Arabidopsis COG Complex Subunits COG3 and COG8 Modulate Golgi Morphology, Vesicle Trafficking Homeostasis and Are Essential for Pollen Tube Growth

Spatially and temporally regulated membrane trafficking events incorporate membrane and cell wall materials into the pollen tube apex and are believed to underlie the rapid pollen tube growth. In plants, the molecular mechanisms and physiological functions of intra-Golgi transport and Golgi integrity maintenance remain largely unclear. The conserved oligomeric Golgi (COG) complex has been implicated in tethering of retrograde intra-Golgi vesicles in yeast and mammalian cells. Using genetic and cytologic approaches, we demonstrate that T-DNA insertions in Arabidopsis COG complex subunits, COG3 and COG8, cause an absolute, male-specific transmission defect that can be complemented by expression of COG3 and COG8 from the LAT52 pollen promoter, respectively. No obvious abnormalities in the microgametogenesis of the two mutants are observed, but in vitro and in vivo pollen tube growth are defective. COG3 or COG8 proteins fused to green fluorescent protein (GFP) label the Golgi apparatus. In pollen of both mutants, Golgi bodies exhibit altered morphology. Moreover, γ-COP and EMP12 proteins lose their tight association with the Golgi. These defects lead to the incorrect deposition of cell wall components and proteins during pollen tube growth. COG3 and COG8 interact directly with each other, and a structural model of the Arabidopsis COG complex is proposed. We believe that the COG complex helps to modulate Golgi morphology and vesicle trafficking homeostasis during pollen tube tip growth.


Introduction
In flowering plants, pollen tubes grow through the style and deliver male gametes to ovules through highly polarized growth of the tips caused by cell expansion occurring exclusively at the apex [1]. Pollen tubes can grow rapidly under both in vivo and in vitro conditions where actomyosin-dependent reverse fountain-like cytoplasmic streaming efficiently drives vesicles into the clear zone of a growing pollen tube [2]. Most of these vesicles fuse with the apex membrane and deposit cell wall materials, membrane lipids, and proteins to support growth. This deposition must be regulated temporally and spatially to balance turgor pressure and cell wall extensibility [3,4]. The massive amount of exocytosis that occurs at the pollen tube tip was calculated to exceed the requirements for maintaining growth rates, and was suggestive of underlying endocytosis and recycling processes [5,6,7,8]. How these processes are coordinated during pollen tube growth remains unclear.
Key components of the vesicle trafficking machinery which operates during pollen tube growth are being characterized [4,9]. Small GTPase NtRAB11B labeled transport vesicles in the apical inverted cone of the growing pollen tube, and play a role in secretory vesicle delivery and possibly vesicle recycling [10]. A knockout mutation of the Arabidopsis pollen-specific RABA4D gene impaired selective targeting of cell wall materials and pollen tube guidance [11]. In addition to Rab GTPases, several mutants of Arabidopsis exocyst subunits exhibited short and swollen pollen tubes, and the tip-localization patterns of EXO70A1, SEC6, and SEC8 were suggestive of a role in polarized exocytosis or recycling at the tips [12,13,14]. Recently, pollen-specific GNL2 was shown to be essential for pollen tube tip growth based on its necessary role in polar recycling [15]. These results demonstrated that the tip-focused delivery of exocytic and recycling vesicles is crucial for polarized and directional pollen tube growth. In addition, a dominant negative form of tobacco NtRAB2 blocked secretory protein trafficking and arrested pollen tube growth [16] indicative of the importance of the endoplasmic reticulum (ER)-Golgi early secretory pathway.
The ER-to-Golgi secretion pathway mediated by the coat protein complex II (COPII) is thought to be counter-balanced by COPI-mediated retrograde trafficking [17,18]. In yeast and mammalian cells, the conserved oligomeric Golgi (COG) complex, which is an octameric tethering complex, is involved in COPI-mediated, intra-Golgi retrograde transport of Golgi-resident proteins such as glycosyltransferases, which are enzymes that glycosylate proteins and lipids [19,20,21,22]. In addition, the COG complex is required for the integrity of the mammalian Golgi apparatus [21,23].
Homologs of each of the COG complex subunits have been identified in the Arabidopsis genome [24] and its presumed partners in COPI-mediated intra-Golgi trafficking, such as COPI vesicles, were observed predominantly at the periphery of cis-and medial-cisternae of plant Golgi [25,26]. Moreover, COPI components and the GTPase ADP ribosylation factor 1 (Arf1), which is responsible for COPI coat assembly and dissociation, have been identified in plants [17,27,28,29,30]. COPI is essential for COPII-mediated anterograde trafficking in tobacco leaf epidermal cells [17]. Aside from the involvement of a presumed association between COPI and the COG complex in intra-Golgi trafficking, little is known regarding the effects of COG complex disruption on COPI-mediated trafficking in plants.
The Golgi apparatus in plant cells consists of numerous individual stacks moving over the ER with the aid of actin-myosin motors. This arrangement differs from the perinuclear Golgi stacks which are arranged side-by-side in mammalian cells [18,31,32]. Understanding the mechanisms by which the plant Golgi structure is maintained, despite its rapid movement and exchange of proteins with other organelles, is important. Long coiled-coil protein golgins have been implicated in the maintenance of Golgi integrity in mammalian cells by linking adjacent membranes in vesicle docking and by linking adjacent cisternae [33]. Among plant golgin homologs that localize to the Golgi [24,34,35,36], Atp115/GC6 [36] and AtCASP [34,35] may function as vesicle tethering factors, but the roles of golgins in plant Golgi structure maintenance have not been demonstrated, nor have the roles of the multi-subunit tethering complexes, TRAPPI/II and COG complex [18].
The Arabidopsis embryo yellow (eye) mutant, in which the COG7 gene is disrupted, exhibits abnormal embryo color and development, mislocalization of ERD2 to the ER, and alterations in the size of the Golgi apparatus [37]. Moreover, the COG complex was implicated in penetration resistance of barley to barley powdery mildew fungus [38]. Our current understanding of COG complex function in plant development remains limited. How Golgi functions dependent on COG complex coupled to plant development, particularly pollen tube tip growth is a very interesting question to answer.

Results
Mutations in COG3 and COG8 genes cause male sterility COG3 is interrupted in intron 21 by a T-DNA insertion in the Arabidopsis line GK_498G10, while COG8 is interrupted by a T-DNA insertion in exon 10 in the Salk_122096 line ( Fig 1A). No homozygous cog3 or cog8 mutant plants were identified using PCR-based genotyping with combinations of different gene-specific primers and a T-DNA border primer (n>300). The progeny from self-pollinated cog3 -/+ and cog8 -/+ plants all segregated at a ratio of roughly 1:1 instead of the expected 3:1 ratio. For 134 heterozygous and 165 wild-type progeny of the cog3 -/+ mutant, a χ 2 test for a 1:1 ratio gave a value of 3.21, P < 0.05. For 181 heterozygous and 174 wild-type progeny of the cog8 -/+ mutant, a χ 2 test for a 1:1 ratio gave a value of 0.14, P < 0.05 (Table 1). Pollination of wild-type plants with pollen from cog3 -/+ or cog8 -/+ mutant plants resulted only in offspring with no insertion (Table 1) indicating that the two mutations cannot be transmitted by the male gametophytes. In contrast, when cog3 -/+ and cog8 -/+ were used as recipients in crosses with wild-type pollen, approximately 45% [113/(113+138)] of cog3 -/+ and 44% [94/(94+120)] of cog8 -/+ progeny contained the T-DNA insertion indicating that the T-DNA insertions in COG3 and COG8 had no discernible effect on female gametophyte transmission.
cog3 and cog8 mutants are defective in pollen tube growth in vitro and in vivo products because they do not separate from each other [39]. A quartet from a resulting cog3 -/+ qrt1 -/plant had two cog3 qrt1 (mutant) and two qrt1 (wild-type) pollen grains. The same genotypic frequency applied to a cog8 -/+ qrt1 -/quartet. As shown in S1 Fig, Alexander staining showed that almost all cog3 -/+ qrt1 -/and cog8 -/+ qrt1 -/quartets had four viable pollen grains  (S1 Fig). Scanning electron microscopy (SEM) and 4 0 ,6-diamidino-2-phenylindole (DAPI) staining revealed that these pollen grains were normal in appearance (S1 Fig) and contained two sperm cells and one vegetative cell (S1 Fig). Therefore, pollen grain development was not affected by the cog3 or cog8 mutations.
We examined in vitro germination and pollen tube growth of cog3 -/+ qrt1 -/and cog8 -/+ qrt1 -/quartets after incubation on solid media for 6 h at 22°C. No more than two normal pollen tubes were observed in the mutants (Fig 1D-1I) in contrast with the wild-type (Fig 1B), while many ruptured, short, swollen, and wavy pollen tubes were observed in the mutants ( Fig  1D-1I), but not in the wild-type (Fig 1C). To analyze the defective pollen tube growth statistically, we established counting criteria. Quartets producing zero or one pollen tube were excluded. For quartets with two normal pollen tubes, the phenotypes of the other two pollen were counted. For example, in Fig 1D, the two pollen with normal tubes were not counted and the two pollen with burst tubes were counted as 'ruptured'. Approximately 26% of cog3 -/+ and 25% of cog8 -/+ counted pollen tubes ruptured immediately after germination or after growth. In addition, approximately 29% of cog3 -/+ and 32% of cog8 -/+ counted pollen tubes were short or swollen and wavy and were classified as 'deformed'. Therefore, the amounts of both 'ruptured'-and 'deformed'types of pollen increased significantly (P < 0.001) ( Fig 1J). In the meantime, the germination rates of counted pollen from cog3 -/+ qrt1 -/-(n = 712) and cog8 -/+ qrt1 -/-(n = 716) were approximately 56 and 58%, respectively, which was comparable to that of qrt1 -/pollen (50%). The same types of defects with similar ratios (S2 Fig) were observed for cog3 -/+ and cog8 -/+ heterozygotes after germination for 4 h on solid media and the mutant pollen tubes grew significantly slower than the wild-type (S3 Fig). Taken together, in vitro germination assays indicated that mutations in COG3 and COG8 genes caused unstable, slow growing, and morphologically abnormal pollen tubes.
The  Fig 1K) and pollen tubes (Fig 1L), which was consistent with the observed phenotypes.
We examined in vivo pollen tube growth using artificial pollination of cog3 -/+ qrt1 -/and cog8 -/+ qrt1 -/pollen grains onto ms1 -/mutant pistils. The ms1 mutation affects the early stage of pollen development and thus provides readily available unpollinated stigmas [40]. After 16 h of pollination, approximately 68% (n = 112) of qrt1 -/quartets grew three or four normal pollen tubes into the pistil as revealed by aniline blue staining (Fig 2A). In contrast, none of the cog3 -/+ qrt1 -/-(n = 73) or cog8 -/+ qrt1 -/-(n = 56) quartets germinated three or four normal pollen tubes; the majority germinated two normal pollen tubes plus one or two aberrant pollen tubes which did not grow into pistils (Fig 2B and 2C). COG3 and COG8 mutations disrupted pollen tube growth both in vitro and in vivo.

1.86
gCOG8/gCOG8 pollen tubes and plants grew normally as the wild-type (S5 Fig). These results showed that loss function of COG3 and COG8 was responsible for the mutant phenotypes of cog3 -/+ and cog8 -/+ plants, respectively.

COG3 and COG8 are localized to the Golgi apparatus in pollen
Determining the localization of COG3 and COG8 proteins in pollen would increase our understanding of the relationship between COG complex function and pollen tube tip growth. The pro-LAT52:COG3-GFP and proLAT52:COG8-GFP fusion proteins were functional because they rescued their corresponding mutants (S2 Fig and Table 2); thus, their subcellular localization in pollen should reflect COG3 and COG8 localization. COG3-GFP and COG8-GFP exhibited punctate fluorescence signals in pollen grains and tubes ( Fig 4A) and moved with the reverse fountaintype of cytoplasmic streaming as shown by live cell imaging (S1 and S2 Movies), suggestive that they are associated with the Golgi apparatus. To determine the nature of the compartments with punctuate signals, the COG3-GFP and COG8-GFP plants were crossed with transgenic lines expressing the LAT52 promoter-driven Golgi marker protein RPA with its C-terminus fused to DsRed (RPA-DsRed) [41,42] or ER marker mCherry-HDEL [43]. The green fluorescence signals of COG3-GFP and COG8-GFP exhibited the same localization pattern as the red fluorescence signals of RPA-DsRed ( Fig 4A) but not of mCherry-HEDL (S6 Fig) indicating that COG3 and COG8 were indeed present in the Golgi apparatus before and after germination.
To verify the subcellular localization of COG8, we generated pro35S:COG8-GFP Arabidopsis transgenic plants. Immunogold labeling with anti-GFP antibodies revealed that COG8-GFP was located at the Golgi apparatus, in particular, it was targeted to the periphery of the Golgi cisternae (Fig 4B and 4C), which resembled γ-COP localization [25]. While it was not present at other organelles, such as the ER and the multi-vesicular body (MVB) (Fig 4D). And applying only the secondary antibodies gave no signal (Fig 4E).
Altered morphology of Golgi labeled by GFP-EMP12 and GAUT14-GFP in cog3 and cog8 pollen Because COG3 and COG8 localized to the Golgi in pollen (Fig 4), we analyzed the morphology of the Golgi apparatus in cog3 -/+ and cog8 -/+ plants using confocal microscopy. The Golgi marker constructs proUBQ10:GFP-EMP12 [44] and proPPME1:GAUT14-GFP [45] were first transformed into wild-type plants, respectively. Individual GFP-EMP12 fluorescence spots looked sharp, discrete and well-distributed in wild-type pollen grains, appeared either as discs or bars depending on the viewing angle (Fig 5A), typical of Golgi morphology [46]. However, after crossing, approximately half of pollen grains (cog3 -/+ : 67/114; cog8 -/+ : 83/148) showing fuzzy, rounded Golgi bodies tend to clump together in a blurry background, which were readily discernible from wild-type, and were counted as mutant pollen grains (Fig 5A). Length to width ratio (length/width) of GFP-EMP12 signal of all distinguishable Golgi bodies in five pollen grains was measured for each genotype (Fig 5B and 5D). Pollen with that in a range of 1.0 to 2.0 is about 14% in wild-type, 91% in cog3, 84% in cog8. In contrast, pollen with that of 2.0 to >3.5 is about 86% in wild-type, 9% in cog3, 16% in cog8 ( Fig 5B). These data demonstrated that Golgi morphology was dramatically altered in cog3 and cog8 pollen. GAUT14-GFP-labeled Golgi [45] also exhibited similar morphological disruption in about half of the pollen grains (cog3 -/+ : 63/143; cog3 -/+ : 65/138), although GAUT14-GFP fluorescent bars were generally shorter than those of GFP-EMP12 in wild-type cells (Fig 5A and 5C).

Change of Golgi apparatus morphology in cog3 and cog8 pollen
To further explore abnormalities in Golgi morphology, ultrathin sections of high-pressure frozen/freeze-substituted cog3 -/+ qrt1 -/and cog8 -/+ qrt1 -/pollen grains were used for transmission electron microscopy (TEM) observation. Wild-type and mutant pollen within a quartet were readily discernible based on differences in Golgi morphology. In wild-type pollen, the Golgi apparatus exhibited normal morphology with clear cis-trans polarity (Fig 6A and 6B), while in cog3 and cog8 pollen, all Golgi observed were disrupted, and were named mini-stacks as shown in Fig 6C-6F and as is more clearly illustrated in Fig 5H. In this case, the number of cisternae (five or six) was not significantly different from that of wild-type, but a few obvious changes were noticed and characterized. First, clear recognition of the cis-trans polarity was difficult. Second, the lengths of the mini-stacks cisternae were reduced significantly with the longest cisterna lengths decreased from approximately 638 nm to 339 nm in cog3 and to 347 nm in cog8 and the shortest cisterna lengths decreased from approximately 293 nm to 244 nm in cog3 and to 255 nm in cog8 (Fig 6I). Concomitantly, the widths of the mini-stack cisternae increased from approximately 195 nm to 267 nm in cog3 and to 282 nm in cog8 (Fig 6J) as the resulted of increased widths of individual cisternae (Fig 6K). The overall changes in mini-stacks morphology were consistent with the changes in Golgi morphology revealed by GFP-EMP12 and GAUT14-GFP fluorescence (Fig 5). These observations suggested that COG3 and COG8 play a role in keeping normal Golgi morphology. Partial loss of the association of the COPI subunit γ-COP and its cargo protein EMP12 with the Golgi apparatus in cog3 and cog8 pollen One possible mechanism by which the COG complex maintains Golgi integrity is the tethering of COPI vesicles, which appear at the periphery of Golgi cisternae [25,26]. γ-COP is a subunit of the COPI coat, which has been shown to localize to the Golgi in plant cells [25]. We introduced proLAT52:γ-COP-mCherry and proUBQ10:GFP-EMP12 constructs into wild-type, cog3 -/+ , and cog8 -/+ backgrounds by crossing with plants harboring either of the constructs. In wild-type, double-labeled pollen, typical disc-or bar-like Golgi labeling was observed for both GFP-EMP12 and γ-COP-mCherry and the fluorescent signals merged well (Fig 7A, n = 43). In about half of cog3 -/+ and cog8 -/+ mutant pollen (32/68, cog3 -/+ ; 28/59, cog8 -/+ ), however, the Golgi labeled by GFP-EMP12 and γ-COP-mCherry exhibited altered morphology in a fuzzy background (Fig 7A). The diffuse GFP-EMP12 fluorescence was indicative of its disassociation from the Golgi apparatus, while γ-COP signals may represent free coatomer complexes in the cytosol (Fig 7A). The Golgi-resident protein EMP12 interacts via its C-terminal KXD/E motif with the COPI subunit γ-COP and thereby achieves steady state cis-and medial-Golgi localization by retrograde transport of COPI vesicles [44]. We performed immunogold electron microscopy analysis using anti-EMP12 antibodies on ultrathin sections prepared as described above. Consistent with the published data [44], the majority of anti-EMP12 antibody conjugated gold particles in wild-type pollen grains appeared at the cis-and medial-Golgi (Fig 7B,  blue arrows). In cog3 and cog8 pollen grains, however, more gold particles (compare Fig 7B to  Fig 7C-7E, red arrows) were found scattered in the cytoplasm distant from the mini-stacks. These data indicated that a large proportion of COPI vesicles containing EMP12 proteins lost their tight association with the Golgi and were released into the cytoplasm in cog3 and cog8 mutant pollen presumably due to disassembly of the COG complex containing COG3 and COG8 subunits.
Cell wall materials and proteins required for pollen tube growth are deposited incorrectly in cog3 and cog8 pollen tubes The COG3 and COG8 loss-of-function mutants exhibited restricted pollen tube growth (S3 Fig) and bulged, wavy, and ruptured pollen tubes (Fig 1), suggestive of altered deposition of cell wall components. Pectin is the major component of the pollen tube cell wall. Ruthenium red, which stains a broad range of methylesterified pectins, strongly labeled the tip region of wild-type pollen tubes (Fig 8A.1). In contrast, cog3 and cog8 pollen tubes exhibited lighter staining in confined regions in the deformed or branched pollen tube tips (Fig 8A.2-8A.4) and pectins appeared to accumulate in the cytosol when the pollen tubes were ruptured (Fig 8A.5).
To examine the distribution of homogalacturonan (HG) with high and low degrees of methylesterification, we used the monoclonal antibodies JIM7 and JIM5, respectively [47]. The JIM7 antibody labeled the crescent of wild-type pollen tubes (Fig 8B.1) with a pattern similar to that previously reported for JIM7 labeling of pollen tubes [11,48]. In some cog3 and cog8 pollen tubes, JIM7 signals were not tip-localized, but were instead distributed over the distal bulging tips (Fig 8B.2). Notably, JIM7-positive dots were present in intact (Fig 8B.2-8B.3) or ruptured (Fig 8B.5) pollen tubes indicating reduced secretion efficiency for JIM7-positive pectins. In addition, JIM7 signals were observed in a much smaller tip region at the curling tips (Fig 8B.4).  In wild-type pollen tubes, JIM5 fluorescence was strong and uniform along the length of the pollen tube, except at the crescent (S7 Fig). In cog3 and cog8, JIM5 labeling was essentially homogeneous along the entire pollen tube length (S7 Fig). Callose is a component of the pollen tube inner wall. In wild-type pollen tubes, aniline bluelabeled callose was present along the flanks of the tube and was excluded from the pollen tube apex (Fig 8C.1). In the cog3 and cog8 mutants, callose was distributed evenly throughout the entire pollen tube wall (Fig 8C.2-8C.4). Based on these results, the distribution of some of the cell wall materials in the cog3 and cog8 mutants was altered; however, the pattern of cellulose was not affected (S7 Fig).
In addition to cell wall polysaccharides, we examined the localization of LM2 epitopes and GFP-CSLD1 fusion proteins. LM2 is a monoclonal antibody which recognizes a subset of arabinogalactan proteins (AGPs) [49]. LM2 labeling appeared more strongly at the tip than at the shank in the wild-type (Fig 8D.1). In cog3 and cog8 pollen tubes, the tip polarity of LM2 labeling was lost (Fig 8D.2 and 8D.3). CSLD1 is a putative cellulose synthase localizing to the Golgi apparatus that could be transported to the plasma membrane of pollen tube apex (Fig 8E) [43]. GFP-CSLD1 in cog3 and cog8 mutants retained the Golgi localization, but the apical membrane labeling was much decreased. Notably, the GFP-CSLD1 labeled Golgi bodies appeared in the clear zone where they were normally excluded ( Fig 8E). Collectively, our data showed that in cog3 and cog8 mutants, polar deposition of cell wall materials and proteins is defective.

Putative COG complex organization in plant cells
The nearly identical phenotypes of the cog3 and cog8 mutants suggested that COG3 and COG8 may both be present in the COG complex and are critical for the function of the complex; therefore, we examined whether COG3 and COG8 interact directly and explored the structural organization of the COG complex. COG3 interacted with COG8 in yeast grown on a synthetic complete, quadruple drop out (QDO) medium lacking tryptophan, leucine, adenine, and histidine (Fig 9A). For bimolecular florescence complementation (BIFC) experiments, split-yellow fluorescent protein (YFP N or YFP C ) was fused to the C terminus of COG3 or COG8 respectively. COG3-2Y N + 2Y C and COG8-2Y C + 2Y N protein pairs served as negative controls ( Fig  9B and 9C). Various combinations of plasmids were cotransfected into Nicotiana benthamiana leaves, and protein expression levels were examined (S8 Fig). YFP signals that localized to the Golgi were detected only with COG3-2Y N and COG8-2Y C cotransfection (Fig 9D and 9E), but not in controls (Fig 9B and 9C). To examine the architecture of the COG complex, we cloned eight subunits of the Arabidopsis COG complex into the pGADT7 and pGBKT7 vectors for pair wise bidirectional interaction tests using a yeast-two-hybrid assay. In the course of our study, we identified another gene in addition to COG4 under the gene ID At4g01400, and based on our 3' RACE data, COG4 was given a new ID AT4G01395 by the Arabidopsis Information Resource (TAIR) (S9 Fig). Only those proteins that interacted bidirectionally were considered to reflect the COG complex structural assembly (Fig 9F). The analyses are summarized in Fig 9G. Interestingly, COG2, COG3, COG5 and COG8 were shown to have the abilities to form homodimers (Fig 9G, curved arrows). The interactions observed identified three possible COG complex subassemblies, COG2-4, COG1/8, and COG5-7, which appear to be well conserved in eukaryotic species.

Discussion
Rapid tip growth of elongating pollen tubes is sustained by a highly dynamic intracellular trafficking system that targets transport vesicles to the apical clear zone. Our results show that the COG complex subunits COG3 and COG8 play important roles in Golgi structure maintenance, COPI vesicle tethering, and efficient anterograde Golgi trafficking, which is crucial for pollen tube stability, rapid tip-focused pollen tube growth, and male fertility.
cog3 and cog8 pollen tube growth defects are caused by wrong deposition of cell wall materials We screened T-DNA insertion mutants of several COG subunits and found that mutations in COG3 and COG8 caused similar defects in pollen tube growth. Consistently, COG3 and COG8 are co-expressed in mature pollen and pollen tube (Fig 1), and COG3 and COG8 interact directly (Fig 9). cog3 and cog8 mutant pollen tubes were burst, short and slow-growing, or wavy (Figs 1, S2 and S3), which resulted in male sterility (Table 1). Wrong deposition of cell wall materials could be the cause of various pollen tube defects. Highly methylesterified pectins which were labeled by the JIM7 antibody were absent from the pollen tube tip, but were instead distributed along the pollen tube wall which decreased the rigidity of the cell wall and caused the pollen tubes to rupture at random sites (S2 Fig). In contrast, JIM5-labeled demethylesterified pectins and calloses appeared at the pollen tube tip where they were normally absent, which restricted tip growth and resulted in short pollen tubes that did not grow efficiently (Figs  8 and S7). Various combinations of these incorrectly-deposited cell wall materials led to pleiotropic phenotypes including short and wavy, short and burst, and branched pollen tubes similar to phenotypes found in mutants in cell wall synthesis or modification [42,45,50,51] or in vesicle trafficking pathways that deliver cell wall materials [11,16]. The broad range of cog3 and cog8 phenotypes further highlights the importance of the COG complex in efficient secretory transport of cell wall materials which in turn is required to maintain the rapid growth rates. Also, we need to bear in mind that COG3 and COG8 are required for plant sporophytic development (Fig 3), and the detailed mechanisms await to be studied in the future.

COG3 and COG8 are required for maintaining normal Golgi morphology
The Golgi apparatus consists of hundreds of individual stacks of flattened cisternae which are highly mobile in plant cells [31,52]. Information on the nature of the molecular interactions that maintain the Golgi morphology is limited.
COG3 and COG8 localize exclusively in the Golgi apparatus, preferably at its periphery (Fig  4), resemble that of γ-COP localization [25]. In the cog3 and cog8 pollen, Golgi morphology were disrupted as evidenced by confocal microscopy (Fig 5) and TEM (Fig 6) analyses. Loss of COG3 and COG8 in the mutants may disassemble the entire COG complex, whereby COPI vesicles containing transmembrane protein EMP12 could not be tethered with the Golgi apparatus, and distributed to the cytosol (Fig 7). Therefore, we speculated that in the cog3 and cog8 pollen, the shorter cisternae were resulted from a loss of their rims by losing peripheral COPI vesicles. The Golgi-localized ARF-GEF GNOM-like 1 (GNL1) protein localizes to and acts mainly at the Golgi stacks [52]. Golgi cisternae were shown to be wider in the gnl1 mutant probably due to the failure of COPI-coated vesicle formation [52,53]. Thus, COPI vesicles located at the Golgi periphery seem to contribute to the maintenance of Golgi morphology in plant cells.
In mammalian cells, the two characteristic features of the Golgi apparatus, flattened cisternae and cisternal stacking, is proposed to be governed by Golgi cisternal adhesion [54]. Major players of cisternal adhesion in mammalian cells, such as GRASP65/55, Glogin45 and GM130, are either not identified or not functioning in Golgi stacking in plant [54]. Instead, multiple weak inter cisternal adhesive processes (e.g., Rab-SNARE, tether-SNARE) could be underlying the plant Golgi morphology maintenance [54]. In the cog3 and cog8 pollen, the widths of individual Golgi cisternae increased about 40% in the mini-stacks (Fig 6K), indicative of disruption in cisternal flatness, and COG3 and COG8 as one of the players of intercisternal adhesion. In mammalian cells, the COG complex might both physically and functionally interacts with Rabs, SNAREs, SNARE-interacting proteins, coiled-coil tethers (golgin 84, p115, CASP, TMF etc.) [23]. If COG3 and COG8 proteins interact with these components in plant cells, cog3 and cog8 mutations might lead to the un-stability of intercisternal holding. However, these possibilities awaits further studies.

Aberrant Golgi membranes support anterograde transport at a lower rate and lower accuracy
The Golgi serves as a station for protein sorting and transport, receiving membrane and proteins from the ER and delivering them to the plasma membrane or other intracellular sites. In the meantime, Golgi-resident membrane proteins are thought to undergo continuous cycling between the ER and Golgi in plant cells [17,32]. In the cog3 and cog8 mutants, two Golgi-resident membrane proteins EMP12 and GAUT14 (Fig 5) were able to localize to the aberrant Golgi bodies, but not redistributed to the ER, indicating that protein exchange between the ER and the Golgi apparatus was not interrupted. What happened to the vesicle trafficking from the Golgi to the plasma membrane in the two mutant pollen tubes? Pectins and structural cell wall proteins are synthesized and modified within the Golgi apparatus after which they are packaged into secretory vesicles for targeting to sites of new cell wall deposition [51]. The disrupted Golgi bodies were shown to support anterograde delivery from the Golgi to the plasma membrane (Fig 8), but with much lower efficiency than the normal ones, as indicated by the presence of JIM7 fluorescence dots in the cytoplasm (Fig 8B), and by GFP-CSLD1 proteins less efficiently delivered to the apex pollen tube membrane (Fig 8E). Similar results were obtained in mammalian cells in which the anterograde transport of vesicular stomatitis virus G protein (VSVG) through the Golgi had minor defects and/or kinetic delays by partial COG3 depletion [21]. Furthermore, cell wall components examined in Fig 8 all exhibited wrong localization in the mutant pollen tubes. Therefore, COG complex is required for highly efficient and accurate anterograde transport.

Arabidopsis COG complex organization maybe similar to that in yeast and mammals
The direct interaction of COG3 and COG8 indicated by the yeast-two-hybrid ( Fig 9A) and BIFC (Fig 9B-9E) assays is not surprising given the similar defects in pollen tube growth caused by mutations in COG3 and COG8 (Fig 1). The cog3 and cog8 phenotypes are more severe than those of cog7 [37] probably because COG3 and COG8 have more interacting partners within the putative Arabidopsis COG complex (Fig 9F and 9G) and disruption of COG3 or COG8 likely results in disassembly of the entire COG complex. In mammalian cells, the eight subunits that make up the COG complex can be divided into three sub-assemblies: COG2-4, COG5-7, and COG1/8. The COG1/8 subassembly bridges the COG2-4 and COG5-7 assemblies, while, in the yeast COG complex, COG1p links the COG2-4p and COG5-7p lobes [20]. Our proposed model of the Arabidopsis COG complex organization largely resembles those of yeast and mammals, but with clear differences. For example, AtCOG8 appears to link the AtCOG2-4p and AtCOG5-7p lobes and certain pairs of subunit interactions are specific to plants. But the proposed model needs supports from further experiments, and the functional significance of the model awaits to be explored.
In this study, we show that the functions of COG complex are coupled to pollen tube growth, and provide important insights into the structure maintenance and the functioning of the Golgi apparatus in plant cells.

Phenotype characterization
To visualize nuclei, pollen grains were collected into a 4 0 ,6-diamidino-2-phenylindole (DAPI) staining solution (0.1 M sodium phosphate pH 7.0, 1 mM EDTA, 0.1% Triton X-100, and 0.5 mg/mL DAPI) and examined 15 min after staining. Pollen viability was examined using Alexander staining [55]. To observe embryo development, developing seeds were clarified overnight in Hoyer's solution [56] and observed with a Zeiss microscope equipped with a Differential Interference Contrast (DIC) system. In vitro pollen germination was performed as described by [51]. Germinating pollen grains were counted under a microscope after incubation at 23°C for 6 h.

Confocal laser scanning microscopy
Confocal images of fluorescent proteins or immuno-fluorescence signals in pollen grains or pollen tubes were collected using an LSM710 system (Zeiss, www.zeiss.com). Green fluorescent protein (GFP) signals were excited at 488 nm and emission was detected at 505-530 nm. RPA-DsRed [42] signals were excited at 543 nm and emission was detected at 585-615 nm. Time-lapse movies and images were obtained using an Ultra View spinning-disc confocal scanner unit (PerkinElmer) and processed using Velocity software (5.3V, Improvision, Inc.).

Immunofluorescence and cytochemical staining
Pollen tubes were adhered to poly-L-lysine-covered glass slides after germination in liquid germination medium for 4 h. Pollen tubes were fixed in 4% (w/v) polyformaldehyde in PIPES buffer (50 mM PIPES, 1 mM EGTA, 5 mM MgSO 4 , 0.5 mM CaCl 2 , 0.1% TritonX-100, pH 7) for 1 h. The fixed pollen tubes were washed three times with PBS (100 mM potassium phosphate, 138 mM NaCl, and 2.7 mM KCl, pH 7.3) and then incubated in blocking buffer (0.8% BSA, 0.1% gelatin, and 2 mM NaN 3 in PBS) for 30 min. After removing the blocking buffer, pollen tubes were treated with primary antibody for 1 h. The carbohydrate antibodies JIM5, JIM7, and LM2 were diluted 1:100 in blocking buffer. The pollen tubes were washed three times with PBS and incubated for 30 min in Alexa488 conjugated secondary antibody (1:100 dilution in blocking buffer). The pollen tubes were washed five times with PBS before analysis using a Zeiss LSM710 confocal microscope. For cytochemical staining, the pollen tubes were stained without fixation after germination in liquid germination medium for 4 h. Calcofluor white (0.01%, w/v), aniline blue (0.1%, w/v), and ruthenium red (0.01%, w/v) in germination medium were used to stain β-glucans (both cellulose and callose), callose, and pectins, respectively in pollen tubes.

Scanning and transmission electron microscopy
For scanning electron microscopy (SEM) observation, pollen grains were mounted on sample stubs. After dehydration in air for 30 min, pollen grains were coated with gold particles (EIKO IB-3). Pollen grains were then observed using a HITACHI S-3000N scanning electron microscope.
To visualize the structure of Golgi stacks by TEM, anthers of qrt1 -/-, cog3 -/+ qrt1 -/-, and cog8 -/+ qrt1 -/flowers were cut and high-pressure frozen and substituted with 2% OsO4 in 100% acetone and infiltrated with Epon resin as described [57]. TEM examination was performed using a Hitachi H-7650 transmission electron microscope with a charge-coupled device camera (Hitachi High-Technologies) operating at 80 kV.

Bimolecular fluorescence complementation (BIFC) assay
The coding sequences of the Arabidopsis COG3 and COG8 genes were subcloned into the p2YN-35S and p2YC-35S vectors, respectively, which containing yellow fluorescent protein (YFP) fragments and HA tag at C terminal, resulting in COG3-YFP N and COG8-YFP C constructs. For transient expression, Agrobacterium tumefaciens cells carrying the BIFC constructs were coinfiltrated with constructs carrying the Golgi marker ManI-mCherry or ST-mCherry into five-to six-week-old Nicotiana benthamiana tobacco leaves. Infiltrated leaves were observed after 24 h. The eYFP and mCherry fluorescent signals were monitored sequentially. The excitation and detection wavelengths for eYFP and mCherry were 514 and 587 nm for excitation, 527 and 610 nm for detection, respectively. For western blot, the infiltrated leaves were ground in liquid nitrogen and extracted with the lysis buffer, following the SDS-PAGE and immunoblotting. Anti-HA antibodies (Abcam ab9110) were used as the primary antibody at a dilution of 1:500.

Yeast two-hybrid assay
Yeast-two hybrid analysis was performed using the Match Maker GAL4 Two-Hybrid System according to the supplier's instructions (Clontech). The cDNAs were cloned into the pGBKT7 and pGADT7 vectors (Clontech). Pairs of pGBKT7 and pGADT7 vectors were cotransformed into the Saccharomyces cerevisiae strain AH109. Diploids were selected on synthetic drop-out (SD) medium lacking Trp and Leu (SD-Trp-Leu, DDO), while the selection of yeast cells expressing interacting proteins was made on SD medium lacking Ade, His, Trp, and Leu (SD-Ade-His-Trp-Leu, QDO), and subjected to a β-galactosidase assay. The experiments (yeast transformation, mating, and selection) were repeated several times and the representative result was presented.