|Home | About | Journals | Submit | Contact Us | Français|
The endoplasmic reticulum (ER) in S. cerevisae is largely divided between perinuclear and cortical compartments. Yeast Nvj1 localizes exclusively to small patches on the perinuclear ER, where it interacts with Vac8 in the vacuole membrane to form nucleus-vacuole (NV) junctions. Three regions of Nvj1 mediate the biogenesis of NV junctions. A membrane-spanning domain targets the protein to the ER. The C-terminus binds Vac8 in the vacuole membrane, which induces the clustering of both proteins into NV junctions. The luminal N-terminus is required for strict perinuclear localization. 3D cryo-electron tomography reveals that Nvj1 clamps the separation between the two nuclear membranes to half the width of bulk nuclear envelope. The N-terminus contains a hydrophobic sequence bracketed by basic residues that resembles outer mitochondrial membrane signal-anchors. The hydrophobic sequence can be scrambled or reversed without affecting function. Mutations that reduce the hydrophobicity of the core sequence, or affect the distribution of basic residues, cause mislocalization to the cortical ER. We conclude that the N-terminus of Nvj1 is a retention sequence that bridges the perinuclear lumen and inserts into the inner nuclear membrane.
The endoplasmic reticulum (ER) is an interconnected network of membranes that encircles the nucleus and extends to the cell cortex. The ER is topologically continuous and shares a common lumen, yet it is divided and further subdivided into increasingly smaller and functionally specialized domains (1, 2, 3). In this fashion many of the diverse functions of the ER, such as protein and lipid biogenesis, vesicular and non-vesicular trafficking, ion homeostasis, and intracellular signaling, are segregated into a mosaic of temporally and spatially dynamic ER compartments. A significant amount of current effort is focused on understanding the structure, function and biogenesis of ER subdomains.
The nuclear envelope is an especially interesting ER compartment because it houses the genome and regulates many aspects of gene expression. The topology of the nuclear envelope is unusual; it consists of two concentric membrane sheets connected by wormhole-like channels called nuclear pores. The inner nuclear membrane is associated with chromatin and, in higher eukaryotes, is linked to the nuclear lamina (4). The fungal nuclear envelope associates with chromatin but lacks a classical nuclear lamina. The outer nuclear membrane is mostly rough ER and, as such, is studded with ribosomes in the process of translating membrane or secreted proteins.
The ER communicates with organelles such as mitochondria and the plasma membrane through membrane contact sites (5), which may be mediated by stable protein-protein interactions (6, 7). The S. cerevisiae nuclear envelope forms unique membrane contact sites with the vacuole called nucleus-vacuole (NV) junctions. NV junctions are Velcro-like patches formed by heterotypic bonds between Vac8 in the vacuole membrane and Nvj1 in the outer nuclear membrane (8). NV junctions are sites of Piecemeal Microautophagy of the Nucleus (PMN), a unique form of selective autophagy that targets non-essential portions of the yeast nucleus (9). During PMN, a portion of an NV junction and the underlying nucleoplasm form a bleb that is pinched-off into the vacuole lumen and degraded. PMN is induced by starvation, during which Nvj1 levels and the surface area of NV junctions increases proportionally (9).
The exclusion of Nvj1 from the cortical ER is the key factor in directing the formation of NV junctions to the perinuclear domain of the ER. Even in vac8-Δ cells, where Nvj1 is free to diffuse within the plane of the membrane, it remains excluded from the cortical ER (8, 10). Not much is known about the selective localization of eukaryotic proteins to the perinuclear ER, also known as the outer nuclear membrane. Beside certain nuclear pore complex proteins, which represent a special case, the best-understood outer nuclear membrane proteins are KASH domain (Klarsicht, ANC-1, and Syne homology)-containing proteins. KASH-domain proteins are retained in the outer nuclear membrane through interactions across the perinuclear lumen with SUN domain proteins (Sad1p and UNc-84 homology), which are themselves anchored in the inner nuclear membrane via interactions with nuclear lamins (11, 12, 13, 14). These fascinating proteins establish a chain of protein-protein interactions that connect the cytoskeleton across the nuclear envelope to the interior of the nucleus.
Nvj1 is a modular protein that contains a single membrane-spanning domain which is sufficient and necessary for ER targeting (8, 10, 15, 16). The C-terminus of Nvj1 is exposed to the cytoplasm, where it binds Vac8 (17, 18). The N-terminus is exposed to the ER lumen. Nvj1 also binds Osh1 and Tsc13, proteins with roles in lipid metabolism (10, 15). Here we show that the lumenal N-terminus contains an inner nuclear membrane anchor that actively retains Nvj1 in the perinuclear ER and prevents its leakage to the cortical ER.
S. cerevisiae strains used in this study were based in the YEF473 (MATa) genetic background (trp1-Δ63 leu2-Δ1 ura3-52 his3-Δ200 lys2-801) (19). Deletions of NVJ1 and VAC8 in YEF473 were described elsewhere (8, 17). Cells were cultured at 30°C in standard YPD, synthetic complete media (SC), or dropout media containing 2% glucose (20). The Saccharomyces castellii strain used in this study was MATa and ura3 (21, 22). The two paralogs of VAC8 were deleted with using kanamycin and nourseothricin (CloNat; Werner Bioagents) antibiotic resistance markers (23, 24, 25). S. castellii cells were cultured at 25°C in standard YPD, synthetic complete media (SC), or dropout media containing 2% glucose (26). The Kluyveromyces lactis strain MW98-8C used in this study was Mat, uraA, argA, lysA, K+ pKD1(0) (27). Deletion of VAC8 in K. lactis was described elsewhere (8, 17). Cells were cultured at 30°C in standard YPD, synthetic complete media (SC), or dropout media containing 2% glucose (20).
Unless otherwise indicated, the expression of plasmid-borne reporters was induced in log-phase cells. EGFP-labeled truncations of Nvj1 were expressed from pRS316 MC-EGFP (N. Shulga, unpublished vector pNS198) by culturing in media lacking methionine for 4 hours. Vacuoles were stained with FM4-64, and nuclei with Hoechst as previously described previously (17, 10). Confocal microscopy was performed on a Leica TCS NT microscope equipped with a 100X Fluorotar lens and UV, Ar, Kr/Ar, and He/Ne lasers (Leica Microsystems, Chantilly VA). Images were processed using Adobe PhotoShop 5.0 (Adobe Systems, CA).
Cells were was grown to mid-log phase overnight at 35°C in YE medium with continuous rotation. Yeast cells were harvested from rapidly growing cultures by gentle vacuum filtration onto Millipore filters. This yeast paste was then removed from the filter with a spatula and placed directly onto a planchette and rapidly frozen in a HPM-010 high-pressure freezer (Bal-Tec AG, Liechtenstein). Cells prepared using an isotonic sucrose buffer were indistinguishable from cells frozen in 20% Dextran (28).
Vitreous cryosections were prepared as previously described (29). Briefly, samples contained in the lower portion of a brass freezing planchette were mounted in a UltraCut UCT/FCS cryo-ultramicrotome (Leica Microsystems, Vienna), using a custom-made collet holder cooled to −145°C. Blocks were trimmed to an appropriate size with a diamond trimming tool (Diatome US, Hatfield, PA) and sectioned using a “Cryo-Platform” diamond knife with a 25° included angle (Diatome). During cutting, ribbons of vitreous sections were controlled with a Leitz micromanipulator (Model M; Leica) and collected onto carbon-coated 200-mesh molybdenum EM grids (Electron Microscopy Sciences, Hatfield, PA). Samples were stored in liquid nitrogen, then transferred to the EM with a cryo-transfer specimen holder (Gatan, Pleasanton, CA).
Low magnification montages (~140X) were collected with the Navigator program in SerialEM to characterize the distribution of cryosections on the EM grids. Images were recorded at 200kv or 300kv using Tecnai F20 or F30 EMs (FEI, Eindhoven). The TF30 was equipped with a post-column energy filter (GIF from Gatan Inc., Pleasantown, CA. Tilt series were recorded on Gatan CCD cameras using the SerialEM tilt series aquisition program. The tilt range was usually from −60° to 60° in increments of 1.5°. Images were recorded with the objective lens defocused between −6 μm and −10 μm. Total dose per tilt series was <100e/Å2. A cos (tilt angle) multiplier was used during the tilt series to increase the dose and keep exposure essentially constant at a higher tilts.
Alignment of multiple tilted views (both with and without fiducial points), tomographic reconstruction by back-projection, and viewing of the resulting 3D maps of cell structure were performed using the IMOD software (see also <bio3d.colorado.edu>). In general, tilted views were aligned by cross-correlation of structural features, including spots of ice contamination on the section. The thickness of vitrified sections was measured directly from the tomograms.
The membrane-spanning domain of Nvj1 is sufficient to target the protein to the ER. Specifically, a 34 residue peptide (aa 87-120) encompassing the membrane-spanning domain of Nvj1 fused to EGFP localized to the perinuclear and cortical ER (16). The N-terminus is not required for targeting to the ER membrane, but is required for exclusive localization to the outer nuclear membrane. Reporters lacking even the first 6 residues of the N-terminus partially mislocalize to the cortical ER in vac8-Δ cells (16). N-terminal truncations are still capable of binding Vac8, Osh1 and Tsc13 via their cytoplasmic domain; thus, their proper orientation in the membrane is maintained (10, 15). In VAC8+ cells N-terminal truncations of Nvj1 are not retained in the perinuclear ER and are free to direct the assembly of both perinuclear and extra-perinuclear ER-vacuole contact sites (16). Extra-perinuclear contact sites directed by N-terminal truncations of Nvj1 arise as expansions of the outer nuclear membrane, and result in the local separation of the inner from outer nuclear membranes (16). These ectopic ER-vacuole junctions rarely form at the periphery of the cell, suggesting that cortical ER is for some reason resistant to forming junctions with vacuoles. Taken together, these results suggested that the N-terminus might retain Nvj1 in the nuclear envelope by anchoring to the inner nuclear membrane (16). An inner nuclear membrane anchor would explain both the strict localization of Nvj1 to the perinuclear ER, and provide a mechanism to coordinate the concerted exvagination of the inner and outer nuclear membranes during PMN (9).
To test this hypothesis we investigated the sequence requirements for the strict perinuclear localization of Nvj1. A prominent feature of the N-terminal sequence is an 11 residue hydrophobic sequence (the “core”) beginning at Ile10 (IFSLGLSVAVL; see Fig. 1). We created a series of mutants to investigate the role of the hydrophobic core in perinuclear localization. Reducing the hydrophobicity of this sequence by replacing three strongly hydrophobic residues with Ala (I10A, L13A L15A) disrupted localization (Fig. 1). Introducing charged residues (L20E, V23E) into this region also disrupted localization (Fig. 1). These results indicate that the hydrophobicity of the core is required for strict localization of Nvj1 to the perinuclear ER.
We next asked whether the specific order of amino acids within the core (IFSLGLSVAV) is important. Reporters were created that contained either a reversed or a scrambled sequence. The hydrophobicities of both the reversed (VAVSLGLSFI) and scrambled (LSVGFASLVI) peptide sequences are identical to the native sequence, and both directed strict perinuclear localization (Fig. 1). Thus the overall hydrophobicity of the core is more important than the particular order of amino acids.
The sequence properties of the N-terminus are similar to the N-terminal signal-anchor sequences of the outer mitochondrial proteins Om45, Tom20 and Tom70 (30; 31). These N-terminal sequences insert post-translationally into the outer mitochondrial membrane, leaving the bulk of the polypeptides exposed to the cytosol (32). Among these, the Om45 signal-anchor sequence is most similar to the N-terminal sequence of Nvj1. Both Om45 and Nvj1 N-termini contain hydrophobic cores with peak hydrophobicities of >1.5 and are bracketed by basic residues (Fig. 1). To determine if the Om45 signal-anchor sequence can replace the function of the native Nvj1 sequence, we replaced the first 35 residues of Nvj1 with the corresponding 35 residues of Om45. This sequence should be sufficient since the analogous 33 residue signal-anchor sequence of rat Tom20 was sufficient to target a GFP reporter to mitochondria in COS-7 tissue culture cells (33). In vac8-Δ cells, the Om45-Nvj1-EGFP chimera localized to both the perinuclear ER and cortical ER, but not to mitochondria. In VAC8+ cells, this chimera formed ectopic ER-vacuole junctions (Fig. 1). Thus, the mitochondrial signal-anchor sequence of Om45 did not replace the perinuclear ER targeting activity of the native Nvj1 sequence. The targeting of the Om45-Nvj1-EGFP chimera to the ER network instead of mitochondria is consistent with previous work showing that the membrane-spanning domain of Nvj1, and not the signal peptide-like N-terminus, plays a dominant role in ER targeting. Although the overall hydrophobicity of the Om45 core is high (ΔG=2.0) relative to the Nvj1 core (ΔG=1.6), it does contain a preponderance of Ala residues (AALAAAITASIMV). The replacement in the Nvj1 core of three strongly aliphatic amino acids with Ala (AFSAGASVAVL) disrupted localization, but here these changes also reduced the hydrophobicity of an 18 residue window encompassing the core (ΔG = −3.5).
The Om45 sequence might not function in the context of Nvj1 for several reasons, including the lack of strongly hydrophobic residues in the core, the particular number and distribution of charged residues bracketing the core (see below) and, possibly, the absence of downstream sequence elements that may contribute to the efficiency of Nvj1 anchoring.
Basic residues, usually Arg or Lys, commonly bracket the hydrophobic core domains of signal peptides, membrane-spanning domains, and signal-anchor sequences. These charged residues are thought to influence the orientation of the sequence in the membrane and probably stabilize the interaction by forming salt bridges with the acidic head groups of phospholipids at the membrane-water interphase (34). To investigate the importance of charged residues near the hydrophobic sequence of Nvj1, we changed most of the charged residues throughout the lumenal domain (aa 1-96) one at a time to Ala (Fig. 2A). Only two of 21 changes (R28A & K29A) affected localization in vac8-Δ cells (Fig. 2A, B). The loss of K29 had a significantly greater effect on localization than the loss of K28 (Fig 2A, B). A reporter replacing K29 with Arg (K29R) showed normal localization (Fig. 2B), indicating that a basic charge at this position is important, and not a lysine residue per se.
The sequence upstream of the hydrophobic core, which is necessary for perinuclear ER localization, contains a mix of hydrophobic, polar, and basic residues. Reporters lacking the first 6 or 11 residues partially mislocalized to the cortical ER (16). Although individual R3A or R8A mutations did not affect localization, the double mutant (R3A R8A) did mislocalize in both vac8-Δ and VAC8+ cells (Fig. 2B). We conclude that basic charges on both sides of the hydrophobic core contribute to the efficient localization of Nvj1 to the perinuclear ER. It should be noted that the R3A R8A mutation increases the hydrophobicity of the N-terminus, which could partially compensate for the loss of these stabilizing basic charges by increasing the affinity of the sequence for the hydrophobic phase of the inner nuclear membrane (see below).
It is possible that Nvj1 is actively excluded from the cortical ER. The cortical ER is a network of tubular membranes that contrasts with the sheet-like structure of the nuclear envelope. Nvj1 could be specialized to promote the local biogenesis and expansion of sheet-like membranes, and might be sterically excluded from tubular membranes. Deleting the reticulon genes RNT1 and RTN2 converts the cortical ER from tubes into sheet-like cisternae (35, 36). Nvj1-EGFP still localized exclusively to the outer nuclear membrane of rtn1-Δrtn2-Δ cells (data not shown). Therefore, the exclusive localization of Nvj1 to the outer nuclear membrane is not due to its exclusion from tubular membranes. Instead, these results support the model that Nvj1 is absent from the cortical ER because it is selectively retained in the outer nuclear membrane.
If the N-terminus actively retains Nvj1 in the perinuclear ER by bridging the perinuclear lumen and inserting into the inner nuclear membrane, then the physical dimensions of the luminal domain should determine the separation between the outer and inner nuclear membranes. As a test we employed high-resolution 3D TEM tomography to quantify the separation between the three membranes comprising the NV junction. The consistent separation between the three membranes within NV junctions is apparent by visual inspection of these images (Fig. 3). The separation between the outer nuclear membrane and vacuole membrane in NV junctions is 18.2 ± 2.7 nm (standard deviation). This distance is determined by the volume of the complex formed between Vac8 and the cytoplasmic domain of Nvj1. The separation between inner and outer nuclear membranes in NV junctions is 8.6 ± 1.7 nm (Fig. 3). In contrast, the separation between inner and outer nuclear membranes in bulk nuclear envelope is about twice what it is in NV junctions (18.7 ± 6.4 nm) and, moreover, is significantly more variable (Fig. 3). These results are consistent with the model that the luminal N-terminus of Nvj1 associates with the inner nuclear membrane to physically link the inner and outer nuclear membranes.
We made several attempts without success to alter the separation between the inner and outer nuclear membranes within NV junctions by increasing the length of the luminal domain of Nvj1. Regardless of whether we duplicated sections of the luminal domain, or inserted sequences from other proteins, none of the engineered proteins either mediated the formation of NV junctions or were expressed at detectable levels. Apparently, this system does not tolerate large modifications of the N-terminal domain.
Instead, we sought to determine the width of the perinuclear lumen within NV junctions formed by a mutant Nvj1 reporter that lacks the putative membrane anchoring domain (Nvj1(1-26Δ)-EYFP). This reporter is not anchored in the outer nuclear membrane and localizes throughout the ER network (16). The pool of Nvj1(1-26Δ)-EYFP that localizes to the perinuclear ER is competent to form pseudo-NV junctions. Images of these structures were used to quantify inter-membrane distances. Without the anchoring sequence, the width of the perinuclear lumen within the NV junctions formed by Nvj1(1-26Δ)-EYFP should not be any different than in bulk nuclear envelope. A confounding concern was the possibility that that the positioning of the vacuole adjacent the nucleus would be sufficient to compress the perinuclear lumen. However, this does not appear to be the case. The width of the perinuclear lumen under the mutant NV junctions was 18.7 ± 4.3 nm (not shown), which is indeed similar to the separation between these membranes in bulk nuclear envelope. We conclude that the N-terminal 26 residues of Nvj1 are necessary for physically linking the inner and outer nuclear membranes.
NVJ1 orthologs are restricted to the subphyla Saccharomycotina (37), including those belonging to both pre- and post-genome duplication lineages. Fig. 5A shows that the S. cerevisiae sequence is only 70% identical to the ortholog in the closely related “stricto sensu” species S. bayanus, and drops off precipitously in more divergent species. Most of the orthologs contain a single putative membrane-spanning domain and homology to the Osh1- and Vac8-binding motifs. The C. glabrata sequence is unique in lacking the signature Osh1-binding motif (Fig 5B). The N-termini of all but the K. lactis and A. gossypii sequences contain stretches of hydrophobic residues (Fig. 5D). The extremely divergent A. gossypii sequence was identified by synteny, and is dissimilar except for a putative membrane-spanning domain and Osh1 binding motif (Fig. 5A, B, C, D). Aspergillus species encode an ORF with a single putative membrane-spanning domain and a downstream Osh1 binding motif (Fig. 5B). This gene contains none of the other motifs characteristic of S. cerevisiae Nvj1, thus its meager similarity to bona fide orthologs may be coincidental.
Natural variation in the N-terminal sequences of orthologous proteins provides an opportunity to explore the sequence requirements for perinuclear targeting. The protein coding sequences of the S. castellii, S. kluyveri, C. glabrata, K. lactis and K. waltii orthologs were fused to EGFP and expressed in S. cerevisiae from plasmids using the MET25 promoter in VAC8+ and vac8-Δ cells. All five EGFP reporters localized to the ER in S. cerevisiae (Fig. 4). Only the S. kluyveri and K. waltii reporters localized exclusively to perinuclear ER in vac8-Δ cells, and formed only normal NV junctions in VAC8+ cells. The cloned A. gossypii ortholog was shown to be correct by DNA sequencing, but did not express in S. cerevisiae and was not pursued. The S. castellii, C. glabrata, and K. lactis reporters mislocalized to the cortical ER in vac8-Δ cells and formed ectopic ER-vacuole junctions in VAC8+ cells (Fig. 4). Because they mediate close interactions between ER and vacuoles, all five of these orthologous proteins presumably interact with S. cerevisae Vac8 via conserved C-terminal Vac8-binding sequences (Figs. 4 & 5). VAC8 itself is extremely well conserved among these yeasts (BLAST E value < 10−100).
The inclusion or exclusion of the orthologous proteins from the cortical ER in S. cerevisiae correlates with the peak hydrophobicities of their N-terminal sequences. Reporters with high hydrophobicities (S. kluyveri and K. waltii) localized strictly to the perinuclear ER, while those with low hydrophobicities (S. castellii, C. glabrata, and K. lactis) co-localized to both the perinuclear and cortical ER (Fig. 4 & 6A). To further investigate this relationship, we increased the overall hydrophobicity of the S. castellii sequence by substituting Ile for Ala10 and Leu for Gly15. This mutant reporter localized strictly to the perinuclear ER in S. cerevisae (Fig. 6A). This result reinforces our hypothesis that the general hydrophobicity of the core, and not a specific sequence of amino acids, is important for function of the anchor.
In addition to a sufficiently hydrophobic core sequence, bracketing basic residues are also important for the strict perinuclear ER localization of Nvj1 (Fig. 2). The three orthologous sequences that localize strictly to the outer nuclear membrane in S. cerevisiae contain basic residues in similar positions, including one immediately upstream of the hydrophobic core. The first eight residues of the S. castellii (MTRPPVMQ) ortholog are nearly identical to the first eight residues of S. cerevisiae Nvj1 (MTRPPLVR). The key difference is the presence of an uncharged Gln at position 8 instead of the Arg present in S. cerevisiae Nvj1. A S. castellii reporter containing a Q8R substitution localized exclusively to the perinuclear ER in S. cerevisiae (Fig. 6A). Therefore, S. castellii Nvj1 can be converted from a protein that localizes throughout the S. cerevisae ER network to one that localizes exclusively to the perinuclear ER by increasing either the hydrophobicity of the core sequence or by adding a second basic residue upstream of the core. We conclude that the anchoring activity of the N-terminus is promoted both by the relative hydrophobicity of the core and by bracketing basic charges. A deficiency in one parameter can be compensated for by an increase in the other.
Although the S. castellii protein mislocalized when heterologously expressed in S. cerevisiae (Fig. 6A), it formed only NV junctions in wt S. castellii cells, and it localized exclusively to the nuclear envelope in vac8-ΔS. castellii cells (Fig. 6B). This result suggests that the cellular environment, possibly involving the protein and/or lipid composition of the inner nuclear membrane, plays an important role in establishing efficient membrane anchoring. In contrast, the divergent K. lactis ortholog, which lacks any semblance of an N-terminal anchor sequence, did not localize exclusively to the nuclear envelope in K. lactis, but instead formed patches both in the nuclear envelope and what appears to be the extranuclear ER network (Fig. 6C). These patches may or may not represent junctions with K. lactis Vac8. Importantly, expression of S. cerevisiae Nvj1-EGFP in K. lactis resulted in strict perinuclear loclalization, and no evidence of perinuclear or ectopic junctions. It is possible explanation for this localization is that the S. cerevisiae protein does not bind tightly to K. lactis Vac8. Still, the strict perinuclear localization of the S. cerevisae protein in K. lactis suggests that it is efficiently anchored in the inner nuclear membrane.
Nvj1 is a remarkable protein that physically links three biological membranes, sequesters Osh1 and Tsc13, and mediates the coordinated bending of the entire structure during PMN. From an Nvj1-centric perspective the assembly of the NV junction is elegant in its simplicity. The membrane-spanning domain of Nvj1 is sufficient and necessary to insert the protein into the ER membrane (16). The C-terminus is exposed to the cytoplasm where the last 20–40 residues bind to Vac8 in the vacuole membrane (8). The N-terminus is exposed to the ER lumen where it somehow recognizes and binds specifically to the inner nuclear membrane. These three membrane-associating elements function hierarchically to direct the protein to increasingly specific locations and, ultimately, to position the outer nuclear membrane between the vacuole membrane and the inner nuclear membrane at discrete structures called NV junctions.
The clustering of Nvj1 into small patches on the surface of the nuclear envelope is the final step in the assembly of NV junctions. Clustering and expansion of NV junctions presumably occurs spontaneously after Nvj1 is targeted to the nuclear envelope. In principle, after the first Nvj1-Vac8 complex forms between the two organelles, additional complexes will become trapped in the nascent junction as the two binding partners diffuse to the site and arrange themselves in phalanx-like arrays. In vac8-Δ cells, Nvj1 localizes around the entire perinuclear ER and remains excluded from the cortical ER. During nutrient depletion, Nvj1 levels rise and an increasing fraction of the cellular pool of Vac8 is recruited into the expanding NV junctions (9). In fact, Nvj1 can be ectopically overexpressed to levels that phenocopy the vacuole fragmentation and vacuole inheritance defects of vac8-Δ cells (17). Under these conditions, Nvj1 levels apparently exceed the available pool of Vac8 and the remainder is left free to spread over the surface of the perinuclear ER. The point here is that Nvj1 is efficiently retained in the perinuclear ER under a variety of conditions that might be expected to promote its leakage into the cortical ER. We only observe Nvj1 in the cortical ER when the membrane anchor is deleted, mutated, or blocked (16; this study).
The major conclusion of the present study is that the N-terminus contains a membrane anchor that bridges the perinuclear lumen and associates tightly with the inner nuclear membrane. A strong prediction of this hypothesis, that the width of the perinuclear lumen underlying NV junctions should be fixed by the physical dimensions of the luminal domain of Nvj1, was verified by quantitative 3D cryoelectron tomography (Fig. 3). Here, the separation between the inner and outer nuclear membranes in NV junctions is a consistent ~9 nm compared to ~18 nm in bulk nuclear envelope. Also, the width of the perinuclear lumen in bulk nuclear envelope varies much more than in NV junctions, where the width is strikingly consistent. By analogy, KASH domain proteins clamp the width of the perinuclear ER by bridging the lumen and binding to SUN domain proteins in the inner nuclear membrane (12, 38). RNAi depletion of both Sun1 and 2 proteins in mammalian tissue culture cells caused the dilation of the nuclear envelope from a relatively constant ~50 nm to a variable ~100 nm or more (12).
The N-terminal sequence of Nvj1 bears no similarity to KASH domain proteins. Instead, the sequence is very similar to N-terminal outer mitochondrial membrane anchor sequences, which are known to insert post-translationally into the outer mitochondrial membrane. Both mitochondrial membrane anchors and the Nvj1 anchor are composed of a core sequence of hydrophobic amino acids bracketed by basic residues. The number and distribution of basic charges within signal sequences (39) and membrane-spanning domains (40) play decisive roles in determining the orientation and of proteins in membranes and promote the stability of their interactions with the lipid bilayer.
Our analysis of the Nvj1 anchor sequence is most consistent with the hypothesis that it inserts spontaneously into the inner nuclear membrane. The strongest argument in favor of this is its similarity to mitochondrial outer membrane anchor sequences. The S. castelii sequence is informative in this regard because it differs slightly from the S. cerevisae sequence in both the distribution of basic residues and the hydrophobicity of its core. When heterologously expressed in S. cerevisae the S. castellii sequence did not direct efficient perinuclear localization. Increasing either the hydrophobicity or adding an extra basic charge converted the S. castellii sequence to an efficient inner nuclear membrane anchor in S. cerevisiae (Fig. 6B). The finding that the S. cerevisae, protein, but not the K. lactis protein localized strictly to the perinuclear ER in K. lactis cells suggests that the S. cerevisae membrane anchor functions in a host that is naïve to such an anchor. We think this is most consistent with the model that the membrane anchor associates with the lipid bilayer of the inner nuclear membrane, and not with an inner nuclear membrane protein, which, if it does exist, does not bind to the native ortholog, which seems unlikely. These experiments, together with those shown in Figs. 1 and and2,2, argue that the function of the anchor is determined by biophysical properties that to a first approximation we expect to interact directly with lipid bilayers. However, we cannot formally rule out the possibility that the anchor binds to an inner nuclear membrane protein by analogy to KASH-SUN protein complexes. Alternatively, other proteins may guide the anchor to the inner nuclear membrane or facilitate its insertion in the membrane. If the anchor inserts spontaneously into the inner nuclear membrane, then we will at some point have to explain why it does not insert into the opposite membrane in extra-perinuclear ER cisternae. These same unresolved issues apply to the factors that determine the strict membrane-specificity of outer mitochondrial anchors. One possibility is that these anchors have a propensity to integrate into the distinctive lipid compositions of their cognate membranes. In this regard, we previously altered the sterol, sphingolipid, and very long chain fatty acid composition of the cell without affecting the perinuclear localization of Nvj1 (10, 15).
We considered that Nvj1 might be sterically excluded from the tubular membranes of the cortical ER in favor of the sheet-like perinuclear ER. However, Nvj1 continued to localize exclusively to the perinuclear ER in cells lacking the reticulons Rtn1 and Rtn2. The cortical ER of reticulon mutants assumes a sheet-like structure that is similar to the outer nuclear membrane (35, 36). We conclude that the luminal N-terminus cooperates with the membrane-spanning domain to actively retain the protein in the nuclear envelope. We can rule out a contribution by the cytoplasmic domain, which includes the Vac8-binding motif, since reporters lacking the entire C-terminus up to the membrane-spanning domain continue to localize exclusively to the perinuclear ER (16).
NVJ1 has evolved extremely rapidly. In Fig 5E the domain structure of representative orthologous Nvj1 proteins are organized according to the currently accepted phylogenetic relationships of their host species. The striking feature of the evolution of NVJ1 orthologs, in addition to their rapid sequence divergence (Fig. 5A), is the modular fashion in which the orthologs acquired, and sometimes lost, functional domains. Species closest to S. cerevisae contain all four conserved motifs: the N-terminal hydrophobic core sequence, membrane-spanning domain, Osh1-binding motif, and the C-terminal Vac8-binding motif. The A. gossypii ortholog is the least well-conserved sequence we could identify with any certainty, and then only by virtue of its syntenic position in the genome. It contains an Osh1-binding motif downstream of a putative membrane-spanning domain, but lacks both a recognizable Vac8-binding motif and an N-terminal membrane anchor sequence (Fig. 5E). Other than the Osh1-binding motif this sequence is almost completely dissimilar to S. cerevisiae Nvj1. We identified a putative Osh1-binding motif in a well-conserved Aspergillus gene whose lineage can be traced to both plants and animals (Fig. 5B). But this weak similarity is poor support for the hypothesis that it an ortholog of Nvj1. The C. glabrata gene appears to have lost the Osh1-binding motif that is present in ancestral K. lactis and A. gossypii genes. Based on evolutionary and functional evidence, we hypothesize that the progenitor Nvj1 was a membrane protein that sequestered Osh1. The later addition of a Vac8-binding sequence positioned the Osh1-binding motif at ER-vacuole junctions. The subsequent evolution of an N-terminal inner nuclear membrane anchor led to the subcompartmentalization of the protein to the perinuclear ER and the innovation of the NV junction. Though other models could explain the presence or absence of various domains within the fungal NVJ1 orthologs, the stepwise evolution of S. cerevisae Nvj1 from an ancestral Osh1-binding progenitor protein is the most parsimonious explanation that is consistent with both the comparative and functional data.
The membrane anchor sequence is clearly absent from A. gossypii and K. lactis proteins, and appears to have “weakened” in S. castellii and C. glabrata, at least in terms of its heterologous function in S. cerevisiae. Preliminary studies indicate that the S. castellii sequence localizes exclusively to the perinuclear lumen when expressed as an EGFP reporter in vac8-ΔS. castellii (data not shown). Thus, this domain may function efficiently as an anchor in its native host. Its mislocalization in S. cerevisiae likely reflects species-specific differences in the lipid and protein composition of the ER network that influence the interaction of the anchor with the inner nuclear membrane. The weak S. castellii membrane anchor provided us an excellent opportunity to test the hypothesis that both basic charges and a sufficiently hydrophobic core sequence contribute to the anchor’s function in S. cerevisae. The K. lactis protein presents a different story, as it localized to both the cortical and perinuclear ER when expressed as an EGFP reporter in vac8-Δ K. lactis cells (data not shown), and, therefore, likely is not excluded from the cortical ER in its native host (data not shown).
In conclusion, the N-terminus of Nvj1 contains a sequence that is necessary for the close opposition of the inner and outer nuclear membranes in NV junctions. The physical properties of the sequence, and its striking similarity to outer mitochondrial membrane anchors, support the hypothesis that it bridges the perinuclear lumen and inserts into the inner nuclear membrane. Based on what we currently know, physically linking the inner and outer nuclear membranes should be a prerequisite to PMN, during which both nuclear membranes and a portion of the underlying nucleoplasm are pinched from the nucleus and degraded in the vacuole lumen. Although we cannot rule out the possibility that the membrane anchor associates with a resident inner nuclear membrane protein, the loose sequence requirements o f the anchor, and its resemblance to outer mitochondrial membrane anchors, argue that it spontaneously inserts into the luminal face of the inner nuclear membrane.
It is known that plastic embedded sections are altered when viewed using an electron beam (43, 44). By cooling the sample to cryo-temperatures or using low-dose imaging techniques the shrinkage can be reduced but not eliminated (45). For tomography studies, the plastic embedded sample is pre-radiated, or pre-shrunk, to stabilize the sample area before acquiring the tilt series.
As shown in the left panel, sample shrinkage is not always uniform. In plastic embedded sections of yeast we noticed that the vacuole shrinks more than the nucleus with electron exposure (Fig. S1A & C). Tomography of a plastic embedded nucleus vacuole junction is shown in Fig. S1A, the nucleus is denoted with N and vacuole with V. When we flip the reconstruction in Z (view the reconstruction in cross section), we notice that the vacuole is reduced more in the Z direction than the nucleus (denoted by the black arrowhead). This non-uniform shrinkage makes is difficult to measure the distances of the nuclear membranes and within the junction.
To alleviate the effects of sample shrinkage we used tomography of frozen hydrated sections. Although this technique is not complete void of its own artifacts, vitreous water does not shrink with electron exposure (Fig. S1B & D). In the cross section of the vitreous section reconstruction (D), the membranes are clearly visible and are not altered after electron exposure.
We are grateful to Eileen O’Toole and Mark Ladinsky for valuable technical assistance, and members of the Goldfarb lab for helpful discussions. We thank Marita Cohn for advise on S. castelli, Nancy Da Silva for the K. lactis strain and Xin Jie Chen for the K. lactis expression vectors. Finally we would like to thank Stephen White for help with hydrophobicity calculations. This study was supported by NIH grants R01GM67838 (D.S.G.) and RR00592 (Andreas Hoenger, Boulder, CO), and NSF grant MCB-072064 (D.S.G.).