion of yeast GRASP homolog Grh1 has no effect on Golgistacking (22). This situation can be extended to plants, where noGRASP or evenGolgin homolog has been identified thus far (23).To address this discrepancy, we studied the relative contribu-

tion of these four “stacking factors” (GM130, Golgin45, GRASP65,and GRASP55) for Golgi stacking by extensive quantitativeanalysis of EM-based studies. The results of these experimentsstrongly indicate that the overriding principle of Golgi stackassembly is simple cisternal adhesion, regardless of whichmolecule mediates the cisternal adhesive process. We show thatadhesive energy that binds cisternae to each other at physio-logical equilibrium can be generated by many different combina-tions of Golgins+GRASPs or even in the absence of GRASPs.On the basis of this new understanding, we propose a simplemechanistic model illustrating how the most ancient form ofGolgi stacking might have been facilitated by the principles de-scribed in our adhesion model and cisternal maturation (i.e., Rabconversion) in organisms, such as yeast (S. pombe) and plants,

Significance

Stacking of cisternae in the mammalian Golgi apparatus isknown to be a highly complicated process that involves a largenumber of adhesive proteins, including Golgi reassembly andstacking proteins (GRASPs) and Golgin tethers, as well as co-ordinated disassembly/reassembly of Golgi stacks during mi-totic cell division. Using extensive quantitative analysis ofelectron microscopic data, we show here that Golgi stackingwith flattened cisternae is maintained in HeLa cells depleted ofGRASP65/55 proteins when the GRASP-deficient cells are com-plemented by exogenous expression of either Golgi matrixprotein of 130 kDa or Golgin45. These results strongly suggestthat the Golgi stack assembly and cisternal morphology are gov-erned by simple membrane adhesion at the core, explaining howGolgi stacking occurs in organisms, which do not express (or use)GRASP-type adhesion proteins such as plants and yeast.

Author contributions: I.L. and J.E.R. designed research; I.L., N.T., M.H.D., M.G., and X.L.performed research; I.L. contributed new reagents/analytic tools; I.L., N.T., and M.H.D.analyzed data; and I.L. and J.E.R. wrote the paper.

The authors declare no conflict of interest.1I.L., N.T., and M.H.D. contributed equally to this work.2To whom correspondence may be addressed. E-mail: james.rothman@yale.edu or intaek.lee@yale.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1323895111/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1323895111 PNAS | February 4, 2014 | vol. 111 | no. 5 | 1849–1854

CELL

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 2

5, 2

021

for which these four adhesive proteins either are not found orare not important for stacking.

ResultsSimultaneous Depletion of GRASP65/55 or GM130/Golgin45 DisruptsCisternal Flatness of Golgi Stacks but Does Not Result in GolgiUnstacking. A recent study reported that double knockdown ofGRASP65/55 results in complete unstacking of Golgi in HeLacells (15). To better understand the relative contribution of

Golgins and GRASPs in Golgi stack assembly and to find outwhether GRASPs are both necessary and sufficient for Golgistacking in vivo, we decided to simultaneously deplete bothGRASPs or both Golgins and study the effects on Golgi stacking,using electron microscopy. In contrast to the earlier report, weobserved that simultaneous depletion of GRASP65/55 or theirGolgin binding partners, GM130/Golgin45, results in disruptionof Golgi cisternal flatness (cf. EM photos in Fig. 1 A–C) withoutsignificant Golgi unstacking. As shown in Fig. S1, double depletion

N

G

G

B C

G

MM

A

N

G

M

D

F

0.5 0.5 m

siRNA Control GRASP65 GRASP55 GM130 Golgin45GRASP65 GRASP55

GM130 Golgin45

Max. Luminal Width(nm) 59±3(125) 73±4(132) 69±5(127) 73±4(117) 65±2(118) 155 ±7(131) 104 ±7(116)

0

50

100

0 100 200 300 400

Perc

ent

Sat

ura

on(%

)

maximum cisternal luminal width (nm)

GRASP65/55

GM130/Golgin45

KD

KD

control

maximum cisternal luminal width (nm)

0

50

100

0 100 300 400

Perc

ent

Satu

raon

(%)

E

Control GRASP65/55 KD

GM130/Golgin45 KD

GRASP55

GRASP65

Golgin45

GM130

KD

KD

KD

KD

control

200

Fig. 1. Double knockdown of GRASP65/55 or GM130/Golgin45 leads to significant disruption in Golgi cisternal flatness, but not Golgi disassembly. (A–C)Representative EM micrographs of HeLa cells treated with siRNAs against indicated adhesive proteins for 96 h. (Scale bar, 0.5 μm.) Control siRNA (A); GRASP65/55siRNA (B); GM130/Golgin45 siRNA (C). Double knockdown of the two GRASPs or two Golgins resulted in significant disruption of Golgi cisternal flatness, butwe did not observe significant Golgi unstacking under any double-knockdown conditions. For morphological criteria of Golgi unstacking, we looked forsignificant separation of cisternae frommain body of the Golgi or significant vesiculation of the Golgi at the EM level. (D and E) K-S plots showing distributionof maximum cisternal luminal width in double-knockdown cells (D) or in single-knockdown cells (E), based on luminal width measurement using ImageJsoftware. (F) Table summarizing maximum luminal width measurement for both single- and double-knockdown cells. Results are expressed as the mean ±SEM. Numbers in parenthesis indicate the number of Golgi cisternae for which luminal width was measured. Note that all analysis passed the Student t test(P < 0.05), except for Golgin45-depleted cells. G, Golgi; n, nucleus; M, mitochondria; tER, transitional endoplasmic reticulum.

GRASP65/55 siRNA → GRASP65-GFP rescue GRASP65/55 siRNA → GRASP55-GFP rescue

GM130/Golgin45 siRNA → ratGM130 rescue GM130/Golgin45 siRNA → HA-Golgin45 rescue

siRNA GM130/Golgin45 GRASP65/55

Rescue None GM130 Golgin45 None GRASP65 GRASP55

Average MaximumLuminal Width(nm)

104±7 155±771±3(187)

71±2(182)

65±3(193)

73±3(192) (131)(116)

0

GRASP55 rescue

GRASP65/55 KD

maximum cisternal luminal width (nm)

control

0

50

100

0 80 160 240 320 400

GRASP65 rescue

GRASP65/55 KD

control

0

50control

Golgin45 rescue

GM130/Golgin45 KD

0 80 160 240 320 4000

50

100

control

GM130 rescue

GM130/Golgin45 KD

0 80 160 240 320 400

maximum cisternal luminal width (nm)

Perc

ent S

atur

aon

(%)

0 80 160 240 320 400

100

50

100

A B

DC

E

Fig. 2. Disrupted Golgi cisternal flatness in the double-depletion cells can be reversed by rescue transfection. (A–D) K-S plots showing rescue of Golgi cis-ternal flatness by exogenous expression of the RNAi-resistant form of GRASP65 (A) or GRASP55 (B) in cells depleted of both GRASPs and by exogenousexpression of the RNAi-resistant form of GM130 (C) or Golgin45 (D) in cells depleted of both Golgins. Representative electron micrographs of Golgi stacks incells rescued with indicated cDNAs are shown in the right panel of each graph. (Scale bar, 0.5 μm.) (E) Table summarizing the results from rescue experiments.Results are expressed as the mean ± SEM. Numbers in parenthesis indicate the number of Golgi cisternae that luminal width was measured.

1850 | www.pnas.org/cgi/doi/10.1073/pnas.1323895111 Lee et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

5, 2

021

of GRASPs resulted in a significant reduction of Golgins, andvice versa. These phenomena suggest that Golgins and GRASPsare likely to affect the stability of each other to a certain extent(GM130–GRASP65 and Golgin45–GRASP55), mainly becauseof the role they play in the membrane association of their bindingpartners. Importantly, however, even with this additional reduc-tion in the level of other adhesion proteins, we did not observeany significant Golgi unstacking in either GRASP or Golgindouble-depleted HeLa cells. To exclude the possibility that thisdifference is a result of siRNA oligos used for GRASP65/55 de-pletion, we designed an additional set of siRNA oligos againstGRASP65 and GRASP55, respectively. The efficiency of GRASP65/55 double knockdown routinely reached above 98% for both sets ofsiRNAs against GRASP65/55 (Figs. S1 and S2). Neither siRNAcombinations led to significant unstacking of the Golgi when ex-amined under EM. We have no explanation for this discrepancy,although we also observed a significant unstacking of the Golgi ina small percentage (<1%) of the knockdown cells we analyzed.Morphologically, we noticed that Golgi membranes appeared

to have lost the characteristic flattened cisternae. As a way toquantify the change in Golgi cisternal morphology, we measuredthe maximum luminal width of the Golgi cisternae (see Fig. S1for an example). In control siRNA-transfected cells, the averagemaximum luminal width of Golgi cisternae was around 59 ± 3nm (Fig. 1F). In single-knockdown cells, we observed a small butconsistent increase (∼10–15 nm) in the average luminal width,whereas GRASP65/55 or GM130/Golgin45 double depletion ledto a two- to threefold increase in the luminal width (Fig. 1 D andE for Kolmogorov–Smirnov (K-S) plots; see Fig. S1 for fre-quency distribution graphs).We reasoned that increased cisternal width in these knockdown

cells might be a result of a block in anterograde cargo transportthrough the Golgi. To test this, we used CD8 fused to conditionalaggregation domain (24) to assess the efficiency of bulk cargotransport under these knockdown conditions. Surprisingly, wefound that GRASP65/55 double knockdown consistently led to a∼30–40% increase in CD8 transport to the plasma membranecompared with control cells and GM130/Golgin45 depleted cells(Fig. S2; see Fig. S3 for raw data), suggesting that the increasedcisternal width in these cells is not likely to be caused by accumu-lation of anterograde cargo within the lumen of the Golgi cisterna.

Additional Knockdown of Golgin45 in GRASP65/55-Depleted CellsResults in Golgi Disassembly. These findings naturally led us tothe question of whether the changes in cisternal flatness in theGRASP- or Golgin-depleted cells are associated with Golgistacking. In other words, the disrupted cisternal flatness may ormay not be indicative of the progression toward Golgi disassemblyin these cells. To find an answer to this seemingly elusive ques-tion, we tested whether additional depletion of Golgin45 orGM130 leads to Golgi disassembly in GRASP65/55-depletedHeLa cells. If so, the result would suggest that both cisternalflatness and cisternal stacking are determined by the collectivecisternal adhesive process from these GRASPs and Golgins. Insupport of this idea, we found that triple knockdown of GRASP65/55 and Golgin45 (and, to a lesser extent, GM130) in HeLa cellsresults in significant Golgi unstacking (Fig. S4 A and B). Qual-itatively, depletion of Golgin45/GRASP65/55 led to almost com-plete Golgi disassembly, but only partial Golgi unstacking wasobserved in GM130/GRASP65/55-depleted cells (compare EMphotos in Fig. S4 A and B).

Is the Disruption in Golgi Cisternal Flatness Reversible? We thenasked whether the disruption in cisternal flatness can be restoredby rescue transfection. To do this, we expressed an RNAi-resistant form of each adhesion protein in double-knockdowncells that eliminate both GRASP proteins or both Golgins.As expected, Golgi stack morphology was restored by either

exogenously expressing GRASP65 or GRASP55 in GRASPsdouble-depleted cells (Fig. 2 A and B) or by expressing eitherGolgin GM130 or Golgin45 in Golgin double-depleted cells (Fig.2 C and D; see Fig. 2E for summary). The average maximumluminal width for rescued Golgi cisternae was very close to thevalues we saw in single-knockdown cells. The knockdown andrescue expressions were confirmed by Western blots, as shownin Fig. S5. Altogether, this is consistent with the idea thatGRASP55 and GRASP65 play a complementary role in Golgistacking and cisternal morphology, as can Golgin45 and GM130,despite their differing physiological cis-trans distributions.

Are GRASP Proteins Dispensable for Golgi Stacking?As shown in thetriple-knockdown experiments (Fig. S4), there was a strong in-dication that GRASPs and Golgins may be playing functionallycomplementary roles in Golgi stacking and cisternal morphology.If this is true, is it possible that Golgins may be able to substitutefor GRASPs either entirely or partially in maintaining Golgistacking and cisternal morphology? To test this, we exogenouslyexpressed Golgins in GRASP65/55-depleted HeLa cells to seewhether Golgins can restore cisternal flatness in GRASP-depleted HeLa cells. Remarkably, crossing the functional barrier

N

G

G G

A GRASP65/55 siRNA → Golgin45 cDNA

0

50

100

controlGolgin45 cDNAGRASP65/55 KD

Perc

ent

Satu

raon

(%)

maximum cisternal luminal width (nm)

0

50

100

controlGM130 cDNAGRASP65/55 KD

Perc

ent

Sat

ura

on(%

)

maximum cisternal luminal width (nm)

0

50

100

control

Golgin97 cDNA

GRASP65/55 KD

Perc

ent

Satu

raon

(%)

maximum cisternal luminal width (nm)

D siRNA GRASP65/55

Rescue None Golgin45 GM130 Golgin97

Average MaximumLuminal Width(nm) (178)

61±2155±7(131) (181)

108±4(186)

141±5

B GRASP65/55 siRNA → GM130 cDNA

C GRASP65/55 siRNA →Golgin97 cDNA

G

G

G

G

G

0 80 160 240 320 400

0 80 160 240 320 400

0 80 160 240 320 400

Fig. 3. Morphologically and functionally normal Golgi stacks in GRASP65/55-deficient HeLa cells (A–C) K-S plots showing rescue of Golgi cisternalflatness by exogenous expression of Golgin45 (A) or GM130 (B) or Golgin97as a control (C) in cells depleted of both GRASP65/55. Representative elec-tron micrographs of Golgi stacks in cells rescued with indicated cDNAs areshown in the right panel of each graph. (Scale bar, 1 μm.) (D) Table sum-marizing the results from substitution experiments. Results are expressed asthe mean ± SEM. Numbers in parenthesis indicate the number of Golgi cis-ternae that luminal width was measured.

Lee et al. PNAS | February 4, 2014 | vol. 111 | no. 5 | 1851

CELL

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 2

5, 2

021

between GRASPs and Golgins was equally effective. Expressingeither Golgin protein in GRASPs double-depleted cells restoredthe Golgi stack morphology, as judged qualitatively by EM (Fig.3 A and B, Right). Golgi cisternal flatness was not restored byexpressing trans-Golgi network localized Golgin97 (Fig. 3C). Asshown in Fig. S6, we also performed reverse experiments, inwhich we depleted GM130/Golgin45, followed by exogenous ex-pression of GRASP55-GFP. Although the degree of restoration issmaller compared with the effect of Golgin overexpression inGRASP-depleted cells, we still found a significant restoration incisternal flatness in these cells (P < 0.05).We measured maximum cisternal luminal width to quantify the

degree of restoration (see K-S plots in Fig. 3 A–D). Exogenousexpression of Golgin45 in GRASP65/55-depleted cells resulted inGolgi with a 61 ± 2 nm average maximum cisternal luminal width(from 155 ± 7 nm in GRASP65/55-depleted cells). Compared withcontrol siRNA-transfected cells (59 ± 3 nm), there was essentiallyno significant difference. Thus, our results strongly indicate that theGRASPs and Golgin45 (and, to a lesser extent, GM130) are likelyto play a functionally redundant and complementary role for Golgistacking under physiological conditions. We measured the ante-rograde cargo transport in these cells (Fig. S6D) and found that theamount of CD8 transported to the plasma membrane in these cellsis slightly lower compared with GRASP65/55-depleted cells.

Golgin45-Dependent Restoration of Golgi Cisternal Flatness andMorphology Is Not Cell Type-Specific. To see whether theseobservations are restricted to HeLa cells, we tested humanfibrosarcoma HT1080 cells depleted of either GRASP65/55 orGM130/Golgin45 for 96 h and processed these knockdowncells for EM. We found that the morphological changes inHT1080 cells depleted of GRASPs or Golgins were essentiallyidentical (Fig. S7A). No significant unstacking of the Golgiwas observed in either knockdown cell. Exogenous expressionof Golgin45 in GRASP65/55-depleted HT-1080 for 18 hresulted in similar restoration of Golgi cisternal flatness as inHeLa cells (Fig. S7 A–C). Because HT-1080 cells are knownto secrete matrix metalloproteinase MMP-2 abundantly, theknockdown cells were tested for the secretion of endogenous

MMP-2. Strikingly, we found that both GRASP and Golgindouble-depleted HT1080 cells secreted significantly increasedamounts of endogenous MMP-2 (∼1.5-fold increase; Fig. S7 Eand F) during the 4-h secretion assays compared with thecontrol transfected cells. These results suggest that flatteningof Golgi cisternae might have a role in determining the rate ofanterograde cargo transport through the Golgi.

Can an Ectopic Adhesion Process Introduce Another Organelle intothe Golgi Stack? Our results so far have suggested that cisternaladhesion, irrespective of its molecular source, is likely to be theoverriding principle of Golgi stack assembly. If so, then addingadhesive bonds between Golgi cisternae and another organelle(mitochondria) should lead to the formation of hybrid stackswithin a few hours, in which mitochondria can be incorporatedinto preexisting interphase Golgi stacks as if they were Golgicisternae, provided the ectopic binding energy exceeds thatnormally stabilizing the Golgi stack.To engineer ectopic adhesion between Golgi cisternae and

mitochondria that can be controlled pharmacologically, we useda “knock-sideways” approach (13, 25) involving the ligand-induced formation of heterodimers between the FK506-bindingprotein (FKBP) and a rapamycin-binding domain (FRB). Thesetwo proteins bind to each other when a rapamycin analog(AP21967, which simultaneously binds to the FRB and FKBPdomains) is added to cells. The Golgi adhesion protein GRASP55was expressed as a hybrid protein with FKBP (and a fluorescentprotein [tag blue fluorescent protein (tagBFP)]) to enable visu-alization (Fig. S8A). FRB was expressed as a hybrid with a pro-tein targeted to the mitochondrial outer membrane (OMP25)also tagged with the myc epitope. In this way, the expressedGRASP55 hybrid protein should relocalize from the Golgi (andcytosol if present in excess) to the mitochondria when the ligandAP21967 is added to the cells (Fig. S8B). More specifically, en-dogenous GRASPs were replaced by the GRASP55 hybridprotein by using GRASP double-depleted cells that also expressan RNAi-resistant form of the GRASP55-FKBP-tagBFP hybrid.To simplify the analysis and focus on the effects of ectopic ad-hesion on the stack (compared with the ribbon), these cells were

G65/55 RNAi+GRASP55 KS (Noc+2 M drug;30min)

G65/55 RNAi+GRASP55 KS

G65/55 RNAi+GRASP55 KS (+Noc )

1 1

(Noc+2 M drug;60min) G65/55 RNAi+GRASP55 KS

(+ Noc→drug;120min)

M

N

M

MG

G

BA

DC

G

G

MM

G

Fig. 4. Golgi–mitochondria hybrid stack formation driven by engineered ectopic adhesive interaction Representative electron micrographs from knock-sideways experiments. HeLa cells treated with siRNAs against GRASP65/55 for 72 h were rescued with GRASP55-FKBP-tagBFP for 24 h, followed by 3 h oftreatment with Nocodazole (A); GRASP55-FKBP-tagBFP-rescued cells treated with AP21967 for indicated time to rapidly target the exogenous GRASP55 fusionprotein onto mitochondria, which leads to Golgi-mitochondria hybrid stacks (B–D). G, Golgi; n, nucleus; M, mitochondria.

1852 | www.pnas.org/cgi/doi/10.1073/pnas.1323895111 Lee et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

5, 2

021

treated with nocodazole to produce ministacks before initiatingknock-sideways with AP21967.Electron microscopy was then used to follow the fate of pre-

existing Golgi stacks and mitochondria (Fig. 4 A–D). The resultwas a dramatic reorganization of the Golgi stacks within 30–60min after addition of ligand (2 μM), in which mitochondriacluster around ministacks at earlier times (Fig. 4B) and theninvade them at later times, often forming hybrid stacks (Fig. 4 Cand D). In some cases, Golgi cisternae appeared to wrap aroundmitochondria similar to a hot dog bun wrapping around a sau-sage. The diversity of the resulting hybrid organelles makes itimpossible to quantify the results. However, it is nonethelessclear that these results support the idea that established Golgistacks can be dissected by ectopic adhesion to mitochondria

within a few hours. This implies that the stacks had been stabi-lized by reversible adhesion that could then be overcome atequilibrium by an engineered alternative.

DiscussionIn this study, we report that cisternal adhesion is likely to bea key governing principle that dictates the two characteristicfeatures of the Golgi apparatus (flattened cisternae and cisternalstacking) as illustrated in Fig. 5A along with EM photos, showingexamples of Golgi stacks at different stages of cisternal adhesion,and although GRASPs seem to play an important role in Golgistacking under physiological conditions, they are not absolutelyessential even in mammalian cells in the sense that they could bereplaced with Golgins, as shown in our experiments. This implies

A

ER ER

RabA

RabB

RabB

RabA

RabA

PM PM

Rab conversion (cisternal matura�on)

Rate of membrane transport

cisternal growth cisternal matura�on

No stacking

or

cisternal growth cisternal matura�on

cisternal stacking

B

ER

RabA

Total amount of cisternal adhesion

physiological

Mito�c Golgi haze

Reassembly of cisternae by SNARE-dependent membrane fusion

RabBRabB

Fig. 5. Illustrations describing the concept of the adhesion model and the simplest form of cisternal stacking that could occur via cisternal adhesion and Rabconversion as two fundamental principles of stacking mechanism. (A) Illustration explaining diverse morphologies of Golgi stacks based on the adhesionmodel: Golgi cisternae are analogous to a water-filled balloon that is densely covered with Velcro-like molecules (tethering proteins, GRASPs, or other ad-hesive proteins). At the low adhesive condition, two of these water-filled balloons (cisternae) would be barely adhered together. As the adhesive energy isprogressively increased broadly across the surface of cisternae, the cisternae gets flattened as the cisternal adhesive strength overcomes the osmotic pressurewithin the cisternae. Examples of Golgi EM photos for progressively adhesive conditions are shown here. (Scale bar, 500 nm.) (B) Illustration depicting thesimplest form of cisternal stacking, based on the adhesion model and Rab conversion. According to this model, cisternal stacking occurs because of signif-icantly faster ER-to-Golgi anterograde cargo transport compared with cargo export from the trans-Golgi to the plasma membrane. Only weak tethering oradhesive processes may suffice if the rate of the ER-Golgi transport far exceeds that of Golgi–plasma membrane transport. Cisternal maturation initiates thestacking of newly forming cis-cisternae to now more distal cisternae.

Lee et al. PNAS | February 4, 2014 | vol. 111 | no. 5 | 1853

CELL

BIOLO

GY

Dow

nloa

ded

by g

uest

on

Mar

ch 2

5, 2

021

that Golgi stacking may be possible even in organisms lackingGRASP-type adhesive proteins.In the case of yeast (S. pombe) and plants, in which none of the

homologs for these four proteins have been either identified orfound to be necessary for Golgi stacking, similar adhesive pro-teins or new adhesive interactions of existing Golgi proteinscould potentially mediate cisternal adhesion. It is possible thateven weak tethering processes between neighboring cisternacould promote cisternal stacking theoretically (i.e., multipletether–Rab binding, tether–SNARE binding, etc). In Fig. 5B, wepropose a model illustrating how the most primitive form ofcisternal stacking might have occurred by this kind of weaktethering (adhesion) processes and Rab conversion, a compart-mental maturation process, in which one Rab-GTP recruits theRab–GTP exchange factor for the subsequent Rab protein and isknown to be well-conserved through the evolution from yeast tomammals (26–29). This model requires that the cargo exportfrom the trans-Golgi be slower than cargo import into the cis-Golgi. In this case, the slower the older (distal) cisternae dis-sipates from the trans side of the Golgi via cargo export to theplasma membrane, the better the chance of cisternal stackingbecause newly formed cis-cisternae would accumulate at the cisface of now more distal cisternae. Naturally, the number of cis-ternae per stack would be determined by dynamic ratio of cargoinflux/outflux through the Golgi in a given cell type.One of the reasons that newly formed nascent cis-cisternae

may stack with older cis-cisternae (rather than fuse with oldercis-cisternae) could be cisternal maturation (i.e., Rab conver-sion) of older cis-cisternae. Thus, newly forming cis-cisternaecontinues to grow in volume as a result of rapid membrane influxfrom the endoplasmic reticulum (ER), until Rab conversion (i.e.,RabA → RabB) is fully completed, at which point the maturedcisternae begins to stack with more nascent cis-cisternae. In thisway, the Golgi stack could be built from the trans side of the Golgitoward the cis side in terms of a purely mechanistic point of view.

If one assumes that Golgi stacking could occur as describedhere, why do these Golgins and GRASPs play crucial roles inmammalian and fly Golgi stacking under physiological con-ditions? The answer to this question may be found from thefact that yeast and plant Golgi stacks do not undergo mitoticdisassembly/reassembly, whereas mammalian and fly Golgi stacksdo. It is possible that the necessity for rapid mitotic disassembly/reassembly of Golgi stacks might have brought about the evolu-tionary adaptation of this particular group of adhesive proteins forGolgi stacking in mammals and fly. Thus, yeast and plants may beusing the more ancient strategy (i.e., multiple weak adhesive pro-cesses) for building stable Golgi stacks, whereas mammals and flieshave adapted a small group of proteins for Golgi stacks that can bedynamically disassembled and reassembled in sync with mitoticcell division.

Materials and MethodsReagents and Antibodies. All common reagents were purchased from Sigma-Aldrich unless otherwise mentioned. The following antibodies were used:mouse monoclonal GM130 (BD Transduction Laboratories), goat polyclonalGRASP65 (Santa Cruz Biotechnology), rabbit polyclonal GRASP55 (Pro-teintech Inc.), rabbit polyclonal p115 (Santa Cruz Biotechnology), HRP-con-jugated mouse β-actin antibody (GenScript). Rabbit polyclonal Golgin97(AbCAM). Rabbit polyclonal Golgin45 antibody was made by injecting syn-thetic Golgin45 peptide (AA40-53) conjugated to KLH (GenScript). pcDNA-hGRASP55-GFP was kindly provided by Adam Linstedt (Carnegie MellonUniversity). Rat GM130 cDNA was obtained from Nobuhiro Nakamura(Kyoto Sangyo University). Human Golgin45 cDNA was purchased fromAddgene. Human GRASP65 cDNA was from Yanzhuang Wang (University ofMichigan, Ann Arbor, MI). Human Golgin97 cDNA was obtained from SeanMunro (MRC Laboratories, Cambridge, United Kingdom).

See SI Materials and Methods for further information.

ACKNOWLEDGMENTS. We thank Adam Linstedt, Gregory Lavieu, NobuhiroNakamura, Yanzhuang Wang, and Sean Munro for kindly sharing reagentsand Stuart Kornfeld for critical reading of the manuscript. We are especiallygrateful to Vivek Malhotra for insightful inputs and discussion during thepreparation of this manuscript.

1. Nakamura N, Wei JH, Seemann J (2012) Modular organization of the mammalianGolgi apparatus. Curr Opin Cell Biol 24(4):467–474.

2. Warren G, Malhotra V (1998) The organisation of the Golgi apparatus. Curr Opin CellBiol 10(4):493–498.

3. Kornfeld R, Kornfeld S (1985) Assembly of asparagine-linked oligosaccharides. AnnuRev Biochem 54:631–664.

4. Wang Y, Seemann J (2011) Golgi biogenesis. Cold Spring Harb Perspect Biol 3(10):a005330.

5. Rabouille C, Misteli T, Watson R, Warren G (1995) Reassembly of Golgi stacks frommitotic Golgi fragments in a cell-free system. J Cell Biol 129(3):605–618.

6. Rabouille C, Levine TP, Peters JM, Warren G (1995) An NSF-like ATPase, p97, and NSFmediate cisternal regrowth from mitotic Golgi fragments. Cell 82(6):905–914.

7. Barr FA, Puype M, Vandekerckhove J, Warren G (1997) GRASP65, a protein involved inthe stacking of Golgi cisternae. Cell 91(2):253–262.

8. Shorter J, Warren G (1999) A role for the vesicle tethering protein, p115, in the post-mitotic stacking of reassembling Golgi cisternae in a cell-free system. J Cell Biol 146(1):57–70.

9. Shorter J, et al. (1999) GRASP55, a second mammalian GRASP protein involved in thestacking of Golgi cisternae in a cell-free system. EMBO J 18(18):4949–4960.

10. Short B, et al. (2001) A GRASP55-rab2 effector complex linking Golgi structure tomembrane traffic. J Cell Biol 155(6):877–883.

11. Barr FA, Nakamura N, Warren G (1998) Mapping the interaction between GRASP65and GM130, components of a protein complex involved in the stacking of Golgi cis-ternae. EMBO J 17(12):3258–3268.

12. Lowe M, et al. (1998) Cdc2 kinase directly phosphorylates the cis-Golgi matrix proteinGM130 and is required for Golgi fragmentation in mitosis. Cell 94(6):783–793.

13. Sengupta D, Truschel S, Bachert C, Linstedt AD (2009) Organelle tethering by a ho-motypic PDZ interaction underlies formation of the Golgi membrane network. J CellBiol 186(1):41–55.

14. Bachert C, Linstedt AD (2010) Dual anchoring of the GRASP membrane tether pro-motes trans pairing. J Biol Chem 285(21):16294–16301.

15. Xiang Y, Wang Y (2010) GRASP55 and GRASP65 play complementary and essentialroles in Golgi cisternal stacking. J Cell Biol 188(2):237–251.

16. Feinstein TN, Linstedt AD (2008) GRASP55 regulates Golgi ribbon formation. Mol BiolCell 19(7):2696–2707.

17. Duran JM, et al. (2008) The role of GRASP55 in Golgi fragmentation and entry of cellsinto mitosis. Mol Biol Cell 19(6):2579–2587.

18. Marra P, et al. (2007) The biogenesis of the Golgi ribbon: The roles of membraneinput from the ER and of GM130. Mol Biol Cell 18(5):1595–1608.

19. Puthenveedu MA, Bachert C, Puri S, Lanni F, Linstedt AD (2006) GM130 and GRASP65-dependent lateral cisternal fusion allows uniform Golgi-enzyme distribution. Nat CellBiol 8(3):238–248.

20. Ramirez IB, Lowe M (2009) Golgins and GRASPs: Holding the Golgi together. SeminCell Dev Biol 20(7):770–779.

21. Kondylis V, Spoorendonk KM, Rabouille C (2005) dGRASP localization and function inthe early exocytic pathway in Drosophila S2 cells. Mol Biol Cell 16(9):4061–4072.

22. Levi SK, Bhattacharyya D, Strack RL, Austin JR, 2nd, Glick BS (2010) The yeast GRASPGrh1 colocalizes with COPII and is dispensable for organizing the secretory pathway.Traffic 11(9):1168–1179.

23. Hawes C, Schoberer J, Hummel E, Osterrieder A (2010) Biogenesis of the plant Golgiapparatus. Biochem Soc Trans 38(3):761–767.

24. Volchuk A, et al. (2000) Megavesicles implicated in the rapid transport of intra-cisternal aggregates across the Golgi stack. Cell 102(3):335–348.

25. Robinson MS, Sahlender DA, Foster SD (2010) Rapid inactivation of proteins by ra-pamycin-induced rerouting to mitochondria. Dev Cell 18(2):324–331.

26. Matsuura-Tokita K, Takeuchi M, Ichihara A, Mikuriya K, Nakano A (2006) Live imagingof yeast Golgi cisternal maturation. Nature 441(7096):1007–1010.

27. Losev E, et al. (2006) Golgi maturation visualized in living yeast. Nature 441(7096):1002–1006.

28. Rink J, Ghigo E, Kalaidzidis Y, Zerial M (2005) Rab conversion as a mechanism ofprogression from early to late endosomes. Cell 122(5):735–749.

29. Rivera-Molina FE, Novick PJ (2009) A Rab GAP cascade defines the boundary betweentwo Rab GTPases on the secretory pathway. Proc Natl Acad Sci USA 106(34):14408–14413.

1854 | www.pnas.org/cgi/doi/10.1073/pnas.1323895111 Lee et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

5, 2

021