Selective Export of HLA-F by Its Cytoplasmic Tail¹

The highly polymorphic classical MHC molecules HLA-A, -B, and -C function in the presentation of peptides for recognition by TCRs (1, 2) as well as being recognized by killer Ig-like receptors (3). In contrast, the functions of the nonclassical class I molecules are more specialized. The surface expression of HLA-E is dependent on the availability of leader sequences from classical class I molecules (4) and the expression of HLA-G is restricted to the placenta (5). HLA-F is the least characterized nonclassical class I molecule. Although HLA-F has been shown to interact with calreticulin and TAP (6, 7), it is not clear whether HLA-F binds and presents peptides. HLA-F was reported to be retained intracellularly, in what was presumed to be the endoplasmic reticulum (ER)⁴ (6, 7, 8). Recently, however, using a new HLA-F specific mAb, HLA-F has been shown to be expressed on the surface of some B cells and monocytic cell lines (9).

After entering the ER, newly synthesized MHC class I H chains bind a variety of chaperones including calnexin, calreticulin, ERp57, and tapasin, that assist in folding and peptide loading (for review, see Ref. 10). Peptides generated in the cytosol by proteasomes are transported by TAP molecules and loaded onto H chain/β₂-microglobulin dimers to form a heterotrimer which is then released from the peptide-loading complex. Following exit from the ER, MHC class I molecules transit through the Golgi on their way to the cell surface for recognition by T cell and NK cell receptors.

It is unclear how MHC class I molecules exit the ER after binding peptide. Traditionally, it has been assumed that they are rapidly exported from the ER by a default exocytic pathway (11), because ER-to-Golgi transport was thought to be associated with release of class I from TAP (12, 13, 14). With the increasing characterization of ER export motifs such as the diphenylalanine FF (15) or diacidic DxE motifs (16), selective export into coat complex (COP)II-coated vesicles via ER exit signals is recognized as an important mechanism of transport for some proteins (for a review, see Ref. 17). The possibility that this may be germane to ER export of MHC class I molecules came with the realization that peptide loaded MHC class I molecules persist in the ER once released from TAP (18, 19). Indeed, Spiliotis et al. (20) showed that mouse MHC class I molecules are recruited into ER exit sites by a selective process. There are two possible mechanisms to explain selective export of MHC class I molecules, namely that they contain ER export sequences themselves, or that they associate with a transport receptor which contains an export motif. As swapping the cytoplasmic tails of H2K^k with H2L^d did not affect the rate of allele-specific ER-to-Golgi transport (20), export was proposed to be via a transport receptor, namely BAP31 (20, 21). However, the same group have not ruled out that HLA molecules contain other unidentified ER exit signals (22). Although the role of cytoplasmic tail residues in endocytosis (23, 24, 25), internalization, degradation (26, 27) and ER retrieval (24) have been revealed, a motif responsible for anterograde transport from the ER to Golgi has never been identified in an MHC class I molecule.

Our aim was to determine whether the cytoplasmic tails of HLA class I molecules are responsible for ER export. In this study, we report the identification of two export motifs in HLA-F and demonstrate the interaction of the cytoplasmic tail of this nonclassical class I molecule with two components of the ER-to-Golgi transport machinery.

Exons 2–8 of HLA-F, HLA-A*0209, and HLA-B*2705 were PCR amplified from vectors (gifts from E. Lepin and J. Goodall, University of Cambridge, Cambridge, U.K.) and cloned into mammalian cell vectors containing the CMV promoter (pcDNA3.0; Invitrogen Life Technologies). Each vector, named PK1, contained the generic N-terminal signal sequence, a GFP cassette (BD Clontech) followed by a myc tag, a GAGA linker and the insertion site. PCR fragments were cloned between the XhoI and XbaI sites. The PCR fragments were also cloned into a vector containing two protein A cassettes in place of the GFP sequence (named ZZ). The cytoplasmic tails of all MHC class I molecules were N-terminally tagged with GST for bacterial expression by cloning PCR fragments into pGEX-6p-2 (Amersham Biosciences) between the EcoRI and XhoI sites. Sequence analysis using BigDye technology was performed.

Site-directed mutagenesis was performed to truncate or mutate specific residues in the cytoplasmic tails (Table I) of HLA molecules using the QuikChange site-directed mutagenesis kit (Stratagene) and transformed into XL2-blue ultracompetent cells (Stratagene).

HeLa cells were grown on glass coverslips in RPMI 1640 (Invitrogen Life Technologies) supplemented with 5% heat inactivated FCS, 20 mM HEPES, 2 mM glutamine, 100 μg/ml streptomycin, and 100 U/ml penicillin. Cells were transiently transfected using FuGene 6 transfection reagent (Roche) and fixed 48-h posttransfection with 4% paraformaldehyde. Cells were permeabilized with 0.5% (v/v) Triton X-100 in PBS, blocked in 20% (v/v) heat-inactivated FCS, 0.5% (v/v) Tween 20 in PBS and stained in the same solution. Mouse mAbs to the cis-Golgi protein GM130 (BD Transduction Laboratories), myc (mAb 9E10), HLA-F (mAb FG1, a gift from E. Lepin) and MHC class I (W6/32) were detected with species-specific Alexa 568 or Alexa 488 secondary Abs (Molecular Probes) and the cells were mounted in fluoromount-G (Southern Biotechnology Associates). Images were obtained using a Zeiss confocal microscope.

To confirm cell surface expression transfected cells where incubated continuously with the mouse mAb 9E10 for 3 h at 37°C before washing, fixing, and permeabilization as above. Myc Ab was detected with Alexa 568 goat anti-mouse IgG.

HeLa cells transiently transfected with untagged HLA-A2 or GFP-A2 were metabolically labeled as in (28). After preclear on Sepharose beads, the mouse anti-HLA-A2 mAb BB7.2 and protein A-Sepharose were used to immunoprecipitate HLA-A2 from lysates. After washing three times in lysis buffer, proteins were eluted from the beads in SDS sample buffer.

Transfected HeLa cells lysed in SDS sample buffer or metabolically labeled, immunoprecipitated samples were treated with 1000 U of EndoH_f (New England Biolabs) in G5 buffer for 2 h at 37°C. For protein blotting, samples were heated at 90°C for 5 min, then separated by SDS-PAGE and analyzed by immunoblotting.

GST fusion proteins were expressed in the Escherichia coli BL21-GOLD strain (Stratagene). Protein expression was induced at OD 600 ∼0.6–0.8 overnight with 0.2 mM isopropyl β-d-thiogalactopyranoside at 22°C. Lysates were prepared by dounce homogenization in 25 ml of lysis buffer (20 mM Tris, 50 mM NaCl, 1 mM EDTA, 2.5 mM MgCl₂ with 1 mM PMSF and protease inhibitors (Roche)) followed by sonication. Lysates were clarified by centrifugation at 17,000 × g for 10 min.

Bacterial lysates were incubated with glutathione Sepharose beads (Amersham Biosciences) at 4°C for 30 min followed by washing in lysis buffer to remove unbound bacterial proteins. Mammalian cell lysates were prepared from ∼1 × 10⁸ HeLa cells per GST sample by dounce homogenization in 10 ml of lysis buffer followed by incubation at 4°C for 30 min with 1% (v/v) Triton X-100. Lysates were cleared by centrifugation at 17,000 × g for 10 min at 4°C followed by incubation with glutathione Sepharose beads at 4°C for 30 min to remove any proteins interacting with beads alone. Precleared HeLa cell lysates were applied to GST-HLA tails or GST control beads and incubated with rotation for 2 h at 4°C. Beads were washed extensively in lysis buffer + 1% Triton X-100 before bound proteins were eluted by the addition of 300 μl of high salt elution buffer (20 mM Tris (pH 7.4), 1.5 M NaCl, 2 mM EDTA, 5 mM 2-ME). Proteins were precipitated, then resuspended in 50 μl of SDS sample buffer. The GST proteins were eluted from the beads in 50 μl of SDS sample buffer post high salt elution. After gel electrophoresis GST proteins were stained with Coomassie blue.

Proteins were separated by SDS-PAGE and transferred to an Immobolin transfer membrane (Millipore). Membranes were blocked for 30 min in PBS, 5% (w/v) dried milk and 0.05% (v/v) Tween 20 before blotting with rabbit anti-COPII (Affinity BioReagents), mouse anti-14-3-3 β (Santa Cruz Biotechnology), Rabbit anti-ERp57 (Santa Cruz Biotechnology) or 9E10 anti-myc mAb for 1 h in blocking buffer. Species-specific HRP secondary Abs were detected by ECL (Amersham Biosciences).

Approximately 48 h posttransient transfection, HeLa cells expressing GFP-F-Ex7 or GFP-A2 were stained with a mouse anti-GFP mAb (Roche) for 30 min at 4°C, which was subsequently detected with goat anti-mouse IgG Alexa Fluor 647 (Molecular Probes). Cells were analyzed on a BD Biosciences FACSCalibur 4-Color Analyzer.

The products of different HLA class I loci have marked differences within their cytoplasmic tails although, unlike polymorphic residues in the Ag presenting groove, these are relatively conserved among alleles of different HLA isotypes (Fig. 1 A). This conservation suggests an important role for the cytoplasmic tail. Differences in the tails between loci imply that traffic of the various isotypes may be controlled independently of each other. Previous studies have suggested that export of some class I molecules can occur independently of their cytoplasmic tails (29). However a selective export mechanism has been suggested for class I (29).

To determine the general importance of the cytoplasmic tails in export to the cell surface we removed the cytoplasmic tails from various HLA molecules and observed the effect on trafficking. We chose to examine the transport of MHC class I molecules throughout the secretory pathway using a GFP chimera to overcome conformational specificity issues of different Abs (Fig. 1,B). Our constructs were tagged with GFP at the N terminus, downstream of a signal sequence, allowing confirmation of cell surface expression. Moreover, because the cytoplasmic tail of type I transmembrane proteins often contain targeting motifs, an N-terminal tagging strategy would avoid interfering with this region of the protein. To confirm that tagging with GFP at the N terminus did not alter trafficking, the maturation of GFP-A2 was compared with untagged A2. This revealed that their trafficking was similar with most transfected A2 gaining EndoH resistance by 3 h (Fig. 1 C). It suggests that tagging with GFP at the N terminus does not dramatically alter the kinetics of transport for HLA class I molecules.

When the classical class I molecule HLA-A*0209 tagged with GFP (GFP-A2) was transiently expressed in HeLa cells it was exported to the Golgi, shown by the perinuclear localization and was expressed mainly at the cell surface, as expected (Fig. 1,D). Although more ER localization was observed, removal of the tail from HLA-A2 (GFP-A2 Δ tail) still resulted in a significant amount of export of HLA-A2 to the Golgi and the cell surface (Fig. 1,D). In contrast to HLA-A2, the nonclassical class I molecule HLA-F tagged with GFP (GFP-F) was expressed in the ER and accumulated in a perinuclear region (Fig. 1,D). This was later confirmed to be the Golgi as shown by colocalization with the cis-Golgi marker GM130 (Fig. 2,B). In addition, removal of its tail (GFP-F Δ tail) resulted in its retention in the ER (Fig. 1 D). We do not believe that the ER retention of GFP-F Δ tail is due to misfolding of the protein as it was recognized by a conformational specific Ab FG1 (data not shown). Thus, while export of HLA-A2 is largely independent of its intracellular sequence, ER export of HLA-F is entirely dependent on its cytoplasmic tail.

Unlike all other HLA class I molecules, the trafficking and function of HLA-F remains relatively uncharacterized. There are reports of expression of HLA-F both in the ER (6, 7) and on the cell surface (9). Two major HLA-F protein sequences are reported in databases. The first contains eight exons (accession number NM_018950) and the second is an alternatively spliced version lacking exon 7, which encodes part of the cytoplasmic domain (accession number AY253270) (Fig. 2,A). We examined the trafficking of both of these gene products. Detailed examination of the expression of full-length GFP-F in HeLa cells confirmed the accumulation of the fusion protein in both the ER and the Golgi where it colocalized with the Golgi marker GM130 (Fig. 2,B). In addition a small amount of cell surface expression was also observed, although this is much lower than that observed for classical class I molecules. GFP-F lacking exon 7 (GFP-F-Ex7) was similarly localized in the ER and Golgi with a small amount present at the cell surface (Fig. 2,B). Golgi localization was not a result of GFP tagging of the HLA-F molecule as a similar localization was found when HLA-F was tagged at the N terminus with a smaller FLAG tag (data not shown). In addition, surface expression of the classical class I molecules HLA-B27 tagged with GFP (Fig. 2,B) and GFP-A2 (Fig. 1,C) was not compromised. Recognition of GFP-B27 transfectants with the conformation dependent, MHC class I-specific mAb W6/32 and GFP-F transfectants with the HLA-F specific mAb, FG1 (Fig. 2 B) confirmed that the N-terminal GFP tag did not alter folding of the HLA molecules.

To quantify the amount of HLA-F expressed on the cell surface compared with the classical class I molecule HLA-B27, sensitivity of the protein to EndoH was tested. EndoH is an enzyme that does not cleave processed N-linked sugars, a modification indicative of transit through the Golgi. The extent of EndoH resistance at steady state was investigated by Western blot analysis. Approximately 50% of the classical class I molecule HLA-B27 was EndoH resistant, implying passage through the medial Golgi (Fig. 2,C). In contrast, the majority of HLA-F was EndoH sensitive. However, a faint resistant band (∗) was reproducibly present in both full-length HLA-F and HLA-F lacking exon 7 (Fig. 2 C). This population of HLA-F likely represents that found at the plasma membrane and is consistent with the results from confocal microscopy. The EndoH-resistant band was more apparent in cells expressing HLA-F lacking exon 7. Interestingly, a smear of higher m.w. bands was consistently observed with HLA-F. This may represent some posttranslational modification of HLA-F, and requires further investigation.

To confirm that a low level of HLA-F is trafficked to cell surface, we incubated cells with Abs against the N-terminal myc tag present in the HLA-F construct (see Fig. 1,B). As only a small population of HLA-F traffics to the cell surface at any given time we allowed cells to be incubated continuously with Ab for a period of 3 h at 37°C. This enabled sufficient Ab to be taken up. The Ab was then removed by washing before permeabilization and detection with secondary Abs. Using this approach, we were able to detect HLA-F clearly at the cell surface (Fig. 2,D). In addition, we could show that HLA-F was internalized from the cell surface into punctate structures in a perinuclear location that most likely represent recycling endosomes. Flow cytometry also revealed surface expression of transfected HLA-F. However, levels were much lower than those for the classical class I molecule HLA-A2, confirming that the majority of HLA-F was indeed retained intracellularly (Fig. 2 E).

As removal of the cytoplasmic tail of HLA-F results in ER retention of HLA-F, we examined the HLA-F cytoplasmic tail for ER export signals. The first that became apparent was a C-terminal valine. This configuration promotes ER export of CD8α chain (30). Both full-length and truncated HLA-F proteins have a terminal valine residue, encoded by exon 8 (Fig. 2,A). To determine whether this residue contributes to forward transport of HLA-F, we examined the trafficking of GFP-F in which the C-terminal valine has been deleted (GFP-F ΔV) (Fig. 3,A). As the majority of residues, including serine, are unable to substitute for valine for ER export (31), we analyzed whether trafficking of HLA-F with a serine for a valine substitution (GFP-F V→S) was also compromised. When transfected into HeLa cells, GFP-F ΔV was clearly retained in the ER and did not colocalize with the Golgi resident protein GM130 (Fig. 3,B). A similar result was observed with the full-length construct in which the valine was substituted for serine (data not shown). When the valine residue was substituted for serine in the truncated version of HLA-F (GFP-F-Ex7 V→S) this construct was also retained in the ER (Fig. 3 B). These results imply that, as with CD8α, the C-terminal valine of both full length and truncated HLA-F is essential for ER export.

A number of groups have proposed that the C-terminal valine functions in ER export by binding to cytosolic factors, resulting in recruitment of cargo to ER exit sites. Here they can become incorporated into COPII-coated vesicles, which bud from the ER and are transported to the Golgi (32). Indeed, when the C-terminal residue of ERGIC-53 is mutated from diphenylalanine to valine the cytoplasmic tail still binds to Sec23, a component of the COPII coat (33).

To identify proteins involved in the forward transport of HLA-F we initially attempted immunoprecipitations from transiently transfected cells followed by Western blotting for candidate proteins. This approach was, however, unsuccessful perhaps due to the transient nature of such interactions. We therefore used a more robust approach in which the cytoplasmic tail of HLA-F was attached to the C terminus of GST. The use of a GST fusion protein results in dimerization (via the GST moiety (34)), and perhaps increases the avidity if the molecule for its binding partners. Following affinity chromatography Western blotting was performed with Abs against potential candidate molecules. GST alone, and GST with the cytoplasmic tails of the other HLA class I molecules, were used as controls. We found that the cytoplasmic tail of HLA-F pulled down Sec23, a component of the COPII coat (Fig. 3,C). The binding of the HLA-F tail to Sec23 was enriched since Sec23 was barely detectable in the input lysate (Fig. 3,C). Sec23 did not bind to GST alone confirming that the interaction was specific to the HLA-F cytoplasmic tail. The cytoplasmic tails of the other class I molecules did not bind to Sec23 either (Fig. 3,C). The terminal valine was confirmed to be responsible for binding to COPII as its deletion reduced binding of Sec23. This was not due to difference in the levels of GST-F and GST-FΔV used, as similar amounts were loaded as revealed by Coomassie staining of the GST proteins (Fig. 3,C). In addition the HLA-F tail did not interact with ERp57 (Fig. 3 C) used here as a negative control to rule out nonspecific binding of GST constructs to cellular proteins. Interestingly the C-terminal valine, which functions in ER export of HLA-F, is the only residue encoded by exon 8 in the molecule. It is intriguing that a separate exon, encoding a single amino acid, is devoted to this specific function.

For classical MHC class I molecules, it is currently thought that binding peptides in the ER results in release from the peptide loading complex and export to the Golgi. Because the C-terminal residue of HLA-A, like HLA-F, is a valine (Fig. 1,A), we examined whether this residue functions in ER export of this classical class I molecule. When the C-terminal valine was removed, the HLA-A2 molecule (GFP-A2 ΔV) was still exported to the Golgi and the cell surface (Fig. 4,A). These findings were confirmed by testing EndoH resistance. Both HLA-A2 and HLA-A2 ΔV were ∼50% EndoH resistant compared with HLA-F ΔV, which was entirely EndoH sensitive at steady state (Fig. 4 B). A smear of higher m.w. bands was again observed with HLA-F and not with HLA-A2 (and HLA-B27) molecules. The smear was only seen with HLA-F that is capable of trafficking to the Golgi and not with HLA-FΔV that remains in the ER. Therefore, this smear likely represents post-ER modification unique to HLA-F.

We suspected that the differences in cytoplasmic tail dependence for export might be due to differences in peptide binding with export of HLA-A2 being peptide dependent and cytoplasmic tail independent. Conversely, export of HLA-F would be largely peptide independent but cytoplasmic tail dependent. To reconcile the fact that HLA-F export was dependent on the C-terminal valine but HLA-A was not, we swapped the tails of HLA-F and A2 (Fig. 5,A) and determined whether export was dependent on the C-terminal valine residue in each case. HLA-F with the cytoplasmic tail of A2 (GFP-F:A) trafficked similarly to GFP-F with some localization to the Golgi (Fig. 5,B). However, when the valine was removed from HLA-F with the A tail (GFP-F:A ΔV), HLA-F again was retained entirely in the ER (Fig. 5 B). Thus, HLA-F is indeed dependent on this export motif in the cytoplasmic tail for forward transport.

In contrast, HLA-A with the cytoplasmic tail of HLA-F (GFP-A:F) was expressed mainly on the cell surface with some localization to the Golgi area (probably the trans-Golgi as limited colocalization was observed with GM130) (Fig. 5,B). When the valine was removed (GFP-A:F ΔV), HLA-A2 was still exported to the cell surface (Fig. 5 B). Therefore, forward transport of HLA-A was independent of such export motifs in its tail. These finding were confirmed by examining sensitivity to EndoH which showed ∼50% EndoH resistance for both GFP-A:F and GFP-A:F ΔV, a small proportion of EndoH resistance of GFP-F:A and total EndoH sensitivity for GFP-F:A ΔV (data not shown).

Taken together, our data suggest that there are two ways of controlling ER export of MHC class I molecules that may be used in different circumstances. The first is a peptide-independent, cytoplasmic tail-dependent pathway. This pathway is exemplified by HLA-F, which when lacking its cytoplasmic tail (GFP-F Δ tail) or a motif important for export (C-terminal valine), is retained in the ER. It is plausible that ER export of HLA-F may be independent of binding peptide because surface expression of HLA-F has been shown to be independent of both TAP and tapasin (9). In addition, peptide has never been detected after elution from its binding groove. One interpretation of the importance of the valine in export of HLA-F is that the C-terminal valine is only essential for ER export of class I molecules in the absence of binding peptide.

The second pathway is a cytoplasmic tail-independent pathway characteristic of classical class I molecules such as HLA-A2, export of which seems to be largely dependent on peptide binding. Although export of HLA-A2 was obviously affected when its tail was removed (GFP-A2 Δ tail), resulting in more ER and Golgi localization and less surface expression, we still observed export of HLA-A2 to the Golgi and the cell surface. This is in agreement with previous studies in which truncation of mouse class I tails lowered surface expression compared with full-length but export was still achieved (29). Our findings with HLA-A2 also agree with those of Spiliotis et al. (20) who, using tail swapping experiments, determined that the export of mouse MHC class I was independent of cytoplasmic tail. In this way, the presence of peptide in the groove may mask the requirement for trafficking motifs such as the C-terminal valine in the cytoplasmic tail of the class I molecule itself. Precedence for such a mechanism has already been proposed, because masking of the di-lysine retrieval motif in HLA-G when high affinity peptide was present prevented recycling of HLA-G to the ER (24).

In addition to the C-terminal valine, we noted that the cytoplasmic tail of HLA-F also encodes an RxR motif (RxRxxxS). This motif has recently been shown to be important for forward transport of both the homodimeric KCNK3 potassium channel (35) and the K_ATP channel α subunit, Kir6.2 (36). The RxR motif is not present in the cytoplasmic tails of other HLA class I molecules (Fig. 1,A). To determine the contribution of this motif to the anterograde transport of HLA-F, we mutated the RxR to GKG, the sequence found at the same position in classical class I molecules and determined the effects on trafficking (Fig. 6,A). When the RxR motif was mutated in HLA-F, trafficking of HLA-F was disrupted (Fig. 6,B). Most of the molecule was retained in the ER, and there was only faint staining in the perinuclear region and no surface expression. Similarly, when the RxR motif was mutated in the shortened form of HLA-F (GFP-F-Ex7) most of the protein localized to the ER (Fig. 6 B). Taken together, these observations are consistent with a role for the RxR motif in forward transport of HLA-F.

Candidate molecules for binding to the RxR motif to assist in anterograde transport are the 14-3-3 proteins (37). For some transmembrane proteins, an ER localization is thought to occur by COPI binding to monomeric RxR motifs and recycling of these proteins back to the ER from the Golgi (35). However, upon dimerization or tetramerization of RxR motifs COPI binding is masked, and 14-3-3 proteins bind, resulting in transport beyond the cis-Golgi (36).

Therefore, to determine whether the cytoplasmic tail of HLA-F bound to 14-3-3 proteins, we again used a GST affinity chromatography approach followed by Western blotting of elutants with a 14-3-3-specific mAb. We found that the cytoplasmic tail of HLA-F interacted weakly but specifically with 14-3-3 proteins. It did not bind to GST alone or to the cytoplasmic tails of the other class I molecules including the classical class I tails which contain GKG instead of an RxR motif (Fig. 6,C). The RxR motif was confirmed to be responsible for binding to 14-3-3 proteins as its mutation to AAA inhibited the interaction of 14-3-3 with the HLA-F tail (Fig. 6 C).

The 14-3-3 proteins are extremely abundant proteins and <0.5% of the total cellular 14-3-3 bound to the cytoplasmic tail of HLA-F. As the 14-3-3-specific mAb used here cross-reacts with all seven isoforms of 14-3-3, the low levels bound to the cytoplasmic tails of HLA-F may reflect binding of only one isoform, or the competition between the many different proteins in the lysates that bind 14-3-3 (38). This requires further investigation.

In this study, we have found that while classical class I molecule can exit the ER in the absence of their cytoplasmic tail, the nonclassical class I molecule HLA-F is entirely dependent on its tail for export. The C-terminal valine is essential for ER export of HLA-F. Other immune system molecules such as CD8α (30), pro-TGFα (39, 40), MTI-MPP (40), in addition to other proteins (31, 41, 42) also require a C-terminal valine to be captured into COPII vesicles for ER-to-Golgi transport. Because C-terminal valine residues are found in ∼10% of human type I membrane proteins this feature may provide a general mechanism for ER export (33).

Once exported from the ER, the RxR motif, which is unique to HLA-F compared with other class I molecules, is responsible for its Golgi localization. The competition of binding between 14-3-3 proteins of the anterograde transport pathways with COPI of the retrograde transport system for the RxR motif is likely to control the recycling from the Golgi back to the ER (35, 36, 37). Interestingly, an interaction between an RxR motif present in the invariant chain component Lip35 with 14-3-3 proteins also controls anterograde transport of MHC class II (35, 43).

Various pathways of ER export have been suggested for class I molecules, depending on the affinity of loaded peptide, including specialized high-affinity ER exit sites and low-affinity escape sites (44). As has been shown here and by others, export of classical class I molecules from the ER via the Golgi to the surface is independent of their cytoplasmic tails (29). This suggests that for most class I molecules ER export is chiefly dependent on peptide binding. Entry into the cytoplasmic tail-independent pathway may be via a default exocytic pathway or may be mediated by a specific receptor such as BAP31 (20, 21).

In contrast, the export of HLA-F is different to peptide-loaded class I molecules. HLA-F uses a cytoplasmic-tail dependent pathway for forward transport. Although it is not known yet whether this pathway requires peptide binding, it is conceivable that export of HLA-F from the ER may be a peptide-independent pathway. Even though HLA-F can bind to TAP molecules (6, 7), it is not currently known whether it binds peptides in the ER. Peptide has never been eluted from the peptide binding groove of HLA-F and surface expression of HLA-F has recently been shown to be independent of both TAP and tapasin (9). Thus, ER export of HLA-F may be totally independent of peptide binding.

It is plausible that suboptimally loaded MHC class I molecule may use a similar exit pathway to HLA-F, using alternative ER exit signals encoded in their cytoplasmic tails. These will most likely be recycled back to the ER by tapasin and COPI (45) to allow another attempt at binding a high-affinity peptide and consequently export via a transport receptor. The importance of ER export motifs in HLA cytoplasmic tails may become apparent in particular circumstances where the normal pathway of export of MHC class I molecules becomes disrupted such as during infection with CMV and EBV. The targeting of residues in the cytoplasmic tail of class I by viral immune evasion products, such as HIV nef (23, 46), Kaposi’s sarcoma-associated herpes virus protein K3 (26, 27), and the HCMV US11 protein (47, 48), may have driven selection of the variation in the tails of class I molecules.

As yet the function of HLA-F remains unknown. Although all class I molecule traffic through the Golgi before being trafficked to the surface, the concentration of HLA-F in the Golgi is unique compared with other class I molecules. This pool of HLA-F in the Golgi may suggest an alternative function for this molecule. As most other MHC class I molecules bind peptides in the ER for cell surface recognition, it is possible that HLA-F has evolved to monitor the Golgi for infections. Intriguingly, endogenous HLA-F is detected on the surface of EBV transformed lymphoblastoid cell lines but not on B cells from the same individuals (9) implicating a role of viral infection or transformation in releasing HLA-F from the Golgi.

The sequence of HLA-F is similar to that of other class I molecules, and the hydrophobic transmembrane domain extends for at least 25 residues which would be consistent with a plasma membrane localization. It is longer than transmembrane domains known to cause Golgi retention (49, 50). Thus, we predict that under certain circumstances the Golgi pool of HLA-F could be exported for expression on the cell surface.

In conclusion, we have discovered an alternative ER export pathway used by the enigmatic HLA-F molecule. Only by investigating the export of HLA-F has a cytoplasmic tail-dependent pathway for anterograde transport of MHC class I molecules been revealed. The retention of HLA-F in the Golgi supports the idea of quality control for MHC class I molecules extending beyond the ER (24). The different mechanisms of export of class I molecule from the ER and Golgi will ultimately affect the nature and quantity of HLA displayed at the cell surface to T cells, NK cells, and other cells expressing class I receptors, such as LILR (51). This alternative mechanism of controlling surface expression of HLA-F is highly suggestive of a unique function compared with other class I molecules.

We thank Rainer Duden for useful discussions, Eric Lepin for the FG1 Ab, Paul Lehner, and Karin Romisch for critical reading of the manuscript.

The authors have no financial conflict of interest.