Septin-dependent compartmentalization of the endoplasmic reticulum during yeast polarized growth

The ER, the major intracellular membrane system of eukaryotes, ensures the biosynthesis of all lipid precursors, as well as the membrane insertion and the translocation through the lipid bilayer of most membrane and secreted proteins (for reviews see Matlack et al., 1998; Meldolesi and Pozzan, 1998; McMaster, 2001; Ma and Hendershot, 2001). As a calcium-storing organelle, the ER plays crucial roles in signal transduction and the regulation of calcium-dependent processes, such as the control of myosin II activity during muscle contraction (Meldolesi and Pozzan, 1998). However, ER function reaches beyond the metabolism and impacts the structural organization of the cell, at least through the formation of the nuclear envelope (Baumann and Walz, 2001). As such, it is “the” eukaryotic organelle par excellence.

The ER is formed of an oxidizing environment enveloped by a single lipid bilayer. It assembles into sheets and reticulated tubules that appear continuous with each other by electron microscopy (Baumann and Walz, 2001). Furthermore, photobleaching experiments showed that ER components freely diffuse throughout the entire ER of fibroblasts (Dayel et al., 1999; Nikonov et al., 2002). Thus, the consensus has emerged that eukaryotic cells contain a single ER. In turn, ultrastructural studies established that the continuous ER membrane is highly organized and forms differentiated domains, such as the nuclear envelope and the rough and smooth ER (Baumann and Walz, 2001). However, we still know little about how these structures differentiate from each other. We also know little about the involvement of the ER in complex cellular processes such as cell polarization and cell division. Particularly, we do not know how the ER is cleaved at or before cytokinesis.

In most cells, the ER is tightly associated with the cytoskeleton, and it colocalizes extensively with microtubules in animal cells (Barr, 2002; Du et al., 2004). This tight association of ER and cytoskeleton suggests that cell polarization might strongly impact on ER organization. Over the last decades, cell polarization has been mainly apprehended as the asymmetric distribution of plasma membrane markers. In epithelial cells, neurons, and yeast, this asymmetry takes the form of functionally and structurally distinct plasma membrane domains that are separated by diffusion barriers (Faty et al., 2002; Boiko and Winckler, 2003). Whether and how the compartmentalization of the plasma membrane affects the internal organization of the cell has not been studied much, and little is known about how cell polarity impinges on ER organization.

The budding yeast Saccharomyces cerevisiae has provided an excellent model to study cell polarity and its molecular mechanism (Pruyne and Bretscher, 2000a,b). This unicellular organism divides by budding; i.e., it polarizes its growth to produce a daughter cell de novo. The restriction of cell growth to the developing bud depends on the polarization of exocytosis and the actin cytoskeleton. Actin cables are nucleated at the bud cortex in a formin-dependent manner and align along the mother-bud axis. These cables serve as tracks for the myosin-dependent delivery of exocytic vesicle to the bud. Thereby, they ensure the polarized delivery of new membrane, cell wall remodeling enzymes, and cell wall material during bud growth. During this process, the yeast plasma membrane is compartmentalized into a bud and a mother domain that are separated by a septin-dependent diffusion barrier (Barral et al., 2000; Takizawa et al., 2000). Septins are GTPases that assemble into membrane-associated filaments. In yeast, these filaments form a ring at the cortex of the bud neck (for review see Faty et al., 2002). Among other functions, this ring establishes the lateral diffusion barrier that helps maintain the compartmentalization of the yeast plasma membrane.

Recent studies suggest that cell polarity not only affects the plasma membrane but also deeper functions of the cell such as protein synthesis. In yeast, at least 24 mRNAs are specifically translated in the bud (Shepard et al., 2003). Localization of these transcripts depends on actin and the type V myosin Myo4 and its binding partner She3 (Jansen, 2001). Little is known about how these mRNAs are maintained in the bud upon transport. However, the fact that at least 18 of them encode membrane proteins (Shepard et al., 2003) suggests that the ER in the bud might play some role in RNA anchoring. In this study, we focused our attention on the impact of cell polarity on the organization of the yeast ER.

The yeast ER is organized into a reticulate network that covers the cell cortex and is connected with the nuclear envelope by cytoplasmic tubules (Preuss et al., 1991; Prinz et al., 2000). To gain further insights into yeast ER organization, we performed fluorescence loss in photobleaching (FLIP) experiments on cells expressing Sec61-GFP. Sec61 is the major subunit of the translocon and resides in the ER membrane (Gorlich et al., 1992). Upon repeated photobleaching of a small cortical area, fluorescence loss in other ER regions indicates exchange of Sec61-GFP molecules with the bleached area. This technique permits the evaluation of exchange rates between different regions of the ER. A small area of the mother cortex was photobleached in medium- to large-budded cells. As seen in Fig. 1 A (Video 1, available), photobleaching led to rapid fluorescence loss over the entire cortex of the mother cell (t_1/2 = 14 ± 2 s; Fig. 1 C). Fluorescence was lost with similar kinetics in the perinuclear ER (t_1/2 = 19 ± 8 s). In contrast, fluorescence decayed slower in the bud ER (t_1/2 = 119 ± 36 s, n = 8). This phenomenon was fully reproducible and the delay of fluorescence loss in the bud was highly significant (Fig. 1 E). It also did not depend on the existence of absence of ER tubules between the perinuclear ER and the bud cortex (Fig. 1, compare A and B; and Video 2, available). The ratio of t_1/2 of the nonbleached compartment over the t_1/2 of the bleached compartment is close to 1 for the perinuclear ER (1.4 ± 0.8), but nine times higher (9 ± 1.7; Fig. 1 G) for the bud cortex. Although the t_1/2 values observed in the different experiments highly depended on the photobleaching protocols used, this ratio remained fairly independent of the size of the bleached area and the duration and intensity of bleaching pulses. Therefore, it is used in the rest of the text to compare experiments and strains. Altogether, these results indicate that Sec61 diffusion is rapid within the mother cortical ER and between the mother cortex and the perinuclear ER. In contrast, Sec61 molecules exchange slowly between mother and bud ERs.

To determine whether the slow loss of fluorescence in the bud is due to its distance to the bleached area, we compared the rate of fluorescence loss between ER regions of the mother and bud that lied at different distances from the bleached area. Fluorescence decayed with similar kinetics in the different domains of the mother cortex (Fig. 1 F, yellow and pink traces). Thus, Sec61-GFP diffusion was instantaneous at the mother cortex, which can be treated as a single compartment. Similarly, the bud cortex also behaved as a single compartment (Fig. 1 F, light and dark blue traces). The region of the bud cortex marked in light blue and the region of the mother cortex marked in yellow (Fig. 1 F) are equidistant from the bleached area. Thus, the fact that these two regions lose fluorescence with distinct kinetics indicated that the difference observed between mother and bud cortices was not due to their different distance to the bleached area.

Translation dramatically slows down the diffusion of Sec61 (Nikonov et al., 2002). Thus, differences in translational activity between mother and bud might cause Sec61 to diffuse slower in the bud than in the mother. This could explain the slow kinetics of fluorescence loss in the bud versus the mother cell. However, photobleaching at the bud cortex (Fig. 2 B and Video 3, available) and FRAP experiments (Fig. S1, available) indicated that Sec61 moved at similar rates in the bud and the mother cell. In these experiments, fluorescence loss was rapid in the bud but slow in the mother. Thus, our FLIP and FRAP data suggest that diffusion of Sec61 through the bud neck is restricted and hence that mother and bud ERs form separate diffusion domains.

A remarkably similar situation was observed for the nuclear envelope during late anaphase. In these cells, exchange between mother and bud cortical ERs remained limited. In addition, the mother cortex exchanged material only with the nuclear half located in the mother cell (Fig. 2 C and Video 4, available; ratio t_1/2 perinuclear ER mother/t_1/2 mother = 2.6, t_1/2 perinuclear ER bud/t_1/2 mother = 11.6, t_1/2 bud cortex/t_1/2 mother = 13.4, n = 6). Concomitantly, the perinuclear ER located in the bud also exchanged rapidly but exclusively with the bud cortical ER (unpublished data). This indicates that in late anaphase the yet undivided nucleus behaved as two distinct domains of diffusion, each exchanging Sec61 molecules only with the cortical ER of the cellular domain in which it was located. Thus, diffusion through the bud neck was limited at all times. At least in the case of the nuclear envelope, a lateral diffusion barrier must have restricted the movement of Sec61-GFP within the continuous ER membrane at the bud neck.

Next, we extended our investigations to other ER proteins. The ER membrane proteins GFP-Sec22 and Hmg1-GFP (Fig. 3, A and B) provided results very similar to those obtained with Sec61-GFP. Sec22 is a v-SNARE required for vesicular transport between the ER and the Golgi (Kaiser and Schekman, 1990; Newman et al., 1990) and GFP-Sec22 localizes almost exclusively to the ER. Hmg1 is one of the two yeast isoforms of the 3-hydroxy-3-methylglutaryl-coenzyme A reductase (Qureshi et al., 1976). Thus, the ER is separated into bud and mother domains by an activity that limits the exchange of ER membrane proteins through the bud neck.

Different results were obtained with the luminal protein GFP-HDEL. Exchange was rapid between the mother and the bud ER (Fig. 3 C and Video 5, available; upon bleaching in the mother: t_1/2 in the bud/t_1/2 in the mother = 2 ± 0.4). GFP-HDEL is targeted to the ER lumen by a signal sequence, while a retrieval sequence (HDEL) ensures ER retention (Pelham, 1991). The fast exchange of GFP-HDEL between mother and bud did not take place through the Golgi because it was not slowed down in the sec18-1, sar1-D32G, and cdc48-6 mutants incubated at the restrictive temperature for at least 30 min (Tables I and II and Fig. 3, D and E). Previous studies indicated that these mutations abolish the function of the protein within 10 min at the restrictive temperature. Sec18 is the yeast homologue of NSF and is required for vesicle fusion with target membranes during antero- and retrograde transport between ER and Golgi (Spang and Schekman, 1998). Sar1 is required for vesicle budding at the ER surface (Barlowe et al., 1993). The AAA-ATPase Cdc48 is required for ER homotypic membrane fusion events (Latterich et al., 1995). Thus, none of the known pathways for ER membrane dynamics was involved in GFP-HDEL exchange between the mother and the bud.

Further data supported the view that GFP-HDEL freely diffused throughout the entire ER. First, in wild type (WT) and mutants, the bud ER exchanged half of GFP-HDEL–associated fluorescence with the mother ER in 12 s (Fig. 3 C). These fast kinetics exclude that it depended on vesicular transport. Furthermore, movement of GFP-HDEL and Sec61-GFP throughout the ER required neither ATP nor GTP (Fig. 4 A). Indeed, when the FLIP experiments were performed in cells treated with sodium azide to deplete metabolic energy, no change in the dynamics of GFP-HDEL and Sec61-GFP was observed. This treatment influenced neither exchange of the markers between mother and bud nor the diffusion of either molecule throughout the mother cortex (Fig. 4 A). In contrast, it immediately stopped nuclear movements and the elongation of the nucleus of anaphase cells, indicating that energy depletion was rapid. Therefore, ER membrane fission, fusion, and vesicular trafficking had little impact on the movement of membrane and luminal proteins within the ER. Thus, the ER must be physically continuous throughout the cell, including the bud neck.

To investigate this possibility, we characterized the morphology of the ER by light and electron microscopy. Using confocal microscopy and GFP-HDEL as a marker, the ER lumen was observed to be continuous through the bud neck (Fig. 4 B) of most cells. The ER membrane appeared also continuous at the bud neck when C₆-BODIPY ceramide was used to visualize the ER bilayer (unpublished data). In contrast, both ER membrane markers Sec61-GFP (see Fig. 6 B) and Hmg1-GFP (not depicted) were seen only very rarely and in those cases only transiently at the bud neck (e.g., Fig. 1 B, Sec61-GFP at 28 s). The difference between the luminal and the ER membrane markers was obvious when GFP-HDEL was coexpressed with Hmg1-CFP, in which case an ER domain labeled with GFP and devoid of CFP was observed at the bud neck (Fig. 4 B). Thus, our data indicate that the ER is continuous through the bud neck but that the ER membrane at this location is different from the rest of the ER membrane because it extensively lacks classical ER membrane proteins such as Sec61 and Hmg1.

Analysis of ER organization by high pressure freezing EM led to similar conclusions (Fig. 4 C). In these images, the ER lumen is apparent as a ribosome-free ribbon around the nucleus and at the cell cortex. ER tubules are also observed in the cytoplasm, some of which emerge from the cortical or the perinuclear ER. In rare instances, these ER tubules linked the perinuclear and cortical ERs. In all images, a ribosome-free ribbon continuous with the mother and bud cortical ERs was observed along the plasma membrane of the bud neck. This ribbon had the same thickness as the ER lumen at other cortical locations and around the nucleus. It was still observed in shs1Δ (Fig. 4 D) and cdc12-6 cells shifted to the restrictive temperature for an hour. Within 2 min of the shift, the cdc12-6 septin allele causes the complete disassembly of the septin ring at the bud neck (Dobbelaere et al., 2003; Dobbelaere and Barral, 2004), which causes the dispersion of all known bud neck proteins (Gladfelter et al., 2001). Shs1/Sep7 is a nonessential septin in yeast (Carroll et al., 1998; Mino et al., 1998). Because the same ribbon is observed independent of septin presence, this ribbon must correspond to the ER lumen and not to the exclusion of ribosomes by septin-specific structures. Thus, both EM and light microscopy images were consistent with the ER being continuous through the bud neck.

Analysis of ribosome arrangement along the ER surface showed the existence of both rough and smooth ER. Ribosomes were well aligned along most of the ER at the mother and bud cortex, which is consistent with the presence of mainly rough ER at these locations (Fig. 4 C, inset 1). In contrast, ribosomes failed to align along the bud neck ER, indicating the presence of smooth ER (Fig. 4 C, inset 2). Furthermore, although the ER did not cover the entire inner surface of the plasma membrane in the bud and the mother, it nearly covered the entire bud neck cortex of WT cells (Fig. 4 C). Quantitative analysis of ER organization (Table I) confirmed that the rough ER was predominant at the mother and bud cortices, whereas the bud neck contained almost exclusively smooth ER. These studies also indicated that the bud neck was reproducibly enriched in ER. These data are in agreement with the exclusion of Sec61-GFP but not GFP-HDEL from the bud neck. Thus, light and electron microscopy data indicate that the tubular cortical ER observed at the mother and bud cortices turns into a ring of smooth ER at the bud neck (see Fig. 7).

Comparison of EM images obtained with the WT and the septin mutant suggested that septins play an important role in the organization of the ER membrane at the bud neck. Indeed, in the cdc12-6 cells shifted to the restrictive temperature, and in the cells lacking Shs1/Sep7, ribosomes nicely aligned along the ER lumen, which was no longer continuous. Thus, differentiation of the ER membrane at the bud neck depended on septin function.

Analysis of ER structure by light and electron microscopy as well as our FLIP data with GFP-HDEL indicated that the yeast ER, like the ER of mammalian cells, is a single and continuous physical entity. However, our FLIP results also indicated that the diffusion of membrane proteins is slowed down in the bud neck. These data suggest the existence of a lateral diffusion barrier in the ER membrane at the bud neck. Alternatively, geometrical constraints might favor the diffusion of luminal over membrane proteins through the bud neck. Comparison of the exchange rate between mother and bud ER for luminal and membrane proteins is not trivial: diffusion in the volume of the ER lumen is not directly comparable to that in the surface of the ER membrane. To circumvent this problem, we developed an in silico modeling approach (see the online supplemental material and Figs. S2 and S3, available). In this model, the exchange between the mother cortical and the perinuclear ERs was set as a standard for the codiffusion of luminal and membrane markers. This analysis indicated that even after standardization the exchange rate of the membrane protein at the bud neck was still reduced by a factor 16 relative to that of the soluble marker. Two models can account for this reduction: either the morphology of the connections offers a larger volume/surface ratio at the bud neck and thereby favors the exchange of the soluble protein over that of the membrane bound, or the diffusion constant is specifically reduced in the membrane. The first model predicts that the connections between mother and bud cortical ERs should be 16 times thicker than those between cortical and perinuclear ER, which is in contradiction to our EM data. Thus, the second model is most likely correct. Our data indicate the existence of a diffusion barrier slowing down the diffusion of proteins in the ER membrane at the bud neck. This conclusion fits with the existence of a differentiated ER membrane at the same location. Thus, we concluded that a diffusion barrier compartmentalized the ER membrane into mother and bud diffusion domains.

Next, we investigated the molecular basis of this diffusion barrier. Most events at the bud neck depend on septins. Furthermore, our EM data indicated that the smooth ER at the bud neck was disrupted in septin mutants. Thus, we investigated whether the septin ring was involved in the separation of the ER membrane into distinct diffusion domains. SEC61-GFP cdc12-6 cells shifted to the restrictive temperature (35°C) for at least 15 min were subjected to FLIP studies. In these cells, fluorescence loss in the unbleached compartments (the bud) was only slightly delayed compared with that in the bleached compartment (Fig. 5, A and C; and Video 6, available; t_1/2 in the bud/t_1/2 in the mother = 3.7 ± 1.4, n = 13, in cdc12-6 vs. 9.8 ± 4.4 in WT, n = 7). Similar results were obtained in cells where septin ring assembly was impaired by the cdc12-1 mutation and in cells lacking Shs1 (Fig. 5 D). In all the FLIP movies, the frequency of cytoplasmic ER tubules connecting mother and buds was not increased in septin mutants compared with WT (e.g., ER tubule growing through the bud neck appeared in 10.4% of the frames of movies of shs1Δ cells, n = 77 frames, vs. 13.4% in movies of WT cells, n = 88 frames), demonstrating that enhanced exchange through the neck of shs1Δ cells was not due to an increased number of cytoplasmic connections. Together, these results indicated that ER compartmentalization depended on septin function. The lack of compartmentalization was not due to the cell cycle delay caused by septin defects because compartmentalization was not restored in cdc12-6 swe1-Δ cells. The swe1-Δ mutation releases the cell cycle block caused by septin defects (Barral et al., 1999; Shulewitz et al., 1999; Longtine et al., 2000).

During budding, the septin ring recruits to the bud neck many proteins involved in actomyosin ring assembly, chitin deposition, cell cycle progression, and spindle positioning (DeMarini et al., 1997; Bi et al., 1998; Lippincott and Li, 1998; Barral et al., 1999; Shulewitz et al., 1999; Segal et al., 2000; Kusch et al., 2002; Longtine and Bi, 2003). Using our FLIP assay, we tested whether any of 30 bud neck proteins representative of these different functions would play any role in ER compartmentalization (Table II). Partial defects were observed in cells lacking both SDKs, Hsl1 and Gin4, simultaneously (Fig. 5 D and Table II). Again, these defects were not due to over-activation of the Swe1 kinase (Table II). SDKs are involved in signaling downstream of septins. In addition, Gin4 is required for proper septin ring assembly, whereas Hsl1 is not (Longtine et al., 2000; Dobbelaere et al., 2003). Because cells lacking Gin4 alone are not defective in ER compartmentalization, septin organization defects do not account for the compartmentalization defect observed in the hsl1Δ gin4Δ double mutant. These results suggest that SDKs act redundantly and downstream of septins on ER organization. They might modify ER proteins at the bud neck.

In addition to SDKs, only Bud6 and Pea2 were involved in ER compartmentalization. In bud6Δ and pea2Δ cells the loss of compartmentalization was similar to that caused by septin mutations (Table II; Fig. 5, B–D; and Video 7, available). Bud6 (Amberg et al., 1997), which is a peripheral membrane protein, has been implicated in actin polarization, microtubule interaction with the cell cortex (Segal et al., 2000), and bud site selection in diploid cells. Together with Pea2, it is a component of the polarizome (Sheu et al., 2000), a complex that also contains either of the formins Bni1 and Bnr1, and the polarity factors Spa2 and Sph1. Analysis of ER compartmentalization in bni1Δ, bnr1Δ, spa2Δ, and sph1Δ cells showed no defect. Thus, the role of Bud6 and Pea2 in ER compartmentalization appeared not to be shared by the other components of the polarizome. Furthermore, like shs1Δ, the bud6Δ mutation did not increase the number of cytoplasmic ER tubules connecting mother and bud (tubules were present in 11.4% of the frames of bud6Δ movies, n = 138 frames, vs. 13.4% in WT movies). Thus, the stimulation of exchange between the mother and the bud in bud6Δ cells was due to a compartmentalization defect and not to an increased number of ER tubules.

Several lines of evidence suggest that Bud6 acts downstream of septins in ER compartmentalization. First, analysis of septin organization using Cdc12-GFP, Cdc3-GFP, or Shs1-GFP as reporters indicated that the bud6Δ mutation did not noticeably affect septin organization (unpublished data). Furthermore, investigation of the localization of Ist2-GFP and Spa2-GFP, which both require a functional septin ring to maintain their asymmetric localization to the cortex of large budded cells (Barral et al., 2000; Takizawa et al., 2000), indicated that the septin-dependent diffusion barrier at the plasma membrane was not affected in bud6Δ cells (Fig. 6 A and not depicted). Thus, bud6Δ does not generally interfere with septin function. Second, investigation of Bud6 localization in the shs1Δ strain indicated that Bud6 failed to accumulate (Fig. 6 C, top) or accumulated to a lesser extent (Fig. 6 C, bottom) at the bud neck of small and medium budded cells, consistent with Huisman et al. (2004). These data suggest that Bud6 acts downstream of septins for the compartmentalization of the ER membrane. Remarkably, Sec61-GFP was still excluded from the bud neck of all bud6Δ cells (Fig. 6 B), in contrast to what we observed in the shs1Δ and cdc12-6 cells (Fig. 6 B and compare Videos 8 and 9, available at http://www.jcb.org/cgi/content/full/jcb. 200412143/DC1). Altogether, these data suggest that septins have two independent effects on ER organization at the bud neck cortex. They govern on one hand the exclusion of the translocon and the formation of a smooth ER domain, and on the other hand the assembly of a Bud6-dependent barrier that limits the diffusion of ER membrane proteins.

In this study, we used photobleaching techniques to investigate ER organization in polarized yeast cells. The diffusion of all tested markers was very fast within the different parts of the ER; i.e, within the ER at the mother cortex, at the bud cortex, and in the nuclear envelope. Remarkably, the yeast ER membrane was separated into distinct domains that reflected cell polarization. This compartmentalization is evidenced by the observation that ER membrane proteins such as the translocon subunit Sec61, the 3-hydroxy-3-methylglutaryl-coenzyme A reductase Hmg1, and the SNARE Sec22 all diffuse very rapidly throughout the bud and mother cortical ERs, but exchange only slowly between these two domains.

Two models can account for such a compartmentalization: either the yeast ER is physically discontinuous, or a diffusion barrier is present within the continuous membrane of the ER at the bud neck. In the first case, residual exchange of ER membrane proteins between the mother and bud must depend on either vesicular transport or transient fusion events. Furthermore, luminal proteins should also be compartmentalized. Our data invalidate both predictions. Indeed, diffusion kinetics of the luminal marker GFP-HDEL and morphological analyses at both the light and electron microscopy levels indicate that the ER lumen is continuous throughout the bud neck. Furthermore, exchange of Sec61 and GFP-HDEL between the mother and the bud was due to passive diffusion because it depended neither on the major players of ER membrane trafficking nor on metabolic energy. Thus, the ER is a physically continuous organelle in yeast like in mammalian cells, and its membrane, but not its lumen, is compartmentalized by a diffusion barrier. Our observation that the membrane of the nuclear envelope is also compartmentalized in late anaphase supports this conclusion, as it indicates that ER compartmentalization at the bud neck does not necessarily require the discontinuity of the ER membrane.

The barrier model was supported by the analysis of our quantitative data by in silico modeling. Indeed, this approach demonstrated that compartmentalization of the membrane markers could not be explained by morphological parameters. The barrier model was also supported by our morphological studies. Indeed, we observe that a specialized ER membrane forms at the bud neck, which is both morphologically (it is a sheet instead of tubules) and structurally different from the rest of the cortical ER. Thus, we conclude that a diffusion barrier compartmentalizes the ER membrane along the same lines as those compartmentalizing the plasma membrane of budded yeast cells. To our knowledge, this is the first report of such an intricate relationship between the organization of the ER and that of the plasma membrane.

What could be the cellular function of this complex ER organization? We see two possibilities: (1) the specialized ER domain at the bud neck might have functions related to cell cleavage and/or (2) the separation of the bud and mother cortical ER membranes might provide support for cell polarization.

At the site of division, a specialized ER domain could play two distinct roles. First, this ER domain might serve as a calcium-storing structure involved in the control of actomyosin contraction. This ER domain could work like the sarcoplasmic reticulum of muscle cells in the control of muscle contraction. In support of this idea, septin defects, which cause the loss of the ring of smooth ER, slow down actomyosin ring contraction during cytokinesis (Dobbelaere and Barral, 2004). However, the isolation of mutants that specifically abrogate the formation of the ring of smooth ER at the bud neck will be required to investigate this possibility more rigorously. Second, the ring of smooth ER might be required for ER cleavage at cytokinesis. Further studies will be required to investigate the timing, control, and mechanism of ER cleavage before the completion of cytokinesis.

The compartmentalization of the ER membrane might also play some roles in cell polarity. For example, it could help restrict the synthesis of specialized membrane proteins to one cellular compartment. Most mRNAs that specifically localize to the yeast bud encode membrane proteins and must be translated at the ER surface (Shepard et al., 2003). Thus, ER compartmentalization might help maintain the localized mRNAs in the bud. This mRNA–ER connection seems strengthened by the observation that mRNA and ER transport to the bud neck both involve the type V myosin Myo4 and its associated factor She3 (Estrada et al., 2003). Remarkably, She3 localizes to the bud ER specifically, suggesting that ER compartmentalization might indeed support some structural and functional asymmetry of the ER. However, due to the leakiness of the barrier indicated by our data, it is unclear which kind of asymmetry it helps maintain. Indeed, even though membrane proteins diffused much slower through the bud neck than through the bud or mother cortex, they did eventually translocate from one domain to the other. Thus, the ER diffusion barrier alone cannot maintain very long the asymmetric distribution of polarized factors. Consistently, it has been shown for several products of asymmetrically localized mRNAs that they do not maintain an asymmetric distribution (Shepard et al., 2003), particularly those that are residents of the ER membrane. However, some other products of polarized mRNAs do maintain an asymmetric distribution, the prototype of which is the plasma membrane protein Ist2. In the case of these products, it might be crucial that the ER diffusion barrier slows down their dispersion before they exit the ER, en route to the plasma membrane. Indeed, at least in the case of Ist2, the final localization of the product is determined by the localization of the mRNA (Takizawa et al., 2000). This would not be possible if the product would freely diffuse through the entire ER. Thus, one function of the diffusion barrier in the ER membrane at the bud neck might be to slow down the diffusion to the mother cortex of both polarized mRNAs and their product. Thereby, it permits dynamic processes, such as the movement of Ist2 to the plasma membrane, to remain asymmetric, whereas slower events become isotropic. These observations are reminiscent of the situation recently described in Drosophila melanogaster embryos, where the localization of the Gurken mRNA determines where at the surface of the embryo this growth factor–like signaling molecule is being secreted (Herpers and Rabouille, 2004). It will be interesting to investigate whether diffusion barriers compartmentalize the ER membrane of the D. melanogaster embryo to ensure the accuracy of Gurken targeting, and perhaps to also ensure that each of the distinct nuclei acquire and maintain a distinct identity despite being all in the same cytoplasm. Indeed, the division of the ER might support the organization of the entire cytoplasm into distinct domains, such as during the polarization of yeast cells. Thus, we suggest that the ER might play more profound roles in cellular organization than previously anticipated.

The observation that a diffusion barrier compartmentalizes the ER membrane opens two mechanistic questions: (1) what is the molecular nature of this barrier, and (2) how is it localized to the bud neck? Remarkably, we found only few of the known bud neck proteins to be involved in ER compartmentalization. Thus, the identified candidates might act in a very specific manner.

On the septin side, both disruption of the entire ring and elimination of the nonessential septin Shs1/Sep7 dramatically affected both the ER diffusion barrier and the accumulation of smooth ER in the bud neck. Because the shs1Δ mutation does not dramatically affect cell viability, septin ring assembly, and morphology (Carroll et al., 1998; Mino et al., 1998), but strongly impacts on ER organization, Shs1 may be specialized for this process. In turn, Bud6 affected only the diffusion barrier and not ER morphology. Thus, the smooth ER at the bud neck does not depend on the presence of the barrier and is not sufficient to form a barrier. Previous data have established that Bud6 is associated with internal membranes (Jin and Amberg, 2000). We propose that Bud6 is a smooth ER protein (Fig. 7), where it might influence protein diffusion.

The role of septins in the recruitment or regulation of ER proteins such as Bud6 might be indirect and mediated at least in part through signaling. Indeed, we found that the SDKs Gin4 and Hsl1 are also involved in the formation of the diffusion barrier. Because SDKs did not seem to function in smooth ER organization, their substrates must be specifically involved in ER compartmentalization. It will be helpful to identify these substrates and to investigate whether Bud6 is one of them. The view that septins act on the ER via signaling mechanisms might explain how they could potentially act at a distance during, for example, the compartmentalization of the nuclear envelope at the bud neck during anaphase.

It is unclear to which extent the observations that we describe here are specific to ER of fungi or relevant for other eukaryotes. At first sight, the fact that the ER of animal cells is mainly cytoplasmic, and not cortical as in fungi, suggests that ER compartmentalization is less likely in metazoans. It will be interesting to perform FLIP experiments on polarized and dividing cells and to investigate whether the role that we ascribe to the cleavage apparatus in the compartmentalization of the yeast ER is conserved during the division of animal cells.

Yeast strains were constructed by standard genetic techniques. Diploids were isolated on selective medium and subsequently sporulated at 23°C. The background is, unless specified otherwise, S288c. ssGFP-HDEL was expressed from the 2-μm vector pG14 (Lesser and Guthrie, 1993); ssGFP-HDEL has the signal peptide of CTS1 and the HDEL retrieval sequence at its COOH terminus (a gift from E. Bertrand, Centre National de la Recherche Scientifique, Montpellier, France). Strains containing Sec61-GFP (provided by D. Liakopoulos, Swiss Federal Institute of Technology [ETH], Zurich, Switzerland), Spa2-GFP, Bud6-GFP, Hmg1-GFP, or pGAL-GFP-Sec22 were made using the PCR-based integration system (Longtine et al., 1998). The different deletions shown in Table I were obtained from the EUROSCARF deletion collection (S288c) and provided to us by M. Peter (ETH, Zurich, Switzerland). Each time that an effect was observed, the mutation was backcrossed several times into our background. The hsl1Δ, gin4Δ, shs1Δ, and swe1Δ strains are isogenic with S288c. Cdc12-GFP was expressed from a centromeric plasmid (Dobbelaere et al., 2003). The sec18-1 and cdc48-6 strains were gifts of R. Collins (Cornell University, Ithaca, NY) and S. Jentsch (Max Planck Institute, Münich, Germany), respectively.

Cells were grown on YPD plates, resuspended in liquid nonfluorescent medium, and immobilized on nonfluorescent medium (Waddle et al., 1996) containing 1.6% agarose. Photobleaching was applied on the area shown on the figures, using a microscope (model LSM510; Carl Zeiss MicroImaging, Inc.) and a Plan-Apochromat 100× objective (NA 1.4). For FRAP, scans were collected at 5-s intervals for a minimum of 120 s using the acquisition software LSM510 (Carl Zeiss MicroImaging, Inc.). Bleaching regions were irradiated with 250 iterations of 50% laser intensity at 30% output of an argon laser (488 nm) and scans were collected with typically 1% laser intensity at the same conditions. All pictures of FLIP experiments shown in the figures were treated to account for the bleaching due to image acquisition, whereas the movies were left untreated.

Pictures shown in Fig. 4 B were taken on a spinning-disc confocal system (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a Plan-Apochromat 100× objective. The overlay picture was taken on a DeltaVision microscope (Applied Precision) and deconvolved using the softwox software.

cdc12-1 mutant cells were grown to early log phase at permissive temperature (24°C) in liquid YPD medium and arrested in G1 with 5 μg/ml α-factor for 2.5 h. After removal of α-factor from the medium by washing twice with fresh YPD, cells were mounted on nonfluorescent agarose beds as described in the previous section and immediately shifted to 35°C on a heated stage. FLIP analysis was performed on buds formed after the temperature shift, starting after 30 min. cdc12-6 cells were grown to mid-log phase on YPD plates at 22°C (permissive), mounted on agarose beds, and shifted to 35°C on the heated stage. FLIP analysis was performed in medium- to large-budded preanaphase cells 30 min after temperature shift.

Analysis of the FLIP experiments was performed using the ImageJ 1.29 software (http://rsb.info.nih.gov/ij). The loss of fluorescence over time was measured in different regions of interest (usually the mother cortex, bud cortex, and perinuclear ER). In addition, we measured the loss of fluorescence on neighboring control cells to account for the loss due to visualization. Finally, we measured the intensity of the background. The intensity in the region of interest was calculated as (region − background)/(control cell − background) and then put in fractions.

For BODIPY staining, cells were grown overnight to early to mid-log phase in YPD, harvested, and resuspended in SC media to 5 OD₆₀₀/ml. The culture was incubated for 10 min at 30°C under agitation. Defatted BSA was added to a final concentration of 5 mg/ml and supplemented with 2.5 μl C6 BODIPY ceramide (4 mM stock in DMSO; Molecular Probes). The cells were incubated for 20 min at 30°C under agitation and mounted for direct inspection. Alternatively, the cells were fixed after the incubation period with 4% formaldehyde. No difference in the staining was detected between life and fixed cells. Light microscopy was performed using either a DeltaVision microscope (DeltaVision Spectris System; Applied Precision) equipped with a Coolsnap HQ camera (Roper Scientific) or an Olympus BX50 equipped with a camera Imago (TiLL Photonics). In all cases, we used 100× objectives of NA 1.4. Deconvolution was performed by a constrained iterative method using the softwox software (Applied Precision). Further image processing was performed using the Photoshop software (Adobe) and was reduced to the optimization of the levels. No gamma adjustments were applied. Unless otherwise indicated, light microscopy was performed at RT (22°C) as described previously (Dobbelaere et al., 2003). EM was performed as described previously (Sandmann et al., 2003).

The quantitative transport model was formulated on the basis of standard physical principles, leading to a system of coupled differential equations. Photobleaching and scanning cycles were modeled as algebraic equations.

The quantitative transport model is fully described and assessed in the online supplemental material.

We are grateful to Markus Aebi and Ari Helenius for helpful discussions. We thank E. Bertrand, R. Collins, S. Jentsch, and M. Peter for strains and constructs; D. Liakopoulos for the Sec61-GFP construct; and J. Dobbelaere for technical help.

This work was supported by the Swiss Federal Institute of Technology (ETH), the Swiss National Science Foundation (Y. Barral), and the Max Planck Society (A. Spang). S. Buvelot Frei was supported by a postdoctoral fellowship from the Roche Foundation and I. Sbalzarini by the ETH Strategic Excellence Project in Computational Science. Y. Barral and A. Spang are European Molecular Biology Organization Young Investigators.