Arabidopsis RHD3 mediates the generation of the tubular ER network and is required for Golgi distribution and motility in plant cells

The endoplasmic reticulum (ER) appears as an extended network of interconnected tubules and cisternae stretching throughout the cytoplasm. The shape of the ER often undergoes drastic changes, but the mechanisms underlying the ER rearrangements are not fully elucidated (Federovitch et al., 2005; English et al., 2009). Recent data have indicated that continuous fusion and fission reactions of ER tubules are necessary for the dynamic restructuring of the ER network (Dreier and Rapoport, 2000; Voeltz et al., 2006; Shibata et al., 2009). It has also been generally accepted that the cytoskeleton plays an important role, possibly as a facilitator in the generation and distribution of the ER network in living cells (Runions et al., 2006; Vedrenne and Hauri, 2006; Shibata et al., 2009).

Two families of ER membrane proteins, the reticulons and DP1/Yop1, have been shown to be structural components that shape the ER tubules in mammalian and yeast cells (Voeltz et al., 2006) as well as in proteoliposomes (Hu et al., 2008). The reticulons and DP1/Yop1 proteins have a wedge-like transmembrane topology and they can oligomerize in the tubular ER, serving as arc-like scaffolds around ER tubules (Shibata et al., 2008). It is assumed that they can deform the lipid bilayer into curved tubules through their wedge-like hydrophobic insertion and/or scaffolding mechanisms (Shibata et al., 2009; Sparkes et al., 2010). Recent work also indicated that a class of ER-membrane-bound, dynamin-like GTP binding proteins, which includes the atlastins in mammals and Sey1p in yeast (Saccharomyces cerevisiae), plays an important role in the generation of the interconnected ER tubules (Hu et al., 2009; Orso et al., 2009). Atlastins and Sey1p do not share sequence homology but the membrane topology of the proteins is similar, with four conserved GTP motifs at the N-terminus, and two transmembrane domains, spaced by a few amino acids, located in the lumen of the ER at the C-terminus (Hu et al., 2009). It has been hypothesized that the atlastins and Sey1p might be required in the generation of local ER membrane curvatures that are necessary for either fission or fusion of ER tubules (Hu et al., 2009). In humans there are three atlastin isoforms (Rismanchi et al., 2008). Mutations in atlastin-1 frequently cause hereditary spastic paraplegia (HSP), a disease primarily affecting the development of the long axons of corticospinal neurons (Rismanchi et al., 2008).

Similar to animal and yeast cells, in growing cells of most plant tissues, the cortical ER is organized into a tubular reticulate network interconnected by three-way junctions (Ridge et al., 1999; Zheng et al., 2004). The ER of plant cells undergoes drastic rearrangements in response to both developmental cues and outside influences (Ridge et al., 1999; Hawes et al., 2001; Runions et al., 2006). In mammalian cells, ER tubules move mainly along microtubules (Vedrenne and Hauri, 2006); in plants, although the motility and orientation of the ER can be microtubule dependent (Mathur et al., 2003; Foissner et al., 2009), actin–myosin-based motility is believed to be the primary means of ER movement (Boevink et al., 1998; Sparkes et al., 2009; Ueda et al., 2010). Compared with other eukaryotic cells, plant cells have developed a unique ER–Golgi interface (Moreau et al., 2007; Staehelin and Kang, 2008). The plant Golgi apparatus is made of functionally independent Golgi stacks, each consisting of several polarized cisternae (Boevink et al., 1998; Nebenfuhr et al., 1999). The Golgi stacks in plant cells are closely associated with the ER tubules and travel along the actin tracks throughout the cytoplasm (Boevink et al., 1998; Nebenfuhr et al., 1999). However, the function and molecular mechanism of Golgi motility on the actin tracks remain unknown. From observations of Golgi stacks alternating motile and stationary phases, a ‘stop-and-go’ model was proposed to correlate Golgi movement with ER-to-Golgi transport (Nebenfuhr et al., 1999; Staehelin and Kang, 2008). However, although disruption of the actin track could arrest Golgi movement, both ER-to-Golgi and Golgi-to-ER transport appear not to be dependent on the actin filaments (Brandizzi et al., 2002b; Saint-Jore et al., 2002). In plants, ER exit sites are found closely associated with and tracked with Golgi stacks (daSilva et al., 2004; Stefano et al., 2006). It was therefore suggested that ER exit sites and Golgi stacks might constitute entire mobile functional units and that the stack movement might not be required to reach ER export sites for ER-to-Golgi transport to occur (daSilva et al., 2004; Stefano et al., 2006).

Arabidopsis thaliana ROOT HAIR DEFECTIVE 3 (RHD3) is a large GTP-binding protein initially isolated in a genetic screen for mutants defective in root hair development (Wang et al., 1997). In rhd3 mutants, the root hairs are short and wavy (Wang et al., 1997), a defect reminiscent of axon growth in HSP. RHD3 is ubiquitously expressed in plants, so the cell growth defect is prominent but not restricted to root hairs in rhd3 mutants (Wang et al., 1997). Arabidopsis RHD3 has been classified as a member of atlastin GTPases (Hu et al., 2009). In Arabidopsis, there are three RHD3 isoforms (Hu et al., 2003) but the cellular function of each and their relationship with other members of atlastin GTPases remain to be determined. A study on the steady-state distribution of the soluble ER marker, GFP-HDEL, and of the bulk flow marker, secGFP in rhd3-1 (a line with an Ala575 to Val mutation in RHD3) suggested that RHD3 is required for both ER morphology and perhaps ER-to-Golgi protein transport (Zheng et al., 2004). Here, we provide evidence that RHD3 is primarily involved in the generation of the interconnected tubular ER network in plant cells. The expression of mutant forms of RHD3 alters the cellular distribution and motility of Golgi stacks but does not affect ER membrane export, indicating a specific role of RHD3 in the morphological integrity of the ER rather than in export functions. Results of complementation experiments suggest that RHD3 proteins might be functionally redundant in planta, but that they do not complement a yeast sey1Δ yop1Δ mutant (Hu et al., 2009), and that Sey1p cannot complement an rhd3 loss-of-function mutant. These results underscore that the detailed mechanisms underlying ER tubular integrity might not well be conserved among eukaryotes.

As the first step to define the cellular role of Arabidopsis RHD3, we tagged RHD3 with GFP at its N-terminus to determine its subcellular location. GFP–RHD3 is functional because it fully complements the Arabidopsis rhd3-1 mutant throughout the development of root, root hairs and adult plants (Fig. 1A–D; see also Fig. 3A and Fig. 8A for rhd3-1 phenotype). Confocal imaging of rhd3-1::GFP–RHD3 root epidermal cells revealed that the fusion protein marked a fine meshwork reminiscent of the cortical ER (Fig. 1E; supplementary material Movies 1, 2). In rhd3-1 cells, the fine polygonal tubular ER appeared severely compromised, as shown in rhd3-1 root epidermal cells stained with Rhodamine B hexyl ester, a dye used to stain the ER (Fig. 1F) (Zheng et al., 2005a). Staining of rhd3-1::GFP–RHD3 cells with Rhodamine B hexyl ester suggested that the cortical network marked by GFP–RHD3 is indeed ER (Fig. 1G–I).

We frequently observed that GFP–RHD3 was also concentrated in punctae along the ER tubules (Fig. 1E, arrows; supplementary material Movie 2), often at the three-way junctions of tubules (Fig. 1E, inset). This distribution of the punctae was very dynamic (supplementary material Movies 1, 2). We also noted that, similar to Sey1p–GFP in yeast cells, the punctate labeling of RHD3 appeared weak in some cells with low levels of fluorescence (Fig. 1G), probably due to the low abundance of RHD3. Simultaneous visualization of GFP–RHD3 and Golgi stacks labeled by either the trans-Golgi marker ST–YFP or the ER and cis-Golgi marker ERD2–YFP (Boevink et al., 1998; Brandizzi et al., 2002a) indicated that the RHD3 punctae do not correspond to Golgi stacks (Fig. 2A,B, arrows in the merged image). The punctae appeared much more enriched in GFP–RHD3 than in ER soluble [YFP–HDEL (Irons et al., 2003)] or membrane [P24σ1d–YFP (Langhans et al., 2008)] protein markers, or could even be devoid of the latter two proteins (Fig. 2C,D, arrows in the merged images). The expression of all GFP and YFP ER and Golgi markers are known to not affect plant cell development (Boevink et al., 1998; Brandizzi et al., 2002b; Zheng et al., 2004; Chen et al., 2009). We have not noticed any impact of the expression of GFP–RHD3 and P24δ1d–YFP constructs on cell and plant development (supplementary material Fig. S1).

Atlastins and Sey1p physically interact with the reticulons and DP1/Yop1 in generating tubular ER network (Hu et al., 2009). In Arabidopsis, a Yop1 homolog has been identified as HVA22d, which has been suggested to function as a negative regulator of autophagy (Chen et al., 2009). We found that HVA22d–YFP was also localized to the tubular ER network and was concentrated in punctae along the ER tubules (Fig. 2E, arrows in the merged image). Interestingly, GFP–RHD3 and YFP–HVA22d were colocalized to the ER tubules and at the punctae (Fig. 2E; supplementary material Movie 3). Together with the evidence that the COPII coat proteins, which are responsible for cargo export, are located at the Golgi-associated ER exit sites (daSilva et al., 2004; Stefano et al., 2006), the data presented here suggest that RHD3 acts at the level of the ER, and that the RHD3 punctae are separate entities from the COPII-mediated Golgi-associated ER export sites, but might contain the protein machinery that is important for the ER morphology.

RHD3 is classified as a member of the class of dynamin-like atlastin GTPases (Hu et al., 2009). The GTPases of this class contain the signature motifs (G1, G2 and G3) characteristic of dynamin GTPases but they also possess a distinct G4 motif comprising three hydrophobic residues preceding an Arg-Asp (RD) sequence (Hu et al., 2009). Transgenic wild-type Col-0 plants expressing either YFP–RHD3(S51N) (a single S51N substitution in G1) or YFP–RHD3(T75A) (a single T75A substitution in G2) phenocopied rhd3-1 (Fig. 3A), whereas expression of YFP–RHD3 had no effect on plant development (data not shown). We noted that both mutant forms of RHD3 were predominantly localized to some condensed tubules in root epidermal cells that did not appear well interconnected (Fig. 3B,C, inset, arrows). YFP–RHD3 in wild type Col-0 root epidermal cells marked a polygonal tubular network and punctae along the tubules (Fig. 3D, arrows). Interestingly, RHD3(T75A) tended to be concentrated at punctae along the tubules, specially at the tubular junctions (Fig. 3C, arrowheads; supplementary material Movie 4), whereas RHD3(S51N) mainly marked condensed tubules (Fig. 3B, arrows). Rhodamine B hexyl ester staining suggested that, similar to rhd3-1 plants (Fig. 1F) (Zheng et al., 2004), the tubular ER network in these transgenic plants was altered, in that it lacked the fine meshwork of interconnected ER tubules (Fig. 3E,F). We also found that, similar to transgenic Arabidopsis expressing RHD3(S51N) and RHD3(T75A), transient expression of both RHD3 mutants in tobacco (Nicotiana benthamiana) leaf epidermal cells led to unbranched ER tubules (supplementary material Fig. S2A,B; supplementary material Movies 5, 6). Together, these data imply that RHD3 is involved in shaping the fine tubular ER and this function requires amino acid residues in the GTP motifs that are conserved among dynamin GTPases.

In order to test whether morphological changes of the ER could be correlated with a perturbation of ER-to-Golgi protein trafficking, we tested the effects of RHD3(S51N) in a transient expression system. In Agrobacterium-mediated transient protein expression in tobacco leaf epidermal cells, expression of a protein starts roughly 24 hours post-infiltration and gradually reaches a peak 67 hours post-infiltration (Zheng et al., 2005b). Leaves were infiltrated with the Agrobacterium strains transformed with either the ER lumenal marker GFP-HDEL or the Golgi marker ST–GFP (Batoko et al., 2000), and 24 hours later, the same sectors were re-infiltrated with the Agrobacterium strain containing RHD3(S51N), as well as the following controls: RHD3, RAB-D2a and RAB-D2a(N121I) (Zheng et al., 2005b). Infiltration of the leaf areas with the ER and Golgi markers followed by a re-infiltration with the mutant protein ensured that the markers are expressed before the mutant protein has any effect on their traffic. We expected that if ST–GFP export to the Golgi was inhibited by the RHD3 mutant, then the marker would be partially distributed to the ER (Zheng et al., 2005b). We found a loss of fine tubular ER network roughly 38 hours after infiltration with RHD3(S51N) (Fig. 4A); however, targeting of ST–GFP to Golgi was largely unaffected, even after 67 hours (Fig. 4C). By contrast, an inhibition of ST–GFP trafficking by RAB-D2a(N121I) was clearly visible as early as 38 hours post-infiltration (Fig. 4D), as expected (Zheng et al., 2005b); in the presence of this mutant, however, no morphological changes in the ER could be detected, even after 67 hours (Fig. 4B). Expression of wild-type RHD3 and RAB-D2a had no effect on either ER and ER-to-Golgi trafficking (Fig. 4E–H). Together, these results indicate RHD3 is primarily involved in the generation of interconnected ER tubules and that alteration of the ER meshwork by the RHD3 mutant does not affect ER-to-Golgi protein trafficking.

Because it has been proposed that the ER and the Golgi are attached in highly vacuolated plant cells (Sparkes et al., 2009), we questioned whether RHD3-induced alteration of the ER morphology could also affect Golgi distribution and motility. To test this we compared the Golgi distribution in cells expressing either RHD3 or the RHD3(S51N) mutant. In cells expressing wild-type RHD3, we found that Golgi stacks were dispersed along the tubular ER network throughout the cytoplasm (Fig. 4G; Fig. 5E–G; supplementary material Fig. S3B) (Boevink et al., 1998; Nebenfuhr et al., 1999) with occasional aggregation of a few stacks. In the presence of RHD3(S51N), however, 8–15 Golgi stacks were aggregated around condensed ER tubules and especially at the junctions of the tubules (Fig. 4C; Fig. 5A–C, arrows; supplementary material Fig. S3A). We noted that these Golgi stacks often underwent slow wiggling-type motions around the condensed ER tubules for a prolonged time (Fig. 5B; supplementary material Fig. S3A, arrows in different colors). We tracked and monitored the motility of randomly selected Golgi stacks in eight different cells expressing wild-type RHD3 and RHD3(S51N). In wild-type cells, Golgi stacks often traveled with velocities between 0.5 and 4 μm/second in the cytoplasm (Fig. 5H), consistent with previous reports (Boevink et al., 1998; Nebenfuhr et al., 1999). Although it is possible to quantify the instantaneous velocity of individual Golgi stacks at a given instance of time, because of the long-range movement and the variability of the stack velocity, we found it was difficult to average the velocity of Golgi stacks over a long period of time. However, we noted that the majority of Golgi stacks (89.1%, n=55; Fig. 5H) in the presence of wild-type RHD3 moved quickly (>1 μm/second), at least at one instance of time over a period of 50 seconds. In the presence of RHD3(S51N), the number of Golgi stacks that moved quickly (>1 μm/second) was greatly reduced (32.8%, n=64). Most Golgi stacks (67.2%, n=64) only moved slowly (<1 μm/second) along the ER (Fig. 5D).

The Golgi is a sorting station for proteins destined to post-Golgi compartments. Because expression of dominant-negative versions of RHD3 alters cellular distribution and motility of Golgi stacks but not ER-to-Golgi trafficking, we asked whether expression of dominant-negative forms of RHD3 could inhibit protein trafficking to the cell surface. In the presence of RAB-D2a(WT), the apoplast markers secGFP (Zheng et al., 2004) and AtCTL1–GFP [a cell-wall-localized chitinase-like1 involved in biosynthesis of cell walls (Zhong et al., 2002)] were secreted, producing dim GFP fluorescence (Fig. 5A,D) owing to sub-optimal pH in the apoplast for GFP fluorescence and protein truncation (Zheng et al., 2004; Zheng et al., 2005b). In the presence of dominant-negative RAB-D2a(NI), as expected (Zheng et al., 2005b), secretion of the these proteins was inhibited, as shown by an accumulation of the GFP signal in the ER (compare Fig. 6B with A, 6E with D; supplementary material Fig. S4B with A, S4E with D, quantification in G). By contrast, transport of these proteins to the cell wall was not impaired in the presence of the mutant RHD3 as no increase in intracellular GFP intensity was observed (compare Fig. 6C with A, 6F with D; supplementary material Fig. S4C with A, S4F with D, quantification in G). The presence of RHD3(S51N) is indicated in Fig. 6C,F (YFP channel). This was further confirmed by a western blot analysis of tobacco leaves expressing secGFP in the presence of either wild-type and mutated RAB-D2a or wild-type and mutated RHD3 (Fig. 6G). SecGFP is known to be partially cleaved in the apoplast (Zheng et al., 2004). On western blots with anti-GFP serum this gives two bands with the top band corresponding to full-length intracellular GFP, and a stronger lower band corresponding to the cleaved apoplastic product (Zheng et al., 2004). We found that the intensity ratio of the two bands was comparable in extracts of leaves expressing either wild-type RAB-D2a, wild-type or mutant RHD3 (Fig. 6G). In the presence of mutant RAB-D2a(NI), the intensity of the signal of the lower band was much reduced compared with the upper band, indicating that GFP is not secreted efficiently to the apoplast, where it would be cleaved. These data indicate that secretion of secGFP is not compromised in the presence of the RHD3 forms. Similarly, targeting of AHA2–GFP, a plasma membrane localized H⁺-ATPase (Kim et al., 2001), to the plasma membrane was also not prevented by RHD3(S51N) (compare Fig. 6H with I, arrows). These data show that the aberrant functioning of RHD3 affects specifically ER morphology and Golgi distribution. This, in turn, does not compromise ER-to-Golgi protein transport or Golgi-to-plasma-membrane protein traffic.

In animal cells, atlastins homo-oligomerize and GTP binding is crucial for this self-interaction (Rismanchi et al., 2008; Orso et al., 2009). Using a mating-based split ubiquitin system in yeast, we found that RHD3 homotypically interacts with another RHD3 protein, but not with controls that included the plasma membrane protein KAT1 (Obrdlik et al., 2004) and the ER membrane protein P24σ1d (Langhans et al., 2008) (Fig. 7A). The homotypic interaction of RHD3 was further confirmed with bimolecular YFP fluorescence complementation in plant cells. When the N-terminal half of YFP (YFPn) fused with RHD3 (YFPn–RHD3) was coexpressed with the C-terminal half YFP (YFPc) fused RHD3 (YFPc–RHD3), YFP fluorescence was re-established (Fig. 7B, arrows; 32% of infiltrated cells, n=247). No YFP fluorescence was discerned when YFPn–RHD3 was coexpressed with YFPc fused to P24σ1d (P24σ1d–YFPc) (Fig. 6C; 0% of infiltrated cells, n=234). Interestingly, coexpression of the two split-YFP fusions to RHD3 used for bimolecular fluorescence complementation with GFP–RHD3 revealed that the self interaction of RHD3 occurred largely at the RHD3 punctae (Fig. 7D,E, arrows).

In Arabidopsis, in addition to RHD3, there are two RHD3-like GTP binding proteins encoded by the genetic loci At1g72960 and At5g45160 (Hu et al., 2003). In this report, we refer to them as RHD3-like1 (RL1) and RHD3-like2 (RL2), respectively. RL2 is ubiquitously expressed at very low levels throughout plant tissues, RL1 is pollen specific (Hu et al., 2003; Schmid et al., 2005). Knockout mutants of RL1 and RL2 produce no obvious developmental defect (supplementary material Fig. S5). To understand the functional relationship within the RHD3 GTPases in plants, we expressed GFP fused RL2 in rhd3-1 under the control of the constitutive promoter CaMV 35S. RL1 was not analyzed because of expression specificity at a tissue level. We found that GFP–RL2 fully complemented rhd3-1 (Fig. 8A). Confocal microscopy of GFP–RL2 expressed in rhd3-1 cells revealed that, similar to RHD3, RL2 was also localized to the ER tubules and often concentrated on punctae along the ER tubules (Fig. 8B). Simultaneous visualization of GFP–RL2 and YFP–RHD3 indicated that RL2 was colocalized with RHD3 on the ER tubules and the RHD3 punctae (Fig. 8B, arrows). Transient expression of YFP–RL2(S54N), a mutated version of RHD3L2 with a S54N substitution in the G1 motif, revealed that RL2(S54N) was also predominantly localized to some condensed tubules (Fig. 8C, arrowheads). The morphology of the tubular ER network indicated by GFP–HDEL was also altered in the presence of this mutant RL2 (Fig. 8C), indicating that the S54N substitution in RL2 also has a dominant-negative effect on the generation of the tubular ER network. Coexpression of wild-type GFP–RHD3 rescued this negative effect of RL2(S54N) (Fig. 8D). Furthermore, RL2 and RHD3 were found to form homotypic and heterotypic interactions when expressed in yeast cells (Fig. 8E). These data indicate that members of the RHD3 family, at least in conditions of overexpression, can replace the function of other members in the maintenance of the ER tubular morphology, suggesting that they might act on similar targets in the cells.

In the family of dynamin-like atlastin GTPases, Sey1p shares sequence homology with RHD3 but not with the atlastins (Hu et al., 2009). It was proposed that Sey1p and RHD3 might work in protozoans and plants, whereas the atlastins might function in metazoan (Hu et al., 2009). Having established the primary role for RHD3 in the generation of the tubular ER network in plants, we expressed Sey1p and RHD3 in plants and yeasts, reciprocally. We found that unlike RHD3 (Fig. 1A–D), Sey1p was unable to rescue the developmental defect of the rhd3-1 mutant (Fig. 8A; supplementary material Fig. S6). When YFP–Sey1 was expressed in tobacco leaf epidermal cells, it was targeted to the ER (supplementary material Fig. S6B). At the same time, when Arabidopsis RHD3 was expressed in sey1Δ yop1Δ yeast cells with Sec63p fused with GFP as an ER marker (Hu et al., 2009), RHD3 was unable to rescue the altered ER morphology in the sey1Δ yop1Δ cells (compare Fig. 9C with A), whereas Sey1p did (compare Fig. 9B with A). When GFP–RHD3 was expressed in yeast cells, it was localized to the ER (supplementary material Fig. S6C–E). These results indicate that although Sey1p and RHD3 have a similar function, in the formation of the tubular ER network in the corresponding organisms, unique functional features of Sey1p and RHD3 probably arose during evolution to suit different cell systems.

RHD3 is a large GTP-binding protein originally identified in a screen for root hair defective mutants (Wang et al., 1997) whose primary cellular function remains elusive (Hu et al., 2003; Zheng et al., 2004). We show here that RHD3 is primarily involved in the generation of the tubular ER network rather than in secretory traffic. RHD3 resides in ER tubules; however, it also concentrates as punctae on the ER tubules (we termed them RHD3 punctae). RHD3 is also colocalized with HVA22d, an Arabidopsis homolog of DP1/Yop1p that is involved in shaping ER tubules (Voeltz et al., 2006) at the RHD3 punctae. This result suggests that the RHD3 punctae define subdomains that control the tubular morphology of the ER.

A recent structural study on recombinant atlastin-1 has indicated that atlastin-1 undergoes a GTP-dependent oligomerization in solution (Byrnes and Sondermann, 2011). Here we show that RHD3 is capable of homotypic interaction, which appears to occur largely within the RHD3 punctae on the ER tubules. Generally in the large dynamin-like GTPases, the amino acid stretch GxxxxGKS (G1) is involved in coordinating phosphate binding and the tyrosine (T) in the G2 motif is required for GTP hydrolysis (Marks et al., 2001). The expression of both RHD3(S51N), a mutant form of RHD3 with a single amino acid change in the G1 motif, and RHD3(T75A), with a single amino acid change in the G2 motif, causes the ER to be less branched. Interestingly, our data suggest that the mode of action of the two RHD3 mutants in cells is different. We found that RHD3(S51N) is largely associated with less branched ER tubules, whereas RHD3(T75A) is seen frequently on large punctae in the junction of the less branched ER tubules. It is known that, in the presence of GTP, dynamins are capable of deforming membranes by transiently polymerizing on them. Upon GTP hydrolysis, triggered by polymerization, the dynamin scaffold disassembles, resulting in a membrane release for spontaneous membrane tubulation and fission (Bashkirov et al., 2008). Although it remains to be demonstrated how RHD3(S51N), as well as RHD3(T75A), could affect homotypic interaction of RHD3 in plant cells, it is tempting to speculate that RHD3(S51N) might have a reduced ability in homotypic interaction and therefore a reduced ability to form RHD3 punctae, but RHD3(T75A) might have an enhanced ability in homotypic interaction and therefore form large RHD3 punctae in the junction of the less branched ER tubules.

It has been proposed that the GTP-dependent conformational changes of atlastin at discrete points along the tubules is crucial in the formation of the tubular ER network with three-way junctions, by either fission of the ER tubules or fusion of two different ER tubules (Hu et al., 2009). Atlastin in Drosophila has been shown to mediate the fusion of different ER tubules (Orso et al., 2009). At the moment, we do not know whether RHD3 is involved in fission or fusion of ER tubules in the generation of tubule ER network. Given the structural similarity between RHD3 and atlastin, it is possible that RHD3 also mediates the fusion of different ER tubules.

The atlastins and Sey1p interact with the previously identified ER tubule-shaping proteins of the reticulon and DP1/Yop1 families (Hu et al., 2009). Here we revealed that RHD3 is colocalized with HVA22d at the RHD3 punctae. HVA22d has been implicated as a negative regulator of autophagy (Chen et al., 2009) although its homologue, DP1/Yop1, is known to play a role in ER tubule shaping (Voeltz et al., 2006). It is possible that autophagy and ER tubular shape maintenance are linked processes (Chen et al., 2009). A role for Arabidopsis reticulons in remodeling of ER tubules has also been proposed (Tolley et al., 2008; Sparkes et al., 2010). It will be interesting to test whether RHD3 could interact with reticulons and HVA22 proteins to coordinate the generation of tubular ER network in plant cells.

We revealed in this study that Sey1p, a yeast functional homologue of atlastin, and RHD3 are not interchangeable in rescuing the respective loss-of-function mutants. Although it is possible that RHD3 does not fold properly in yeast and that Sey1 might not fold properly in Arabidopsis, the evidence that fluorescent protein fusions of RHD3 and Sey1 are targeted to the ER in the alternative system suggests that this possibility is unlikely. Therefore, it seems that although the function of the atlastin GTPase members in the generation of the tubular ER network is conserved among eukaryotes (Hu et al., 2009) (this study), their molecular mechanisms might have diversified in the different organisms. A possible reason for the diversified mechanisms could be the different relationship between the ER and the cytoskeleton in animal, plant and yeast cells (Bola and Allan, 2009).

In mammalian cells, ER tubules are moved mainly along microtubules (Vedrenne and Hauri, 2006). Atlastin-1 interacts with spastin, a katanin-like ATPase microtubule severing factor to disassemble microtubules (Lee et al., 2009) in order to remodel the ER network. In plants, actin–myosin-based motility is the primary means of moving the ER (Boevink et al., 1998; Sparkes et al., 2009; Ueda et al., 2010), although the motility and orientation of the ER can be microtubule dependent (Mathur et al., 2003; Foissner et al., 2009). In rhd3-1 cells, the organization of cortical actin filaments is affected (Hu et al., 2003) and the transition of cortical microtubules in response to low doses of propyzamide, a β-tubulin-binding pronamide that affect microtubule dynamics, is also impaired (Yuen et al., 2005). It will be interesting to test whether and how RHD3 could physically and genetically interact with actin and microtubule elements to control the stability of the cytoskeleton in the generation of tubular ER network and cell growth in plants. One obvious candidate to be tested is the Arabidopsis homologue of katanin. Consistently, Arabidopsis plants with mutations in the katanin gene are dwarf (Burk and Ye, 2002).

In Arabidopsis, there are three isoforms of RHD3 (Hu et al., 2003). RL2 is ubiquitously expressed but at very low level, whereas RL1 is pollen specific (Hu et al., 2003; Schmid et al., 2005). In our functional analysis of RL1 and RL2, we revealed no obvious developmental defect in knockout mutants of RL1 and RL2. However, rhd3-1 can also be rescued by the overexpression of RL2. Expression of an RL2 mutant [RL2(S54N)] leads to ER with less branched tubules. Coexpression of wild-type RHD3 with RL2(S54N) rescues the defect. Furthermore, RHD3 and RL2 can form a heterotypic interaction when expressed in yeast cells. Together, these results indicate that there is functional overlap between members of the RHD3 protein class in the generation of the tubular ER network. It seems that RL2 is functional but at a low level copy number so that the loss of RL2 in the knockout line is compensated by RHD3. rhd3-4, a T-DNA insertional knockout mutant results in slightly longer root hairs than rhd3-1 (Wang et al., 1997), an allele with an A575V mutation in RHD3. It is possible that the loss of RHD3 in rhd3-4 is also partially compensated by low level RL2. In humans, three atlastins are differentially localized in ER domains, and atlastin-1, but not atlastin-2 or -3, interacts with spastin (Rismanchi et al., 2008). It remains to be determined whether RHD3 proteins function differently in the generation of tubular ER network in different regions of the ER.

Compared with other eukaryotic cells, a unique feature of the plant ER–Golgi interface is that individual plant Golgi stacks are closely associated with ER tubules and are motile along the actin tracks with variable velocities (Boevink et al., 1998; Nebenfuhr et al., 1999). In the presence of dominant-negative RHD3 mutants, Golgi stacks tend to aggregate; many of them undergo slow wiggling motion along the condensed ER tubules for a prolonged time. This suggests that, in addition to the actin tracks, the tubular ER network could also be a molecular highway for fast linear movements of the Golgi stacks. Interestingly, although the distribution and motility of Golgi stacks is altered in the presence of mutant RHD3, ER-to-Golgi transport is not prevented. This result is consistent with the observation that, when actin filaments are disrupted the traffic of cargo from the ER to Golgi stacks is not affected (Brandizzi et al., 2002b). We think this can be at least partially explained by the fact that in plants Golgi stacks are closely associated with ER exit sites where they act as a functional unit (daSilva et al., 2004; Stefano et al., 2006).

Here we show that in the presence of mutant RHD3, ER-to-Golgi protein transport is not affected, and we also noted that general secretion of cell wall proteins and targeting of plasma membrane proteins are not prevented. This is consistent with the findings in mammalian cells that in atlastin-depleted HeLa cells the plasma membrane targeting of VSVG, a vesicular stomatitis virus glycoprotein, is not impaired (Rismanchi et al., 2008). It seems, therefore, that the minor transport defect detected in some cells of rhd3-1 (Zheng et al., 2004) might be a secondary effect of the ER disorganization, perhaps resulting from stress during the period of plant growth. In particular, we cannot exclude that the fluorescent signal found in some rhd3-1 cells (Zheng et al., 2004) is due to accumulation of autofluorescent molecules, as might happen in conditions of stress (Chalker-Scott et al., 1999). It is also possible that reduced vacuolar expansion in rhd3-1 mutants (Galway et al., 1997) is linked to altered pH of the lumen of secretory organelles of some rhd3-1 cells. The secGFP reporter used by Zheng et al. is pH sensitive (Zheng et al., 2004). Therefore, accumulation of GFP signal in some rhd3-1 cells might also be linked to pH changes during growth.

Our analyses indicate that the Arabidopsis rhd3 mutants could be a useful tool for investigating the regulation and functional roles of ER remodeling in directional cell elongation. Approximately 50% the known dominant forms of HSP have been linked to mutations in atlastin-1, REEP1(DP1-like) and spastin, three proteins that interact in the generation of tubular ER (Park and Blackstone, 2010). However, the actual functional role of ER restructuring in the development of the long axons of corticospinal neurons has not been established. Searches for genes that suppress rhd3-1 could provide valuable therapeutic insights into HSP, and identification of rhd3-1 enhancers could shed light on how the generation of the tubular ER network is controlled and how it is involved in cell elongation. For example, synergistic genetic interaction of rhd3 with scn1-1 (RhoGDI), tip1-2 (S-acyl-transferase) and shv2-1 (a plasma-membrane-localized COBRA-like protein) (Parker et al., 2000) certainly warrants further investigation.

RHD3, p24σ1d, HVA22d, AtCTL1 and RHD3L2 were amplified by RT-PCR using a Superscript III first-strand synthesis system kit (Invitrogen) with RNAs from wild-type Col-0 Arabidopsis. RHD3-1(A575V) was amplified from rhd3-1. RHD3 was first cloned in pBluescript-II(KS) as an XbaI–SalI–BamHI fragment and sequenced. The sequenced RHD3 was then subcloned as a SalI–BamHI fragment into pVKH16–GFPN (Zheng et al., 2005b) to generate pVKH18–GFP–RHD3. RHD3 and RHD3L2 was also cloned into the entry vector pDONR222 and sequenced. p24σ1d, HVA22d, AtCTL1 were cloned into the entry vector pCR8/GW/TOPO and sequenced. The corresponding genes were then subcloned into destination vector pEarleyGate104 [Arabidopsis Biological Resource Center (ABRC) stock DB3-686] to generate YFP–RHD3; into pEarleyGate101 (ABRC stock DB3-683) to produce P24σ1d–YFP and HVA22d–YFP; into pEarleyGate103 (ABRC stock DB3-685) to generate AtCTL1–GFP; and into pMDC43 (Curtis and Grossniklaus, 2003) to create GFP–RHD3L2. The gene encoding Sey1p from Saccharomyces cerevisiae was amplified from pRS315 (Hu et al., 2009), sequenced and subcloned into the binary vector pFGC5941. YFP–Sey1 was made in pEarleyGate104. All ER, Golgi and transport markers were described by Zheng et al. (Zheng et al., 2005b).

Site-directed mutagenesis of RHD3 and RHD3L2 was performed in pDONR222 using Quick Change Lighting Site-Directed Mutagenesis Kit (Stratagene). RHD3(S51N) was generated using primers 5′-CCACAAAGTAGTGGGAAGAATACGCTCTTGAATCATTTG-3′ and 5′-CAAATGATTCAAGAGCGTATTCTTCCCACTACTTTGTGG-3′. RHD3(T75A) was created using primers 5′-GAGGAAGGTCTCAGACGGCTAAGGGAATTTGGATT-3′ and 5′-AATCCAATTCCCTTAGCCGTCTGAGACCTTCCTC-3′. RHD3L2(S54N) was produced using primers 5′-GGTCCTCAAAGTAGTGGGAAGAATACACTTTTGAATCATCTTTTT-3′ and 5′-AAAAAGATGATTCAAAAGTGTATTCTTCCCACTACTTTGAGGACC-3′. The mutated genes were sequenced and then cloned into destination vector pEarleyGate104 to produce YFP–RHD3(S51N), YFP–RHD3(T75A) and YFP–RHD3L2(S54N).

rhd3-1 expressing GFP–RHD3, YFP–RHD3, GFP–RHD3L2 and Sey1p, wild-type Col-0 expressing YFP–RHD3(S51N) and YFP–RHD3(T75A) were generated by Agrobacterium-mediated transformation of corresponding plasmids into rhd3-1 and wild-type Col-0 plants, respectively (Clough and Bent, 1998). Transformed Arabidopsis seeds were selected on AT (Somerville and Ogren, 1982) or MS (Murashige and Skoog, 1962) growth medium supplied with either kanamycin (50 μg/l; Wisent, Montreal, Canada) or hygromycin (20 μg/l; Wisent) or BASTA (20 μg/ml; Calbiochem). To confirm the expression of Sey1 in transgenic plants, RNA was isolated from 2-week-old seedlings (RNeasy Plant Mini Kit; Qiagen). The first-strand cDNA was then synthesized using the SuperScript III first-strand synthesis system (Invitrogen). PCR was performed using the High-fidelity DNA Polymerase (Finnzymes, Espoo, Finland) for 35 cycles. For an internal control, the constitutive gene Ubiquitin10 was used. Seedlings on AT or MS plates or plants in soil were grown at 20–22°C under constant light.

Transient expression of YFP–RH3D, RAB-D2a and other GFP-based markers in Nicotiana benthamiana leaf lower epidermal cells was performed using Agrobacterium transformation as described by Batoko et al. with slight modifications (Batoko et al., 2000). Plants were cultivated in short-day (8 hours light) conditions. After the Agrobacterium culture reached the stationary growth phase at 28°C with agitation, cells were pelleted and resuspended in the infiltration buffer (100 μM acetosyringone in 10 mM MgCl₂). The final OD₆₀₀ for Agrobacterium cultures was 0.01 except for RAB-D2a(WT) and RAB-D2a(NI) (OD₆₀₀=0.03) as used by Zheng et al. (Zheng et al., 2005b).

Total proteins were extracted with 1× SDS loading buffer [50 mM Tris-HCl, pH 7.5, 100 mM dithiothreitol, 2% SDS (w/v), 0.1% Bromophenol Blue (w/v) and 10% glycerol] from tobacco leaves expressing secGFP with RabD2a(NI), RabD2a(WT), RHD3(WT) and RHD3(SN). Proteins were boiled for 5 minutes and loaded on a 12% SDS-PAGE gel. SDS-PAGE was performed on a Protean III apparatus (Bio-Rad) and separated proteins were transferred onto a PVDF membrane. Western blotting was carried out with a rabbit anti-GFP antibody (Sigma-Aldrich) at 1:2000 dilution and the secondary anti-Rabbit IgG-peroxidase (Sigma-Aldrich) at 1:10,000 dilution. Signals were detected using Invitrogen Novex ECL (HRP Chemiluminescent Substrate Reagent Kit) according to the manufacturer's recommendations.

For ER staining with Rhodamine B hexyl ester, seedlings of rhd3-1::GFP–RHD3, Col-0::YFP–RHD3(S51N) and Col-0::YFP–RHD3(T75A) were immersed in Rhodamine B hexyl ester solution (1.6 μM) for 10 minutes before imaging. For fluorescence imaging of GFP, YFP and Rhodamine B hexyl ester, samples were analyzed with an inverted Zeiss LSM510meta laser scanning microscope with either a C-apochromat 40×/1.2W numerical aperture oil-immersion lens, or a C-apochromat63×/1.2W numerical aperture oil-immersion lens as described by Zheng et al. (Zheng et al., 2005b). Golgi motility was analyzed using volocity quantification (PerkinElmer). Zeiss LSM software (Carl Zeiss), Photoshop (Adobe) and NIH ImageJ (http://rsbweb.nih.gov/ij/) were used for post-acquisition image processing.

To rescue the sey1Δ yop1Δ yeast double mutant and locate RHD3 in yeast cells, RHD3 was subcloned into a yeast expression vector pGREG505 and pGREG574, respectively (Jansen et al., 2005). sey1Δ yop1Δ cells expressing Sec63–GFP (Hu et al., 2009) were transformed with Sey1p in pRS315 (Hu et al., 2009) and RHD3 in pGREG505. Sec63–GFP in yeast cells were imaged with a Leica DMI 6000B microscope (Leica), 100× oil lens and Volocity Acquisition software. GFP–RHD3 expressed in yeast cells was examined with a spinning disc confocal microscope equipped with an EM-CCD camera (Hamamatsu C9100-B).

For the yeast two-hybrid analyses, the corresponding wild-type and mutated genes in the entry vectors were subcloned into the destination vector pNCW-GWRFC.1 to generate all the N-terminal Cub fusions used in this paper, and pNX32 or pXN22 for N- or C-terminal NubG fusion used. KAT1-Cub (ABRC stock CD3-815) and KAT1-NubG (ABRC stock CD3-816) (Obrdlik et al., 2004) were obtained from ABRC. The yeast two-hybrid analysis was carried out according to the method of Obrdlik et al. (Obrdlik et al., 2004).

For bimolecular fluorescence complementation analysis, YFPn–RHD3, YFPc–RHD3 and P24σ1d–YFPc were made by fusion of either residues 1–174 of YFP (termed YFPn) or residues 175–239 (termed YFPc) in corresponding Gateway destination vectors pMDC32–YFPn-C, pMDC32–YFPc-C and pMDC32–N-YFPc. These Gateway compatible destination vectors were created by inserting HindIII–SstI fragments of pUGW2–nEYFP, pUGW2–cEYFP and pUGW0–cEYFP (Nakagawa et al., 2007) into the same cutting site of binary vector pMDC32 (Curtis and Grossniklaus, 2003). YFPn–RHD3 was coexpressed with YFPc–RHD3 or P24σ1d–YFPc at OD₆₀₀=0.03 in tobacco leaf lower epidermal cells. Fluorescence of YFP was then assessed 68 hours post-infiltration with a Zeiss LSM510meta system (Zeiss).

We thank Sylvie Lalonde (Carnegie Institution, Stanford, CA, USA) for pNX32 and pXN22 mbSUS Gateway vectors; David Bird (University of Manitoba, Winnipeg, Canada) for pNCW-GWRFC.1; Tsuyoshi Nakagawa (Shimane University, Matsue, Japan) for pUGW2-nEYFP, pUGW2-cEYFP and pUGW0-cEYFP BiFC gateway vectors; Mark Curtis (University of Zurich, Zurich, Switzerland) for pMDC43; William Prinz (National Institute of Diabetes and Digestive and Kidney Diseases, NIH, Bethesda, MD, USA) for the Sey1p clone and sey1Δyop1Δ::Sec63–GFP yeast double mutant; Inhwan Hwang (Pohang University of Science and Technology, Pohang, Korea) for AHA2–GFP; and Tamara Western for critical reading of manuscript. This work was supported by a grant from The National Science and Engineering Research Council of Canada and a startup grant from McGill University (Montreal, Canada) to H.Z. and a grant from National Science Foundation (MCB-0948584) and from the Department of Energy Great Lakes Bioenergy Research Center and the Chemical Sciences, Geosciences and Biosciences Division, Office of Basic Energy Sciences, Office of Science, US Department of Energy (award number DE-FG02-91ER20021) to F.B.