COPII Subunit Interactions in the Assembly of the Vesicle Coat*

In vitro analysis of COPII vesicle formation in the yeast Saccharomyces cerevisiae has demonstrated the requirement for three cytosolic factors: Sec31p-Sec13p, Sec23p-Sec24p, and Sar1p. Convergent evidence suggests that the peripheral endoplasmic reticulum (ER) membrane protein Sec16p also represents an important component of the vesicle assembly apparatus: SEC16 interacts genetically with all five COPII genes; Sec16p binds to Sec23p and Sec24p, is found on ER-derived transport vesicles, and is required in vitro for the efficient release of ER-derived vesicle cargo. In this report, we demonstrate an important functional interaction between Sec16p and Sec31p. First, we map onto Sec31p binding regions for Sec16p, Sec23p, Sec24p, and Sec13p. Second, we show that a truncation mutant of Sec31p specifically defective for Sec16p binding is unable to complement asec31Δ mutant and cannot rescue the secretion defect of a temperature-sensitive sec31 mutant at nonpermissive temperatures. We propose that Sec16p organizes the assembly of a coat that is stabilized both by the interaction of Sec31p with Sec23p and Sec24p and by the interaction of these three components with Sec16p.

In eukaryotic cells, proteins in the secretory pathway are transported from the endoplasmic reticulum (ER) 1 to the Golgi apparatus via membrane-bounded vesicles (1) that are formed by the recruitment and assembly of cytosolic coat components upon the ER membrane (1)(2)(3). In the yeast Saccharomyces cerevisiae, three cytoplasmic factors, collectively termed COPII, have been shown to be required for vesicle formation: the Sec31p-Sec13p protein complex, the Sec23p-Sec24p protein complex, and the small GTP-binding protein Sar1p (4,5). The addition of these three factors to urea-washed ER membranes stimulates the formation of coated, fusion-competent vesicles (4) .
Although in vitro analysis has defined soluble components required for vesicle assembly, membrane-associated factors have remained less accessible to biochemical study. Thus, the mechanism of COPII recruitment and assembly onto the membrane is still unknown. However, converging evidence suggests that Sec16p, an essential 240-kDa multidomain protein, may be involved intimately in this process. SEC16 is required for transport vesicle formation (6,7) and exhibits genetic interactions with all five COPII genes (7)(8)(9)(10). Sec16p is tightly associated with the periphery of the ER and is also found on ER-derived transport vesicles (6); Sec16p cannot be extracted from membranes by urea (6). Membranes prepared from sec16 mutant strains exhibit a marked deficit in the release of vesicle cargo molecules (11). Finally, Sec16p directly binds the COPII subunits Sec23p and Sec24p (6,12). We now report that Sec16p also binds the COPII subunit Sec31p; we show this interaction is required for ER to Golgi transport.
Two-hybrid Analysis-A two-hybrid screen was performed in the strain CKY554, which is the indicator strain L40 (16) with the plasmid pDS99, carrying amino acids 447-1235 of Sec16p fused to LexA in the pBTM116 vector (19). CKY554 was transformed with an activation domain fusion library (generously provided by M. White) in the pGADGH vector (20). Interactions were tested as described (18,21). Sec31p truncation constructs in pGADGH were subsequently constructed using the cloned SEC31 locus (10), generously provided by R. Schekman. LexA fusion constructs were made in pBTM116 and represent the entire coding sequence of yeast Sec13p (pDS138) (15) and human Sec13Rp (pDS168) (22,23) and amino acids 666 -926 of Sec24p (pDS272) (8) and amino acids 447-1043 of Sec16p (pDS116) (6). The entire coding sequence of yeast Sec23p (14) was inserted into pGilda to generate pDS72. pGilda represents a derivative of the pEG202 lexA fusion vector (18) that retains the multiple cloning sites of pEG202 but that utilizes the GAL1 promoter instead of the ADH1 promoter; the vector backbone of pGilda is from pRS313 (24). The strain CKY556, which is EGY40 (17) transformed with the indicator plasmid pSH18-24 (18), was co-transformed with LexA and activation domain fusion plasmids. Transformants containing a pBTM116-derived LexA fusion plasmid were grown to exponential phase in selective medium containing glucose. Transformants containing a pGilda-derived LexA fusion plasmid were grown to exponential phase in selective medium containing 2% raffinose and then galactose was added to 2%, and growth continued for another 4 h. At least three independent transformants were assayed for ␤-galactosidase activity (13). The mean activity of the transformants is given and expressed in Miller units (1000 ϫ A 420 /(reaction time ϫ A 600 units assayed)) (25). All values above background were within 35% of the mean value.
In Vitro Binding Studies-Recombinant Sec23p and Sec24p were were expressed as GST fusion proteins in Escherichia coli, purified by affinity chromotography, and then released by thrombin cleavage of the GST moiety (12). Recombinant Sec13p (kindly provided by K. Saxena and E. Neer) was purified as a His 6 fusion protein from E. coli (26). The HA epitope-tagged Sec16 protein was prepared by insertion of a fragment encoding amino acids 447-1043 of SEC16 into the pGAL10-HA expression vector pRH165 2 to generate pDS216. This plasmid was transformed into the S. cerevisiae strain CKY557 (MAT␣ ura3-52 trp1::hisG GAL ϩ ). Cells were grown to exponential phase in selective medium containing 2% raffinose and then supplemented with 2% galactose for 4 -6 h to induce expression of the epitope-tagged protein.
DNA regions encoding the entire Sec31p protein, a fragment of Sec31p lacking the initial 490 amino acids, or a fragment of Sec31p lacking the final 98 amino acids were fused to the 3Ј end of the GST coding sequence expressed from pGAL1 promoter in pPE127, a vector identical to pRD56 but in a different reading frame (6). Clarified cytosolic extracts were prepared as described above, except that the concentration of NaCl used was 0.6 M; this buffer is referred to as LBB-600. Glutathione-Sepharose 4B beads (Pharmacia Biotech Inc.) were incubated with the extracts for 30 min at 25°C and then washed three times with LBB-600. Beads prepared in this fashion were decorated with 2-5 pmol of the fusion protein. For the binding reactions involving Sec23p, Sec24p, and Sec13p, the decorated beads were washed twice with binding buffer (25 mM K-Hepes, pH 6.8, 0.1% Triton X-100, 1 mM MgCl 2 , 0.25 mg/ml bovine serum albumin). The beads were then resuspended in 45 l of salt-supplemented binding buffer, and 5 l of the relevant recombinant protein was then added. Binding of Sec23p (2 pmol) and Sec24p (1 pmol) was carried out in 50 mM NaCl, whereas binding of Sec13p (3 pmol) was carried out in 150 mM NaCl. After incubation for 1 h at 25°C, the beads were washed two or three times with binding buffer (no additional salt) and then resuspended in extract sample buffer (12). For the reactions involving Sec16p, the beads were washed twice with LBB-100 and then incubated with a yeast cytosolic extract prepared as described. These binding reactions were carried out in a volume of 100 l and utilized extract containing 126 g of total protein; NaCl was added to a final concentration of 0.2 M. Following a 1-h incubation at 25°C, the beads were washed twice with LBB-100 and then resuspended in extract sample buffer. All proteins were subjected to SDS-PAGE and Western blotting.

RESULTS AND DISCUSSION
Portions of the SEC16 coding sequence were surveyed for regions that would not by themselves act as transciptional activators when fused to a DNA-binding domain and would therefore be suitable for two-hybrid analysis. pDS99, representing the coding sequence for amino acids 447-1235 of SEC16 inserted into the pBTM116 LexA fusion vector (19), was one of the constructs that fulfilled this criterion. This fragment of Sec16p included the region known to bind Sec24p (Sec16p amino acids 565-1235) and was different from the region known to bind Sec23p (Sec16p amino acids 1638 -2194) (6). The L40 reporter strain (16) was transformed with both pDS99 and a S. cerevisiae cDNA library constructed in the activation domain fusion vector pGADGH (20). Library plasmids were recovered from strains positive for expression of both the HIS3 and lacZ reporter genes. A screen of 8 ϫ 10 5 S. cerevisiae cDNA clones yielded seven positives whose activation of lacZ reporter expression depended upon the presence of the LexA-Sec16p fusion protein. Six of the positive clones contained overlapping cDNA segments derived from the 3Ј region of the SEC31 gene (10). The smallest of these clones, 2a8, encodes a peptide of 127 amino acids, representing the extreme C terminus of the 1273amino acid Sec31p molecule.
In an effort to define the functional domains of SEC31, a series of SEC31 deletions were constructed in pGADGH and evaluated by two-hybrid analysis against a series of potential interactors constitutively expressed as LexA fusion proteins in the pBTM116 vector (Fig. 1). From this study, the Sec13pbinding region mapped to the N-terminal third of the Sec31p protein, a region that contains six WD-40 repeats (28,29). Constructs expressing at least the first 490 amino acids of Sec31p, such as pDS131 (Fig. 1), interacted strongly with Sec13p, whereas constructs lacking this region failed to interact. Because Sec13p itself consists almost entirely of WD-40 repeats (26), the interaction of Sec13p with the N-terminal region of Sec31p indicates that these regions of WD-40 repeats associate in a homotypic fashion.
We also examined the interactions of the human Sec13p homolog, Sec13Rp (22,23). Human Sec13Rp exhibited a twohybrid interaction profile identical to that of yeast Sec13p, interacting specifically with pDS131. These results emphasize the degree of conservation between yeast and mammalian COPII structures (23, 30 -32) and strongly imply the existence of a mammalian Sec31 protein.
We next asked whether Sec31p could interact with either Sec23p or Sec24p. Utilizing a domain of Sec24p that does not interact with Sec23p (12), we determined that Sec24p interacts with the central region of Sec31p (pDS134), a region that does not interact with either Sec13p or Sec16p. The evaluation of Sec23p binding to Sec31p required the use of an inducible LexA-Sec23p fusion protein, because the constitutive overexpression of fusions to SEC23 fusion was lethal (data not shown). We constructed the vector pGilda, which allowed for the galactose-inducible expression of toxic LexA-Sec23p fusion proteins. Fusion proteins expressed from pGilda required different growth conditions and were present at higher levels than those expressed from pBTM116 but gave the internally consistent result that Sec23p interacted specifically with a central, 325-amino acid region of Sec31p (pDS135).
To confirm the interactions detected by two-hybrid analysis, we asked whether Sec16p, Sec23p, Sec24p, and Sec13p could bind to different GST-Sec31p fusion proteins isolated from yeast. This approach has been used previously to demonstrate the direct binding of Sec23 and Sec24p to different regions of Sec16p (12). We evaluated three different GST fusion proteins: full-length Sec31p, Sec31p lacking the N-terminal 490 amino acids (Sec31⌬Np), and Sec31p lacking the C-terminal 98 amino acids (Sec31⌬Cp). For the binding experiments, purified recombinant Sec23p, Sec24p, and Sec13p were used. The source of Sec16p for this experiment was clarified extracts of yeast overexpressing the putative Sec31p-interacting region of Sec16p (amino acids 447-1043) tagged with a hemagglutinin (HA) epitope. The results from these binding experiments were in complete agreement with the two-hybrid data (Fig. 2). Fulllength Sec31p was able to bind Sec13p, Sec23p, Sec24p, and Sec16p. However, Sec31⌬Np was specifically defective for Sec13p binding, whereas Sec31⌬Cp was specifically defective for Sec16p binding. Sec23p and Sec24p bound to both of the truncated proteins, but not to GST alone; these binding reactions were performed in buffer containing 50 mM NaCl because very little binding of Sec24p and only about half-maximal binding of Sec23p were observed under the higher salt concentrations (150 -200 mM NaCl) used for the Sec16p and Sec13p binding experiments (data not shown).
The identification of Sec31⌬Cp, which is specifically defec-tive for Sec16p binding, allowed us to investigate the in vivo significance of this interaction (Fig. 3). First, we asked whether sec31-⌬C could functionally substitute for wild-type SEC31. A sec31-null strain bearing wild-type SEC31 on a URA3-marked plasmid was transformed with LEU2-marked plasmids carrying either SEC31 or sec31-⌬C. Transformants were grown with selection for the LEU2-marked plasmid and then plated on medium containing 5-fluoroorotic acid (5-FOA). Only yeast capable of growing in the absence of the URA3-marked plasmid would be expected to grow under these conditions. The strains carrying the plasmid with sec31-⌬C did not produce segregants that could grow on 5-FOA (Fig. 3A), showing that the truncated protein lacks an essential function of Sec31p. As a control for protein expression levels, the wild-type yeast strain CKY8 (6) was transformed with either a CEN plasmid carrying sec31-⌬C or with vector alone. Extracts from both strains were examined by immunoblotting using anti-Sec31p antibody (Fig. 3D) (5). Sec31⌬Cp was present in equivalent amounts to the endogenous Sec31p, indicating significant production of the truncated protein.
To address more directly the role of the C-terminal region of Sec31p in secretion, we utilized a temperature-sensitive allele of SEC31, designated sec31-2, which was isolated in a screen for new mutants defective for ER to Golgi transport. By testing whether Sec31⌬Cp could rescue the secretion defect of sec31-2

FIG. 2. Binding of COPII subunits to Sec31p in vitro.
A, GST fusion proteins representing full-length Sec31 (1273 amino acids), a N-terminal truncation lacking the first 490 amino acids (Sec31⌬N), a C-terminal truncation lacking the final 98 amino acids (Sec31⌬C), or GST alone were immobilized on glutathione-agarose and incubated with recombinant Sec23p, Sec24p, Sec13p, or cytosol from yeast expressing amino acids 447-1043 of Sec16p tagged with the HA epitope. Samples representing the total protein added and the fraction bound were resolved by SDS-PAGE and Western blotting. The loading ratio of total to bound is 1:1 for the three recombinant proteins, and 1:10 for the Sec16p extract. The recombinant Sec13p migrates more slowly than endogenous Sec13p, and the doublet in the Sec13p lanes indicates that both proteins co-purify with Sec31p and Sec31⌬Cp. B, the affinitypurified GST-Sec31 fusion proteins used in the binding reactions were separated by SDS-PAGE on a 10% gel and stained with Coomassie Brilliant Blue.
FIG. 3. Sec16p-binding region of Sec31p is essential and required for ER to Golgi transport. Plasmids expressing either Sec31p (CEN plasmid) or Sec31p⌬C (both CEN and 2 plasmids) were transformed along with a vector-only control into either RSY1109, which carries a chromosomal deletion of the entire coding sequence of SEC31 covered by SEC31 on a URA3-containing plasmid, or CKY555, which carries the temperature-sensitive allele sec31-2. A, RSY1109 transformants were plated on medium containing 0.1% 5-FOA to assay for complentation of the null allele. B, CKY555 transformants were plated at both permissive (24°C) and nonpermissive (36°C) temperatures to assay complementation of the sec31-2 allele. C, the transport of the vacuolar protein CPY in the CKY555 transformants at nonpermissive temperatures was examined by pulse-chase analysis. Exit of CPY from the ER is observed only in the strain expressing the full-length wildtype Sec31p protein. observed at nonpermissive temperatures, the ability of Sec31⌬Cp to fulfill the function of Sec31p in ER to Golgi transport could be assessed. The sec31-2 mutant was transformed with plasmids encoding either Sec31p or Sec31⌬Cp; growth of the transformants at the nonpermissive temperature of 36°C was then evaluated. Although mutants transformed with the SEC31 plasmid grew at 36°C, mutants transformed with the sec31-⌬C plasmid remained temperature-sensitive for growth (Fig. 3B); these results were observed in mutants transformed with either a CEN-based or a 2-based sec31-⌬C plasmid. The kinetics of ER to Golgi transport of the marker cargo protein CPY was followed by pulse-chase analysis of the transformants at 36°C. The sec31-⌬C plasmid did not rescue the CPY transport defect (Fig. 3D). Because the binding studies showed that the only apparent defect of Sec31⌬Cp is in binding to Sec16p, the transport defect exhibited by the truncated allele of SEC31 argues that the binding of Sec31p to Sec16p is required for ER to Golgi transport. However, we cannot eliminate the possibility that the C-terminal region of Sec31p performs an additional function that has not yet been defined that is necessary for secretion.
Reconstitution studies in both yeast and mammalian cells demonstrate that vesicle coat formation can be stimulated by the addition of a defined set of cytosolic factors to washed donor membranes (4,5,33). For transport between Golgi cisternae in mammalian cells, these factors are the small GTP-binding protein ARF and the coatomer complex, consisting of seven subunits that coassemble in the cytosol and bind en bloc to the donor membrane (34). For transport between yeast ER and Golgi, two different cytosolic protein complexes in addition to the small GTP-binding protein Sar1p are needed to form the COPII vesicle coat (3,14,15). The interaction that we detected between purified components of the Sec31p-Sec13p complex and the Sec23p-Sec24p complex suggested that these two complexes might preassemble in the cytosol. To examine this possibility, we expressed a GST-Sec31p fusion protein in yeast and asked whether Sec23p or Sec24p could be found associated with this fusion protein in a cytosolic extract prepared under conditions of the in vitro transport assay (4,15). We were unable to detect either of these proteins in the bound fraction. This observation was consistent with our measurements of the stability of the interactions between isolated proteins: binding of GST-Sec31p to recombinant Sec23p and Sec24p was demonstrated at 50 mM NaCl but (as noted above) was significantly weaker at 150 mM NaCl, a salt concentration equivalent to that used for the in vitro assay (data not shown). Thus, it is likely that the physiological association between the two COPII complexes requires the context of the ER membrane.
Given that SEC16 interacts genetically with all five COPII genes and encodes a peripheral ER membrane protein that is present on ER-derived transport vesicles, required for vesicle formation, and binds directly to Sec23p, Sec24p, and Sec31p, we propose that Sec16p functions as a foundation for the construction of the COPII coat from soluble protein complexes (Fig. 4). Moreover, the demonstration that Sec31p binds directly to both Sec23p and Sec24p suggests that the assembling COPII subunits are stabilized not only by interactions with Sec16p but also by interactions with each other.