Calreticulin-dependent recycling in the early secretory pathway mediates optimal peptide loading of MHC class I molecules

Abstract

Calreticulin is a lectin chaperone of the endoplasmic reticulum (ER). In calreticulin-deficient cells, major histocompatibility complex (MHC) class I molecules travel to the cell surface in association with a sub-optimal peptide load. Here, we show that calreticulin exits the ER to accumulate in the ER–Golgi intermediate compartment (ERGIC) and the cis-Golgi, together with sub-optimally loaded class I molecules. Calreticulin that lacks its C-terminal KDEL retrieval sequence assembles with the peptide-loading complex but neither retrieves sub-optimally loaded class I molecules from the cis-Golgi to the ER, nor supports optimal peptide loading. Our study, to the best of our knowledge, demonstrates for the first time a functional role of intracellular transport in the optimal loading of MHC class I molecules with antigenic peptide.

Introduction

Major histocompatibility complex (MHC) class I molecules are used by the immune system to survey the intracellular proteome of all nucleated cells for the presence of viruses, parasites, and tumor-specific antigens. They bind peptide fragments of eight to ten amino acids in the lumen of the endoplasmic reticulum (ER) and transport them to the plasma membrane, where they are presented to cytotoxic T lymphocytes (CTLs). Specific high-affinity peptides that fulfil the canonical binding requirements for length and sequence increase the conformational stability of class I molecules, leading to long-lived peptide-class I complexes in vivo (Elliott et al, 1991), which are generally required for efficient CTL responses.

The binding of high-affinity peptides to class I molecules within the ER is an iterative process that depends on an assembly of proteins called the peptide-loading complex (PLC) (Sadasivan et al, 1996). This comprises the transporter associated with antigen processing (TAP) heterodimer that translocates peptides from the cytosol to the ER lumen in an ATP-dependant manner; the soluble lectin chaperone calreticulin (CRT); the protein disulphide isomerases ERp57 and PDI (Gao et al, 2002; Garbi et al, 2006; Park et al, 2006); and the MHC class I-specific accessory protein, tapasin (Sadasivan et al, 1996). MHC class I molecules that have not yet bound optimal high-affinity peptide (and which are called here sub-optimally loaded or peptide-receptive) cycle between the ER and the cis side of the Golgi apparatus; they can proceed to the cell surface only when they become bound to a high-affinity peptide (Garstka et al, 2007). Although the molecular mechanism of the regulation of class I trafficking and quality control is unclear, we and others have proposed that the PLC, or some of its components, may accompany sub-optimally loaded class I molecules to the ER–Golgi intermediate compartment (ERGIC) or the cis-Golgi and help retrieve it to the ER (Wright et al, 2004). In this scenario, the retrieval of the PLC or of its individual members may be mediated by the putative ER retrieval sequences found in calreticulin (KDEL in the single-letter amino acid code; Opas et al, 1991), tapasin (KKXX; Ortmann et al, 1997), and possibly ERp57 (QEDL; Bennett et al, 1988; Mazzarella et al, 1994). Importantly, such signals are not apparent in MHC class I heavy chains themselves.

Calreticulin binds proteins that carry the glycan-processing intermediate Glc₁Man₇₋₉GlcNAc₂ (Ware et al, 1995; Spiro et al, 1996; Vassilakos et al, 1998). This glycan moiety is maintained on unfolded polypeptides in the ER by a glucosyltransferase that is specific for unfolded proteins, leading to the binding of the protein to calreticulin (and/or calnexin) until it has acquired its folded conformation (Sousa et al, 1992; Hammond et al, 1994; Peterson and Helenius, 1999; Ritter et al, 2005; Caramelo and Parodi, 2007). Direct protein–protein interactions also contribute to the chaperone activity of calreticulin and thus to MHC class I folding (Ireland et al, 2008).

The role of calreticulin in MHC class I antigen presentation has been studied in calreticulin-deficient cells derived from knockout mouse embryos. Interestingly, in such cells, MHC class I molecules are transported to the cell surface loaded with low-affinity peptides. As the PLC can form in the absence of calreticulin, this suggests a function for calreticulin in the quality control of class I–peptide complexes. In addition, calreticulin-deficient cells cannot efficiently retain sub-optimally loaded class I molecules inside the cell; instead, class I molecules move to the surface with accelerated kinetics (Gao et al, 2002). This has led us to propose that calreticulin may determine the localization of sub-optimally loaded class I molecules.

In this study, we investigate the role of calreticulin in the intracellular localization of sub-optimally loaded MHC class I molecules. We find that calreticulin does not prevent them from exiting the ER, but that it travels to the Golgi apparatus where it co-localizes with them and that its retrieval from there to the ER—by means of its own C-terminal KDEL retrieval sequence—is required for their retrieval to the ER and for efficient presentation of a model antigen. Remarkably, this function of calreticulin seems to be independent of its role in the PLC.

Results

Loss of the KDEL sequence leads to the secretion of calreticulin

To investigate the role of calreticulin in the intracellular localization of MHC class I molecules, we used the murine embryonic fibroblast cell line K42, which carries a homozygous deletion of calreticulin (Nakamura et al, 2001), to express two variants of calreticulin: a deletion of the C-terminal KDEL retrieval sequence (termed CRT-HAΔKDEL), and a mutation of the KDEL sequence to KDEV (CRT-HAKDEV) (Andres et al, 1990; Haugejorden et al, 1991). Both proteins contained an internal hemagglutinin (HA) tag for detection at the very C terminus of CRT-HAΔKDEL and just before the KDEL sequence in CRT-HAKDEV. As controls, we used the constructs CRT-KDEL (wild-type calreticulin) and CRT-HAKDEL (wild-type calreticulin with an HA tag inserted before the KDEL sequence; Figure 1A). Transfected cells were cloned to near-homogeneous expression of the co-transduced truncated nerve growth factor receptor, resulting in a uniform expression of calreticulin constructs (data not shown), and were then assessed for expression levels of calreticulin constructs by western blotting (Figure 1B). None of the constructs was found to be produced at levels significantly above that of endogenous calreticulin in the syngeneic wild-type K41. Importantly, the steady-state levels of CRT-HAKDEV and CRT-HAΔKDEL proteins were only 33 and 19% of CRT-HAKDEL, respectively, suggesting that they may be secreted from the cells.

Figure 1.

The calreticulin retention mutant HAΔKDEL is deficient in antigen presentation. (A) Rat calreticulin (CRT) constructs used in this study. Hemagglutinin (HA) tags were inserted as indicated. (B) Expression of the calreticulin constructs in K42 cells. Immunoblot of cell lysates probed with anti-calreticulin (left panel) and anti-HA (right panel) antibodies. Glycerinaldehyde phosphate dehydrogenase (GAPDH) was used as a loading control. In this and the following panels, KDEL, HAKDEL, HAΔKDEL, etc. stand for K42 cells expressing the respective calreticulin construct. The bar chart shows the ratio of calreticulin to GAPDH signals, normalized to unity for HAKDEL independently in each panel, to provide a comparison between the values in the two panels. (C) The HAΔKDEL construct is secreted into the supernatant medium. Immunoblot from cell lysate (lys) and cell supernatant (sn) collected at the start of the experiment (day 0), after one (day 1), and after two days (day 2). Calreticulin and constructs were detected by western blotting with anti-calreticulin (for K41 and KDEL) or anti-HA antibodies. Numbers on the right show the ratio of supernatant to lysate signals on day 2. Equivalent amounts of lysate and supernatant were applied in all samples (see the Materials and methods section), but as the supernatants were analysed on different membranes, direct comparison of the band strengths between blots is not possible. (D) FACS analysis of the surface presentation of SIINFEKL peptide on H-2K^b (detected using MAb 25D1.16) at different GFP–ubiquitin–SIINFEKL expression levels (x-axis). Eight sectors used for quantification in (E) are separated by vertical lines. (E) Analysis of data from 1D. For each of the eight GFP intensity sectors of each sample in 1D, the mean intensity of MAb 25.D1.16 staining was calculated and plotted against the GFP mean fluorescence. Controls are K42 cells transfected with GFP only (no peptide) and K42 stained without first antibody.

To estimate the loss of intracellular retention of the calreticulin variants, we measured their accumulation in the culture supernatant (Figure 1C). Although CRT-KDEL and CRT-HAKDEL were efficiently retained inside the cells, CRT-HAΔKDEL was secreted, as observed previously by Sönnichsen et al (1994). In contrast, CRT-HAKDEV was only secreted to a small extent, suggesting that the C-terminal KDEV sequence retains part of the KDEL function. This was surprising because in other studies, a KDEL-to-KDEV change abolished the retention of some proteins (Tang et al, 1992). We conclude that the HA tag does not interfere with the retention of calreticulin, and that KDEL-mediated retrieval is required for keeping calreticulin inside the cell.

Loss of the KDEL sequence of calreticulin leads to impaired presentation of a model antigen

We next asked how the loss of the KDEL sequence of calreticulin would affect the loading of peptides onto class I molecules, and how this would compare with the lack of the entire calreticulin protein. To measure the presentation of an endogenous model peptide, we expressed a chimeric construct of the green fluorescent protein (GFP) and the H-2K^b-binding peptide, SIINFEKL, bridged by ubiquitin. When the construct is produced in cells, the SIINFEKL peptide is released by the ubiquitin C-terminal hydrolase. The peptide is then available for transport into the ER, loading onto class I molecules, and subsequent presentation at the cell surface; its original amount in the cell can be deduced from the intensity of GFP fluorescence (Neijssen et al, 2005). We detected the complex of SIINFEKL and H-2K^b at the cell surface through flow cytometry with the SIINFEKL-specific monoclonal antibody 25D1.16 (Porgador et al, 1997), and we simultaneously recorded GFP fluorescence in these cells. This enabled us to relate the amount of surface presentation of the SIINFEKL peptide to its intracellular concentration, and thus calculate the efficiency of its presentation through class I molecule (Figure 1D and E). Although in the group of cells with the lowest GFP expression, the signals for all calreticulin-expressing cells showed the same background (which is expected as the antibody only recognizes the introduced SIINFEKL peptide), we found that when the peptide was expressed, K41 cells were far more efficient than K42 at processing and presenting SIINFEKL, consistent with published data (Gao et al, 2002). CRT-HAKDEL restored the presentation of SIINFEKL to 60–80% of the wild type at different intracellular peptide levels, with the incomplete restoration seen perhaps due to the species difference between rat calreticulin and mouse cells, or the genetic variation between K41 and K42 cells outside the calreticulin locus. Importantly, in striking contrast to the full-length construct, CRT-HAΔKDEL did not increase the loading above that seen for K42. Thus, the KDEL Golgi-to-ER retrieval sequence of calreticulin is essential for optimal peptide loading over a range of endogenous peptide concentrations. Interestingly, the debilitating effect of the loss of calreticulin (or its KDEL tail) on peptide loading can be overcome by an increased antigen dose, as a two-fold increase in GFP–ubiquitin–peptide expression leads to a restoration of the surface levels of the peptide–class I complex. This may be the reason why some (presumably the most abundant) antigens are calreticulin-independent for their presentation to T cells (Gao et al, 2002).

Loss of the KDEL sequence of calreticulin does not impair the PLC

One possible reason why the KDEL sequence of calreticulin is necessary for efficient antigen presentation is that its absence may impair the incorporation of calreticulin into the PLC and/or the function of the PLC in general, which would in turn affect peptide loading and class I localization. To assess the composition of the PLC in the presence of different calreticulin variants, we lysed the cells in digitonin (which preserves the weak interactions in the loading complex (Diedrich et al, 2001)) and immunoprecipitated the PLC with an anti-TAP1 antibody. The co-precipitating MHC class I heavy chain, tapasin, ERp57, and calreticulin were then detected by western blotting. Figure 2 shows that CRT-HAΔKDEL was detected in the PLC at normal levels, showing that the KDEL sequence is not required for the incorporation of calreticulin into the PLC. When we investigated the association of ERp57, a direct binding partner of calreticulin, with the PLC, we observed variable levels in K42 cells, but never in K41 cells or in K42 transfected with calreticulin constructs (Figure 2A). This variation in K42 cells may be due to a loss of stability of the PLC when calreticulin is absent, such as a more rapid equilibrium of ERp57 association to and dissociation from tapasin in K42 cells. Indeed, when we inhibited disulphide exchange using the alkylating agent methyl methanethiosulfonate (MMTS) before lysis to stabilize the PLC, the majority of ERp57 was consistently detected as a disulphide-bonded conjugate with tapasin as reported previously (Peaper et al, 2005), both in K41 cells and in K42 cells expressing the calreticulin constructs (Figure 2B). Despite the difference in the steady-state levels of CRT-HAKDEL and CRT-HAΔKDEL (Figure 1B), calreticulin was precipitated in equal amounts from the PLC of cells expressing either of the constructs (Figure 2A). We conclude that the assembly of a functional PLC occurs even in the absence of the C-terminal KDEL sequence of calreticulin, and that presence of calreticulin in the PLC alone does not warrant efficient peptide loading of class I molecules.

Figure 2.

The calreticulin mutant HAΔKDEL is incorporated normally into the peptide-loading complex. (A) Immunoprecipitation of peptide-loading complexes from lysates of the indicated cell lines (left panel: anti-TAP1 antiserum, detection of the component proteins by immunoblotting). A sample of each cell lysate (corresponding to 10⁴ cells) was analysed in the same way (right panel). (B) ERp57 is present in the PLC of CRT-ΔKDEL containing cells as a disulphide-bonded conjugate with tapasin. Cells were pre-treated with the alkylating agent MMTS before lysis, and anti-TAP1 immunoprecipitates were separated by non-reducing (to detect the disulphide-bonded ERp57–tapasin conjugate) or reducing SDS–PAGE as indicated, followed by immunoblotting.

Calreticulin deficiency does not alter the packaging of class I into COPII vesicles

Calreticulin can retain associated proteins inside the cell (Molinari et al, 2004). We therefore hypothesized that the effect of calreticulin on class I trafficking and antigen presentation may, at least in part, be due to such a localizing effect of calreticulin on sub-optimally loaded class I, that is, calreticulin may redirect those class I molecules that are associated with it to the ER. Since sub-optimally loaded class I molecules cycle between ER and cis-Golgi (Garstka et al, 2007), we hypothesized that calreticulin may act either at the level of exit of class I molecules from the ER, impeding their packaging into COPII vesicles (which form at ER exit sites and transport proteins between the ER and the ERGIC (Lee and Miller, 2007)); or else, that it may act at the level of the Golgi, promoting the packaging of sub-optimally loaded class I molecules into retrograde COPI vesicles for retrieval to the ER. To distinguish between these possibilities, we asked whether packaging of endogenous H-2D^b and H-2K^b class I molecules into COPII vesicles was affected by the lack of calreticulin. Microsomal membranes were prepared from radiolabelled K42 CRT-KDEL and K42 cells, incubated with cytosol as a source of COPII components and with ATP and GTP to stimulate COPII vesicle formation, and COPII vesicles were isolated by differential centrifugation as described previously by Garstka et al (2007). The vesicles were lysed overnight in the presence or absence of specific peptide (FAPGNYPAL, which binds to both D^b and K^b molecules). In the absence of exogenously added peptide, peptide-receptive class I molecules dissociate into heavy chain and β₂m (Townsend et al, 1989), whereas when peptide is added, they are converted to the peptide-bound form. We then immunoprecipitated class I molecules with the conformation-specific monoclonal antibodies: B22.249 (for peptide-bound H2-D^b) and Y3 (for peptide-bound H2-K^b), treated the samples with endoglycosidase F₁ (EndoF₁) (which has the same substrate specificity as EndoH; (Trimble and Tarentino, 1991)), and detected class I molecules by SDS–PAGE and autoradiography (Figure 3A). Vesicle samples lysed in the presence of peptide gave the total amount of class I (peptide occupied plus peptide receptive), whereas vesicles lysed in the absence of peptide gave only those class I molecules that had been loaded with high-affinity peptide before vesicle budding (Garstka et al, 2007). To compare and average multiple experiments, we standardized all packaging efficiencies to those of the Na⁺/K⁺ ATPase, a protein of the plasma membrane that is exported well from the ER. The class I molecules from the vesicle fraction were indeed contained in COPII vesicles (and not in another kind of vesicle) because in the presence of the dominant inhibitor of COPII budding, Sar1 (T39N, Barlowe et al, 1994), much smaller amounts were immunoprecipitated.

Figure 3.

Calreticulin does not influence packaging of class I molecules into COPII vesicles. (A) COPII vesicles (lanes 1–7) were generated in an in vitro reaction from K42 CRT-KDEL (top three panels) or K42 cells (bottom three panels). Controls in this reaction were the omission of cytosol (lane 4) or ATP (lane 5), or addition of dominant-negative Sar1 (T39N; lanes 6 and 7). COPII vesicles or the corresponding donor microsome membranes (after the reaction; lane 8–11) were lysed with detergent (in the presence of 10 μM peptide as indicated in lanes 3, 7, and 9), and H-2D^b, H-2K^b, and Na⁺/K⁺ ATPase were sequentially immunoprecipitated from the lysates. Immunoprecipitates were treated with EndoF₁ (lanes 2–9), PNGase (lane 10), or no glycosidase as indicated. Lane 7 was moved to a different position of the gel to facilitate comparisons (and is flanked by white lines to indicate this fact). The numbers below the panels are quantifications of the EndoF₁-sensitive (pre-cis-Golgi) bands. The position of the class I molecules with undigested glycans on the gels (visible especially in the last lane in the donor microsomes) is indicated with an asterisk. One out of three independent experiments is shown here. (B) COPII packaging efficiency of H-2D^b and H-2K^b in K42 CRT-KDEL cells is the same for peptide-occupied class I (black bars) and total class I molecules (white bars). Evaluation of the three independent experiments from (A). Packaging efficiencies of the peptide-occupied (lane 2 over lane 8) and total (lane 3 over lane 9) class I molecules were normalized to the packaging efficiency of the Na⁺/K⁺ ATPase in same reactions (see the Materials and methods section for a detailed description). The error bars show the s.e.m. values. (C) COPII packaging of all forms of H-2D^b and H-2K^b is higher in K42 CRT-KDEL (black bars) than in K42 cells (white bars). Evaluation of the three independent experiments mentioned in (A). Packaging efficiencies of total class I molecules (lane 3 over lane 9) were normalized to the packaging efficiency of the Na⁺/K⁺ ATPase in same reactions, and are shown here compared between K42 CRT-KDEL and K42 (see the Materials and methods section for a detailed description). The error bars show the s.e.m. values.

For both H-2D^b and K^b, the rate of incorporation into the vesicles of the peptide-occupied forms was not significantly different from that of total class I molecules in K42 CRT-KDEL cells (Figure 3B). Thus, at the level of the ER, there is no distinction in the rate of COPII packaging of peptide-occupied and peptide-receptive forms of class I molecules. This agrees with our previous observations on HLA-A^*0201, HLA-B^*5101, and H-2 K^b (Garstka et al, 2007). As expected, in K42 cells, there was a much higher proportion (about 85%) of peptide-receptive class I molecules than in K42 CRT-KDEL (about 50%; in Figure 3A, the difference between band strengths in lanes 9 and 8 corresponds to the amount of peptide-receptive class I molecules). Nevertheless, these peptide-receptive molecules were not exported more efficiently in K42 cells, that is, in the absence of calreticulin; in contrast, in K42 cells, export of total D^b and total K^b from the ER was actually decreased (Figure 3C). Interestingly, the export of K^b molecules from the ER was significantly more efficient than that of D^b, mirroring the faster trafficking of H-2K^b in vivo. These findings demonstrate that calreticulin does not delay the export of peptide-receptive class I molecules from the ER.

Calreticulin exits the ER and is found accumulating in the cis-Golgi where it co-localizes with peptide-receptive class I molecules

We next investigated the alternative hypothesis that, in a manner analogous to the chaperone protein, BiP/GRP78 (Pelham, 1988; Schweizer et al, 1991; Hammond and Helenius, 1994), calreticulin may travel to the Golgi apparatus together with unfolded proteins and retrieve them back to the ER. In agreement with this proposition, we detected small amounts of endogenous calreticulin in COPII vesicles of K41 cells (Figure 4A). In contrast, using immunofluorescence microscopy, we were unable to detect post-ER localization of either endogenous calreticulin in K41 or of CRT-HAKDEL in K42 cells, perhaps because its retrieval is very efficient, resulting in a low steady-state concentration in the Golgi (Supplementary Figure S5). A GFP fusion of calreticulin (Snapp et al, 2006) that is easier to detect than the endogenous calreticulin, was accumulated in a post-ER compartment in COS-1 cells, and in K41 cells, an accumulation was visible, although weakly (Figure 4B). The detection of these weak accumulations in post-ER compartments was difficult because of the strong ER background. We reasoned that overexpression of sub-optimally loaded class I molecules may move a visible amount of endogenous calreticulin to a post-ER compartment, in analogy to the situation described by Hammond and Helenius (1994), in which overexpression of the vesicular stomatitis virus glycoprotein (VSV-G) led to a movement of BiP to the Golgi apparatus. This proposition was testable since we had found earlier that murine class I molecules, when overexpressed in primate fibroblast cells, are inefficiently loaded with peptide and do not appreciably reach the cell surface (Garstka et al, 2007); thus, overexpressed class I molecules in COS cells mostly cycle between ER and cis-Golgi. To see whether such class I overexpression would affect the localization of calreticulin, we transfected COS cells with a GFP fusion of H-2D^b and observed the localization of endogenous calreticulin. We found that in many of the cells that expressed large amounts of H-2D^b–GFP fusion protein, a post-ER accumulation of calreticulin was visible, which co-localized with the accumulated H-2D^b. In contrast, such accumulation of endogenous calreticulin was visible only in few of the cells that did not express H-2D^b. The calreticulin accumulation co-localized with the ERGIC and partially with the Golgi apparatus, suggesting that calreticulin was able to follow the class I molecules to ERGIC and cis-Golgi (Figure 4C and D). We conclude that the tagging of calreticulin with an HA or GFP tag did not significantly influence its intracellular steady-state localization in the ER, but that calreticulin can be partially moved to the cis-Golgi by an abundance of substrate. This confirms that even under wild-type conditions, a small amount of calreticulin may travel to the Golgi bound to (partially) unfolded proteins such as sub-optimally loaded class I molecules.

Figure 4a-c.

Calreticulin leaves the ER and co-localizes with class I molecules in the ERGIC and the cis-Golgi. (A) Calreticulin is packaged into COPII vesicles. An in vitro COPII vesicle formation experiment was carried out on K41 cells, and calreticulin was immunoprecipitated from lysates of vesicle fractions or donor membranes with PA3-900 antiserum. Controls are as described in Figure 3A. The numbers below the panels are the quantified band densities. (B) Calreticulin (CRT)–GFP is visible in a post-ER compartment in K41 and COS cells. Cells were transfected with CRT–GFP and stained for the cis-Golgi marker, giantin. Bars, 10 μm. (C) Calreticulin accumulates in COS cells that express H-2D^b–GFP (rows 1 and 2). Calreticulin and H-2D^b–GFP partially co-localize with the ERGIC marker, p58 (Rows 3–5), and with the cis-Golgi marker, GM130 (Row 6), in COS cells, but the bulk of co-localization is with the ER marker, PDI (rows 7 and 8). Row 2 shows an untransfected cell (arrow) that does not express D^b-GFP, in which calreticulin shows no accumulation. Row 5 shows another untransfected cell (arrow), in which calreticulin shows no colocalization with the ERGIC. Row 8 shows untransfected cells and ER stain (PDI) with near-complete colocalization. Bars, 10 μm.

Figure 4d-e.

(D) Calreticulin is shifted to the ERGIC/Golgi in COS cells that also accumulate H-2D^b. Quantification of the localization of calreticulin in the ERGIC/Golgi (white portion) or in the ER (black portion) of cells that accumulate (right bar) or that do not accumulate transfected H-2D^b–GFP (left bar). The error bars show the s.e.m. values for four independent experiments (n=191). (E) Calreticulin accumulates in COS cells that express a single chain fusion of β₂m to H-2D^b–GFP. K41 or COS cells were transfected with H-2D^b–GFP (heavy chain, top rows) or a single chain β₂m–H-2 D^b–GFP fusion and stained for calreticulin and the cis-Golgi marker, giantin. Bars, 10 μm.

To demonstrate that the re-localization of calreticulin to the Golgi apparatus in this experiment was indeed because of sub-optimally loaded D^b/β₂m dimers and not because of potential denatured free heavy chains of class I that exited the ER in complex with calreticulin, we performed two different experiments. First, we repeated the overexpression experiment with a fusion protein consisting of β₂m that was covalently connected to the amino terminus of D^b–GFP by a linker of glycines and serines (Tourdot et al, 2005). This constitutive single chain D^b/β₂m dimer led to calreticulin accumulation in post-ER compartments of COS cells in the same manner as the Db heavy chain alone (Figure 4E). Accumulations of calreticulin were only observed in COS cells and not in K41 cells that expressed heavy chain or single chain constructs, suggesting a different morphology of the COS cells or a requirement for class I overexpression (Figure 4E). Second, to show that sub-optimally loaded class I molecules were indeed found in the cis-Golgi, we transfected COS cells with H-2K^b–GFP and stained them with a K^b-binding peptide, SIINFEKL, that was modified with the TAMRA fluorescent dye (Supplementary Figure S1). We found K^b-specific fluorescence in the ER and in an accumulation that corresponded to the early secretory pathway, similar to previous observations (Day et al, 1995).

In TAP-deficient lymphocytes, the post-ER accumulation of class I molecules is more easily visible than in fibroblasts, as it is often spatially separated from the ER signal, and possibly because of higher class I expression levels (Garstka et al, 2007). We, therefore, stained TAP-deficient T2 lymphocytes for endogenous HLA-B^*5101 and calreticulin. Intriguingly, calreticulin showed a clear post-ER accumulation that co-localized in many cells with that of class I molecules, suggesting a functional post-ER association (Supplementary Figure S2). As calreticulin accumulation parallels class I accumulation, it is now understandable that in fibroblasts, in which class I accumulation is much less pronounced, a much smaller degree of calreticulin accumulation is observed.

Taken together, our data support the idea that calreticulin travels together with sub-optimally loaded class I molecules to the ERGIC and cis-Golgi. In some individual COS cells that strongly overexpressed H-2 D^b-GFP we found green fluorescence in lysosome-like structures that also stained with the calreticulin antibody (data not shown), suggesting that if the retrieval system from the cis-Golgi is saturated, calreticulin may be able to follow class I molecules even further along the secretory pathway.

Calreticulin mediates the retrieval of sub-optimally loaded class I from the Golgi to the ER

Since sub-optimally loaded class I molecules accumulate, before their transport back to the ER, in the same compartment as calreticulin, we next investigated whether class I retrieval was causally connected to calreticulin trafficking. We first asked whether the accumulation of sub-optimally loaded class I molecules that is apparent in human T1 and T2 lymphoblastoid cells (Garstka et al, 2007) and in COS cells (above) was also visible in K41 cells, and we found a weakly visible but highly reproducible accumulation of transfected H-2 D^b–GFP that co-localized with the cis-Golgi. Intriguingly, in K42 as well as in K42 CRT-HAΔKDEL cells, this accumulation no longer existed (Figure 5A–D). Instead, class I molecules in these cells appeared in large vesicular structures that were not part of the early secretory pathway but co-localized with markers of early endosomes and lysosomes, suggesting that class I molecules in K42 and in K42 CRT-HAΔKDEL cells do not remain at the surface for a long time but are rapidly endocytosed, presumably due to the dissociation of their bound pool of low-affinity peptides (Gao et al, 2002). As a control, a very good co-localization of H-2D^b–GFP with the Golgi apparatus was observed in all three cell lines on imposition of a 20°C Golgi exit block. We noted that under these conditions, CRT-HAΔKDEL also accumulated in the cis-Golgi (Supplementary Figure S3), suggesting that like the wild-type calreticulin, it was also associated with class I molecules, but in this case on its way to the cell surface.

Figure 5a-b.

Accumulation of class I molecules in the cis-Golgi depends on the KDEL signal of calreticulin. Cells were transfected with H-2D^b–GFP (green images) and stained with organelle markers (red images). Bars, 10 μm. (A) H-2D^b–GFP shows an accumulation in the early secretory pathway in K41 cells at 37°C (arrows), which is exaggerated on a 20°C Golgi exit block (arrowhead). There is little co-localization of H-2D^b–GFP with endosomes and lysosomes in K41 cells. (B, C), In K42 (B) and K42 CRT-HAΔKDEL cells (C), the accumulation of H-2D^b–GFP in the early secretory pathway is not observed at 37°C but becomes visible upon Golgi exit block at 20°C.

Figure 5c-e.

(B, C), In K42 (B) and K42 CRT-HAΔKDEL cells (C), the accumulation of H-2D^b–GFP in the early secretory pathway is not observed at 37°C but becomes visible upon Golgi exit block at 20°C. (D) Quantification of data from (A–C) with SEM (n=221, 99, 395, 258, 272, 120 cells). (E) H-2D^b(T134K)–GFP accumulates in the ER and cis-Golgi of COS cells, co-localizing with calreticulin. Cells were transfected with H-2D^b(T134K)–GFP and stained for endogenous calreticulin, p58 (ERGIC), and giantin (Golgi).

We conclude that in the absence of calreticulin or its C-terminal retrieval sequence, KDEL, sub-optimally loaded class I molecules no longer accumulate in the cis-Golgi, and that calreticulin retrieval is necessary for the retrieval of class I molecules from this compartment to the ER and for the efficient presentation of endogenous antigens. The simplest explanation for this finding is that Calreticulin interacts with peptide-receptive class I molecules in the cis-Golgi and, through its association with the KDEL receptor, mediates their uptake into retrograde COPI transport vesicles (Majoul et al, 1998).

Calreticulin acts in class I retrieval most likely outside the PLC

So far, our data do not address the question whether calreticulin retrieves class I molecules from the Golgi in the context of the PLC, outside the PLC, or both. We therefore made use of a T134K (lysine 134 to arginine) mutant of H-2D^b. T134K mutants of class I do not detectably interact with the PLC, as the mutation impairs the binding of tapasin. Their trafficking rate to the surface is faster than that of wild-type class I molecules, but their peptide optimization is poor, resulting in the transport of a greater proportion of sub-optimally loaded class I molecules to the cell surface (Lewis et al, 1996; Peace-Brewer et al, 1996; Sadasivan et al, 1996). When we transfected H-2D^b(T134K)–GFP into COS cells, we found a general increase in endosomal and lysosomal staining corresponding to a lower quality of the peptide ligands at the surface (data not shown), but nevertheless, in most cells, post-ER accumulation of the protein (indicative of retrieval to the ER) and co-localization with endogenous calreticulin was still visible (Figure 5E). The same effect was seen using a single chain fusion of β₂m and H-2D^b(T134K)–GFP, demonstrating that the calreticulin association was not with denatured heavy chain (Supplementary Figure S4, analogous to the experiment shown in Figure 4E). Thus, calreticulin may be able to retrieve class I molecules from the cis-Golgi even if it is not associated with them in the context of the loading complex.

Discussion

Besides its other cellular roles, such as in calcium homeostasis, calreticulin is crucial for protein folding and quality control in the ER (Bedard et al, 2005). In this function, calreticulin transiently associates with transmembrane and soluble proteins alternatively through a lectin–glycan or through a protein–polypeptide interaction (for review, see Williams, 2006), and it promotes their correct folding and maturation (Nauseef et al, 1995; Peterson et al, 1995; Wada et al, 1995). Its exact mechanism of action and the way in which it supports protein folding are unclear, but in addition to earlier reports that emphasized direct binding-mediated stabilization of unfolded proteins, more recent data have suggested that calreticulin may slow the transport of partially folded proteins along the secretory pathway, giving them more time to achieve their mature conformation. For example, misfolded influenza HA and other viral and cellular membrane proteins are retained intracellularly by calreticulin (Molinari et al, 2004; Popescu et al, 2005).

In the context of MHC class I antigen presentation, calreticulin was first described as a member of the PLC (Sadasivan et al, 1996), in which its only universally accepted role so far is to promote the tight association of tapasin and ERp57 with class I molecules. In this study, we have demonstrated an additional function of calreticulin in class I antigen presentation, independent of the PLC, by showing that the defect in class I peptide loading and maturation observed in calreticulin-deficient cells stems from the lack of intracellular retention of sub-optimally loaded class I molecules that is normally afforded by calreticulin.

Calreticulin deficiency strongly affects class I trafficking and localization. In K42 cells that lack calreticulin, intracellular retention is very inefficient, and class I molecules travel to the surface with a sub-optimal complement of peptides bound to them (Gao et al, 2002). We observe that in such cells, peptide-receptive class I molecules no longer accumulate in the cis-Golgi, but rather, after moving to the cell surface, class I molecules appear in endosomes and lysosomes, suggesting that the bound sub-optimal peptides dissociate rapidly and the class I molecules are then endocytosed.

In general, the intracellular retention of immature glycoproteins by calreticulin may involve retrieval from the cis-Golgi, inhibition of exit from the ER into COPII vesicles, or both. We find that calreticulin does not slow the exit of class I from the ER, and the accumulation of both peptide-receptive class I and calreticulin in the cis-Golgi (the site of residence of the KDEL receptor) suggests that calreticulin binds to sub-optimally loaded class I molecules in the ER, travels with them to the cis-Golgi, and mediates their retrieval by virtue of its C-terminal KDEL sequence (Figure 6 and Supplementary Figure S6). There are indeed numerous studies on the accumulation of calreticulin in post-ER compartments of mammalian cells, even in complex with class I (Spiro et al, 1996; Arosa et al, 1999; Zuber et al, 2001; Bedard et al, 2005; Neeli et al, 2007; Panaretakis et al, 2009). One study, which implies both retention and retrieval as a mechanism for the ER localization of calreticulin (Sönnichsen et al, 1994), finds that 29% of KDEL-deleted calreticulin is secreted from the cell within 3 h; if the wild-type protein leaves the ER at a similar rate (to be retrieved from the Golgi apparatus via the KDEL receptor), then within 1 min, 0.2% of the cellular calreticulin would be exported from the ER, corresponding to about 20 000 molecules (taking a cytosolic calreticulin pool into account; Means et al, 2006; Holaska et al, 2001). This is more than enough to accompany the 150 or so class I molecules that travel to the cell surface each minute, even if a 10-fold greater number is involved in an ER–Golgi cycle (Porgador et al, 1997; Schneeweiss et al, 2009). Importantly, the very efficient retrieval of KDEL proteins to the ER can lead to a very small steady-state population in the Golgi apparatus, which may be very difficult to detect by microscopy (Stornaiuolo et al, 2003; Supplementary Figure S5). Through deliberate overexpression of sub-optimally loaded murine class I molecules in COS cells (Figure 4C; analogous to the overexpression of the VSV glycoprotein by Hammond and Helenius (1994)), we, to the best of our knowledge, have shown here for the first time that calreticulin can leave the ER to accompany a sub-optimally folded protein to the Golgi apparatus.

Figure 6.

Model of calreticulin-mediated class I recycling. MHC class I heavy chains (HC) assemble with β₂m and the PLC in the ER. Optimally (pentagon) and sub-optimally loaded (triangle) class I–β₂m dimers (the latter in complex with calreticulin) exit the ER via COPII vesicles, travelling to the ERGIC and the cis-Golgi. In the cis-Golgi, sub-optimally loaded class I molecules accumulate before their retrieval, which is mediated by the KDEL tail of calreticulin. Optimally loaded class I molecules continue to the cell surface. A colour version is provided in Supplementary Figure S6.

The role of the PLC in class I retrieval is especially interesting. It is conceivable that calreticulin retrieves only, or preferentially, those class I molecules that are associated with it in the context of the PLC. This scenario would require the entire PLC to leave the ER along with the sub-optimally loaded class I molecules and to travel to the cis-Golgi to be retrieved from there. Indeed, there is some evidence for TAP and tapasin in post-ER compartments (Paulsson et al, 2002; Ghanem et al., submitted). Alternatively, the role of calreticulin in class I retrieval may be separate of that in the PLC and independent of it. An interaction between calreticulin and sub-optimally loaded class I molecules outside the PLC is conceivable, since calreticulin needs no co-factor to recognize its substrates (Peterson and Helenius, 1999), and since it was detected in a complex with class I in TAP-deficient cells (Sadasivan et al, 1996). Indeed, calreticulin binds to class I in the absence of tapasin (Solheim et al, 1997; Turnquist et al, 2002), and a number of mutant forms of H-2D^d, which do not bind tapasin, were shown to maintain a robust association with calreticulin (Paquet and Williams, 2002). Furthermore, the intracellular retention of sub-optimally loaded class I is known to be independent of the PLC: first, in tapasin-deficient cells (where no association of class I with the PLC is possible), tapasin-dependent class I molecules do not reach the cell surface (Lauvau et al, 1999; Purcell et al, 2001; Garstka et al, in preparation); and second, T134K mutants of class I molecules (which do not associate with tapasin) are still dependent on the peptide transporter, TAP, for surface expression, and are thus subject to normal quality control (Lewis et al, 1996; Sadasivan et al, 1996). In agreement with this, when we expressed the T134K mutant form of H-2D^b in COS cells (Figure 5E; Supplementary Figure S4), we found that it accumulated in a post-ER compartment like the wild type, suggesting retrieval to the ER in the absence of association with the PLC. In summary, our and others' data suggest that calreticulin-mediated quality control of the binding of high-affinity peptides to class I molecules is independent of the PLC. This opens up the possibility that sub-optimally loaded class I molecules can repeatedly associate with and dissociate from the PLC (as suggested previously by Wright et al (2004)) while still being held in the early secretory pathway.

Recently, the interaction between the KKXX tail of tapasin and the protein coat of COPI vesicles has been implicated in the retrieval of class I from the cis-Golgi (Paulsson et al, 2006), in agreement with earlier experiments that showed retention of class I by tapasin in insect cells (Schoenhals et al, 1999). Importantly, such a retrieval function by tapasin is not essential for localization of sub-optimally loaded class I, since in the absence of tapasin, tapasin-dependent class I molecules like HLA-B^*4402 are tightly retained in the cell; the same is true for the class Ib molecule, H2-M3 (Lybarger et al, 2001). Nonetheless, tapasin has been detected in ERGIC and/or the Golgi (Paulsson et al, 2002) in association with class I and the PLC, and retrieval of the entire PLC from ERGIC or Golgi may constitute a second, additional, pathway of class I localization, which may be especially important for those class I molecules that bind very tightly to the PLC.

It is unknown whether the functionally important interaction between calreticulin and class I takes place through the glycan moiety on the class I heavy chain (Radcliffe et al, 2002) or through a direct protein–polypeptide interaction. In support of the latter, it was recently shown that a calreticulin variant with an inactivated lectin function fully supports class I loading (Ireland et al, 2008). Although the lectin–glycan interaction between calreticulin and its substrates is known to depend on the modification of the latter by the conformational sensor, glucosyltransferase (see the Introduction section), the mechanism by which calreticulin detects its substrates through the protein–polypeptide interaction is not known. In molecular dynamics simulations, those ends of the lateral helices of the MHC class I peptide binding groove that are next to the peptide C terminus show a much increased mobility in the peptide-free form (Zacharias and Springer, 2004), which suggests that calreticulin might sense the difference between the peptide-bound and the peptide-free (or sub-optimally loaded) forms of class I through the differences in their conformational dynamics, and not their average structure.

Taken together, our data, to the best of our knowledge, link for the first time the optimal loading of MHC class I molecules with their recycling between two intracellular compartments, and they are consistent with an iterative mechanism for peptide cargo optimization, in which exchange between PLC-bound and -unbound states of class I molecules is essential for appropriate cargo optimization and quality control.

Materials and methods

Antibodies and cell lines

The monoclonal antibody (MAb) Y3 (Hammerling et al, 1982) recognizes a peptide-dependent epitope on the α₁ domain of H-2K^b. MAb B22.249 (Lemke et al, 1979) recognizes a peptide-dependent epitope on the α₁ domain of H-2D^b. MAb 25-D1.16 (Porgador et al, 1997) recognizes H-2K^b in complex with SIINFEKL (residues 257–264 of ovalbumin). MAb HC10 is specific for HLA-B and HLA-C molecules (Stam et al, 1990). PA3-900 antiserum (Affinity Bioreagents, Golden, CO, USA) was raised against recombinant human calreticulin. ERp57 antiserum (Zhang and Williams, 2006) was raised against glutathione-S-transferase (GST)-fused mouse ERp57. Anti-GST antibodies were removed by adsorption to GST–agarose before use. Anti-HA probe (Santa Cruz Biotechnology, USA; Chen et al, 1993) recognizes amino acids 98–108, YPYDVPDYASL, of influenza HA. Anti-glyceraldehyde-3-phosphate dehydrogenase (anti-GAPDH) was purchased from Abcam (Cambridge, UK). Anti-tapasin antiserum (Li et al, 1999) recognizes the C-terminal peptide sequence, CATAASLTIPRNSKKSQ. T18 antiserum (Rigney et al, 1998) that was raised against H-2D^b residues 1–193, recognizes epitopes present on the α₁ and α₂ domains of both H-2D^b and H-2K^b. Anti-rat TAP1 antiserum has been described earlier by Powis (1997). HRP-conjugated, anti-mouse and anti-rabbit, IgG Fc-specific antibodies were obtained from Sigma. Alexa Fluor 642 anti-mouse IgG was purchased from Molecular Probes. T2 (721 × CEM.T2) cells are human lymphocytes that have no TAP function (Salter and Cresswell, 1986). They were a kind gift from Peter Cresswell. Calreticulin-deficient K42 cells and their cognate wild type, K41, have been described earlier (Gao et al, 2002). COS-1 cells were obtained from the DSMZ (German Collection of Microorganisms and Cell Cultures).

Generation of calreticulin constructs

Rat calreticulin was cloned from the rat thymoma cell line C58. Briefly, RNA was extracted from 10⁷ cells using RNazol (Sigma). RNA was chloroform-extracted, precipitated with isopropanol, and re-suspended in DEPC water. cDNA was generated using 4 μg RNA, 1 × RNAguard (Roche), 1.5 U AMV reverse transcriptase, and 1 μM of the rat calreticulin-specific reverse oligonucleotide, 5′-CTAGTCTAGACTACAGCTCATCCTTGGCTTGGCCAGT-3′. Reactions were incubated at room temperature for 10 min, followed by incubation at 42°C for 1 h. Rat calreticulin cDNA was amplified using 1 μl of the cDNA reaction, 1 μM of the rat calreticulin forward oligonucleotide, 5′-CCGGAATTCATGCTCCTTTCGGTGCCGCTCCTGCTT-3′, and high-fidelity polymerase (Roche) by PCR (cycle: 1 × 95°C for 120 s; 15 × (94°C 30 s, 65°C-1K per cycle 60 s, 68°C 180 s); 25 × (94°C 30 s, 50°C 30 s, 72°C 180 s+10 s per cycle); 72°C 10 min). Rat calreticulin cDNA was then cloned into pCDNA3(+)hygro and sequenced; it revealed rat calreticulin to be 98% identical to the mouse gene on the amino acid level (data not shown). Calreticulin variants were amplified using the following primers: Forward, 5′-CCACTCGAGGCCACCATGCTCCTTTCGGTGCCG-3′; Reverse: for KDEL: 5′-GGTGTTAACCTACAGCTCATCCTTGGCTTG-3′; for HAKDEL: 5′-GGTGTTAACCTACAGCTCATCCTTGAGGCTAGCGTAATCCGGAACATCGTATGGGTAGGCTTGGCCAGTGGCATC-3′; for HAKDEV: 5′-GGTGTTAACCTACACCTCATCCTTGAGGCTAGCGTAATCCGGAACATCGTATGGGTAGGCTTGGCCAGTGGCATC-3′; for HAΔKDEL: 5′-GGTGTTAACCTAGAGGCTAGCGTAATCCGGAACATCGTATGGGTAGGCTTGGCCAGTGGCATC-3′. The viral recombination vector CMV-bipep-ΔNGFR (Tolstrup et al, 2001) expresses an insert and the cytosolic tail-truncated p75 nerve growth factor receptor (ΔNGFR) from the same promoter via an internal ribosomal entry site (IRES). Retroviral transduction was carried out according to the Nolan group protocols (http://www.stanford.edu/group/nolan). Briefly, 2 × 10⁶ cells were plated overnight in a 6-well plate and transfected the following day (day 1) at a density of 70–80% using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions (1 μg DNA). On day 2, cells were given 3 ml fresh medium (DMEM supplemented with 10% FCS) and incubated at 32°C overnight. On day 3, cell medium was collected and centrifuged at 1500 r.p.m. for 5 min. Cells to be transduced were re-suspended in this supernatant that contained retroviral particles, 5 μg/ml polybrene (hexadimethrine bromide; Sigma) and 25 mM HEPES (Sigma), and then centrifuged at 2500 r.p.m. for 2 h at 32°C, incubated at 37°C, and re-suspended in fresh medium the following day. ΔNGFR-expressing cells were purified once using antibody-coated CELLection Pan Mouse IgG Dynabeads (Dynal Biotech) and clones were established by limiting dilution.

Immunoprecipitations from lysates and medium supernatants

Cultures were seeded at the same density and cultured for up to two days (Figure 1C). At time points 0, 24, and 48 h after seeding, cultures were collected, counted, the cells were pelletted, supernatant was removed, pellet was washed four times in cold PBS and lysed in 25 μl lysis buffer, the nuclei were spun out, and the remaining lysate was mixed with 25 μl loading buffer. The supernatants were concentrated by TCA precipitation and proteins were washed thrice in ice-cold acetone and re-suspended in 50 μl loading buffer. Ten microlitres of either type of sample were loaded onto an SDS gel, and calreticulin was detected by western blotting using PA3-900 (K41, KDEL) or anti-HA (HAKDEL, HAKDEV, HAΔKDEL) antibody. Blots were also probed with anti-GAPDH antibody to control for variations in cell density. Over the experimental time period, the cell number increased, but by a factor of less than two in each culture. Viability at the end of culture was greater than 98% for each, thus ruling out the possibility that the calreticulin we detected in supernatant was from small (non-pelletable) debris or from the lysis of cells.

Detection of TAP1-associated proteins

A total of 7 × 10⁶ cells were washed once in PBS, pelleted, and lysed on ice in 700 μl Tris buffered saline (pH 7.4; TBS: 151 mM NaCl, 10 mM Tris, 1% digitonin (WAKO, Richmond, USA), 2 mM phenylmethylsulfonyl fluoride, 5 mM iodoacetamide, and 3 μl sheep anti-rat TAP1 antibody), and rotated at 4°C for 15–20 min. Lysates were clarified by centrifugation at 15 000 g for 15 min. A total of 40 μl of a 50% suspension of a mixture of Protein A and G sepharoses (1:1) was added and samples were rotated at 4°C for 1 h. Beads were then washed thrice in 0.1% digitonin, and excess liquid was aspirated using a hypodermic needle. Proteins were then eluted with 40 μl non-reducing loading buffer at room temperature for 20–30 min, separated by SDS–PAGE, transferred onto Protran nitrocellulose paper (GE, Amersham, UK), and detected with the West Pico (or Femto) Chemoluminescent reagent (Pierce, Rockford, USA).

Antigen presentation assay and FACS analysis

Cells (2 × 10⁶) were re-suspended in 100 μl of Amaxa (Cologne, Germany) nucleofection ‘V' solution and mixed with 3 μg plasmid DNA encoding GFP fused to ubiquitin and SIINFEKL peptide (provided by Dr Jacques Neefjes, Amsterdam; Neijssen et al, 2005). Electroporation was carried out using a Nucleofector I instrument (Amaxa) set to program T-30. Subsequently, cells were re-suspended in pre-warmed RPMI 1640 medium containing 10% FCS and dispensed into a 6-well plate. After 24 h, cells were collected and prepared for flow cytometry using the K^b–SIINFEKL-specific mAb 25-D1.16 followed by Alexa Fluor 642 anti-mouse antibody. Analysis was undertaken on a FACSCanto (Becton Dickinson, US) using FACSDiva software, with at least 40 000 events collected. To generate the plot in Figure 1E, we calculated the mean antigen dose and mean 25D1.16 staining intensity for each of the eight gates (sectors) shown in Figure 1D. These mean values were then plotted as the eight data points shown in Figure 1E for each transfectant.

COPII vesicle formation assays and evaluation

The assays were carried out as described by Garstka et al (2007). Briefly, cells were radiolabelled with [³⁵S]-methionine for 30 min and microsomes were prepared by repeated freezing and thawing, and differential centrifugation. Budding reactions consisted of microsomal membranes, pig brain cytosol (isolated as described in Garstka et al (2007)), ATP regenerating system, and 0.2 mM GTP. Vesicles were isolated from the supernatant resulting from a 15 000-g spin and sedimented by a 100 000-g centrifugation; lysed in 1% Triton X-100 in 50 mM Tris–Cl (pH 7.5) and 150 mM NaCl; and radiolabelled proteins were immunoprecipitated using antibodies against the HA tag (Figure 4A), the Na⁺/K⁺ ATPase (Figures 3A and 4A), or with the conformation-specific antibodies Y3 (against H-2K^b) and B22.249 (against H-2D^b); separated on SDS–PAGE; and detected by autoradiography. In Figure 3A, some samples were treated before SDS–PAGE with PNGase or with EndoF₁ that is identical in its cleavage specificity to EndoH (Trimble and Tarentino, 1991). Both glycosidases were obtained as kind donations from P van Roey and AL Tarentino, fused to the maltose-binding protein in a plasmid of the pMAL expression system, and prepared as described by the manufacturer of the system (New England Biolabs).

To generate the graphs in Figure 3B and C, the COPII vesicle packaging efficiencies for Db, Kb, and the ATPase were determined separately for each of the three independent experiments. First, the EndoF1-sensitive (pre cis-Golgi) bands were quantified: for the peptide-occupied class I molecules (without peptide added to the lysate), lanes 2 (vesicles) and 8 (donor membranes), and for total class I (with peptide added to the lysate), lanes 3 (vesicles) and 9 (donor membranes). The total amount of class I in the budding reaction was calculated from the sum of the signal in the spent donor membranes and the vesicles (it is important to note that in lanes 8–11, only 50% of the donor membranes are applied to the gel), and the packaging efficiency was calculated as the class I molecule or ATPase content of the COPII vesicles expressed as a percentage of the class I molecule or ATPase amount in the total reaction. Next, to compensate for the experiment-to-experiment variation in the overall formation of COPII vesicles, budding efficiencies of the class I molecules were divided by the budding efficiencies of the ATPase. The resulting relative budding efficiencies from three independent experiments were averaged and are shown in Figure 3B and C with their s.e.m. values.

Generation of the fluorescent protein constructs

H-2K^b–GFP and H-2D^b–GFP have been described by Garstka et al (2007). To generate the single-chain fusion, β₂m–H-2D^b–GFP, the gene for human β₂m (including the signal sequence) was cloned into pCRII-TOPO (Invitrogen), cut using NcoI and XhoI, and cloned into the NcoI and SalI sites of pASK-IBA5 (IBA, Göttingen, Germany). The oligonucleotides 5′-CATGGGTGGCGGAGGTAGTGGTGGCGGTGGCTCCGGTGGTGGCG-3′ and 5′-GATCCGCCACCACCGGAGCCACCGCCACCACTACCTCCGCCACC-3′, encoding the linker sequence (GGGGS)₃, were annealed and ligated into the BamHI and PciI sites of the resulting vector. The BamHI site of pTM1 (Moss et al, 1990) was removed by filling in with Pfu polymerase and blunt re-ligation, and the β₂m/linker sequence was moved into the NcoI and SacI sites of that vector to obtain the single chain expression vector, pCSM27. The gene for H-2D^b was amplified using the primers 5′-GCGGGATCCGGCCCACACTCGATGC-3′ (forward) and 5′-CCCGTCGACTTATCACGCTTTACAATCTCG-3′ (reverse), and cloned into the BamHI and SalI sites of pCSM27. The entire β₂m–H-2D^b fusion gene was then amplified using the primers 5′-CCCCGTCGACCATGGCTCGCTCCGTGGCC-3′ (forward) and 5′-CCCACCGGTGGCGCTTTACAATCTCGGAGAGA-3′ (reverse), and cloned into the SalI and AgeI sites of pEGFP-N1 (Invitrogen). To generate H-2D^b(T134K)–GFP, the point mutation was introduced into the H-2D^b–GFP gene by QuikChange site-directed mutagenesis using the following primers: 5′-CCTGAAAACGTGGAAGGCAGCTGACATGGCGGCGCA-3′ and 5′-CATGTCAGCTGCCTT CCA CGTTTTCAGGTCTTCGTTCAGGG-3′. To generate β₂m–H-2D^b(T134K)–GFP, the H-2D^b(T134K) gene was amplified and cloned into pCSM27, following PCR and transferred into pEGFP-N1 as explained above. All constructs were verified by DNA sequencing. CRT–GFP was a gift from Erik L. Snapp (Albert Einstein College of Medicine).

Immunofluorescence microscopy and organelle markers

Immunofluorescence microscopy was carried out as described by Garstka et al (2007). Compartment markers used were as follows: anti-PDI (protein disulphide isomerase) antibody for the ER, anti-p58 antibody for ERGIC (a gift from Ralf Pettersson), p23-CFP (a gift from Jennifer Lippincott-Schwartz), GM130 antibody (BD Biosciences), and anti-giantin antibody (Alexis Biochemicals, Lörrach, Germany) for the Golgi apparatus; rab11–YFP for early endosomes (a gift from Marino Zerial, Dresden); and CD63–CFP (a gift from Paul Luzio, Cambridge) for lysosomes. Secondary antibodies (anti-mouse or anti-rabbit conjugated with Cy3 or Cy5) were purchased from Jackson ImmunoResearch Laboratories (Soham, UK). The fluorescent SIINFEKL peptide (with TAMRA attached to the lysine residue) was obtained from Biosyntan (Berlin, Germany). For the peptide stain, cells were fixed with 1% paraformaldehyde, permeabilized with 0.1% saponin, and incubated in a 500-nM solution of the peptide in PBS for 40 min.

Supplementary Material

Supplementary Information

Review Process File