Abstract
Major histocompatibility complex (MHC) class I molecules present thousands of peptides to allow CD8+ T cells to detect abnormal intracellular proteins. The antigen-processing pathway for generating peptides begins in the cytoplasm, and the MHC molecules are loaded in the endoplasmic reticulum. However, the nature of peptide pool in the endoplasmic reticulum and the proteolytic events that occur in this compartment are unclear. We addressed these issues by generating mice lacking the endoplasmic reticulum aminopeptidase associated with antigen processing (ERAAP). We found that loss of ERAAP disrupted the generation of naturally processed peptides in the endoplasmic reticulum, decreased the stability of peptide–MHC class I complexes and diminished CD8+ T cell responses. Thus, trimming of antigenic peptides by ERAAP in the endoplasmic reticulum is essential for the generation of the normal repertoire of processed peptides.
Similar content being viewed by others
Main
Major histocompatibility complex (MHC) class I molecules present peptides on the cell surface for immune surveillance by the CD8+ T cell repertoire1. The peptides, derived from intracellular proteins, are used by CD8+ T cells to detect the presence of otherwise hidden viruses, bacteria or genetic mutations. Each polymorphic MHC class I molecule presents a large mixture of peptides with a distinct consensus motif2. The motif is defined by a length of eight to ten amino acids with conserved residues at the C terminus and at one or two internal positions. Yet the variability at the remaining six to eight positions ensures that millions of distinct peptides can be presented by any MHC class I molecule. The understanding of how degradation of cellular proteins produces these highly variable yet precisely cleaved sets of peptide–MHC (pMHC) class I complexes is incomplete.
MHC class I molecules are loaded in the endoplasmic reticulum (ER) with peptides that are generated in the cytoplasm3. Transport of antigenic peptides from the cytoplasm to the ER is accomplished by the transporter associated with antigen processing (TAP)4. In the cytoplasm, most peptides are generated by the multicatalytic proteasome, which can fragment newly synthesized but defective polypeptides5 or those destined for turnover6. Tripeptidyl peptidase II and a few other proteases have also been linked to the generation of some of these cytoplasmic peptides7. Whether these cytoplasmic proteases, acting alone or in concert, generate the final peptides before they enter the ER is not clear.
After the first suggestion that peptides may require trimming in the ER8, several lines of evidence subsequently indicated that antigenic precursors could be generated in the ER itself9. Notable among those were studies showing that TAP is unable to transport peptides containing a proline residue at position 2 (X-P-Xn, where 'P' is proline, 'X' is any amino acid and 'n' is a variable number of residues)10,11. Despite that serious handicap, up to 20% of MHC class I molecules actually present X-P-Xn peptides12. To resolve that paradox, it was suggested that X-P-Xn peptides could be transported as N-terminally extended intermediates that could be subsequently trimmed in the ER. Compelling evidence that peptide trimming is actually accomplished by an aminopeptidase came from studies examining the fate of antigenic peptides in isolated microsomes or in the ER of TAP-deficient cells9,13,14,15,16.
The aminopeptidase ERAAP ('ER aminopeptidase associated with antigen processing' in mice; human ortholog, ERAP1) had been identified in earlier work17,18. The same enzyme had been described as a member of the M1 family of zinc metalloproteases and had been linked to endocrine and anti-inflammatory processes19,20,21,22. In contrast to earlier studies, but consistent with its involvement in antigen processing, ERAAP was found to reside in the ER and its expression was found to be upregulated by interferon-γ (IFN-γ). Furthermore, as would be expected of an enzyme involved in generating a vast peptide repertoire, ERAAP also has a broad substrate specificity with the exception of peptides containing proline at position 2 (ref. 17). Notably, when ERAAP is 'knocked down' with small interfering RNA in tissue culture cells, expression of pMHC class I is inhibited17,23. However, the nature of the peptide pool that arrives in the ER and the function of ERAAP in generating the pMHC class I repertoire in normal antigen presenting cells is not known. To assess its physiological function(s), we generated mice lacking ERAAP by homologous recombination in embryonic stem cells. We show that ERAAP is a unique aminopeptidase in the antigen processing pathway that trims antigenic peptides in the ER to generate the normal pMHC class I repertoire.
Results
Generation of ERAAP-deficient mice
We generated mutant mice lacking ERAAP by homologous recombination in embryonic stem cells. In the mouse genome databases, the structure of the gene encoding ERAAP showed that it contained 18 exons distributed over 53 kilobases (kb). Because of its large size, eliminating the entire locus was impractical. We therefore chose to delete exons 4–8 because they encoded the highly conserved active site residues characteristic of aminopeptidases.
We excised the genomic sequences flanking those exons from bacterial artificial chromosome clones derived from the 129/J strain that was also the source of the embryonic stem cells. The structure of the targeting vector, the strategy and the final outcome of homologous recombination are presented in Figure 1a. In a litter of ten pups generated by the mating of two mice heterozygous for the mutated allele, three carried the homozygous mutation, as determined by Southern blot analysis of DNA obtained from tail tissue (Fig. 1b). As expected, in the same litter, other mice were either wild-type or heterozygous.
Immunoblot analysis of tail fibroblasts with or without IFN-γ treatment showed that in contrast to wild-type cells, in which ERAAP expression was strongly induced by IFN-γ, ERAAP was undetectable in mice with homozygous ERAAP deficiency (Fig. 1c). As positive controls, we assessed expression of the proteasome component LMP7, or H-2Kb MHC class I; this expression was upregulated by IFN-γ in both wild-type and ERAAP-deficient cells. We conclude that targeted deletion of exons 4–8 of the gene encoding ERAAP caused the loss of immunologically detectable ERAAP protein. Despite the loss of this enzyme, the ERAAP-deficient mice were fertile and did not show any apparent abnormalities.
ERAAP-deficient mice have reduced MHC class I
We assessed the expression of different MHC class I and class II molecules on the surfaces ERAAP-deficient splenocytes by flow cytometry. Each of these MHC molecules binds peptides with a different consensus motif and therefore presents a distinct peptide repertoire12. Expression of each of the five MHC class I molecules (H-2Kb, H-2Db, H-2Kd, H-2Dd and H-2Ld) was lower in ERAAP-deficient cells than in wild-type splenocytes (Fig. 2a). Notably, in contrast to the decrease of approximately 20% for the first four MHC molecules, H-2Ld expression decreased by approximately 70%. We noted this substantial reduction in H-2Ld expression with an antibody specific for peptide-bound H-2Ld (B22.249) or for its α3 domain (28-14-8S). Among the five MHC class I molecules, only the H-2Ld molecule presents peptides with the X-P-Xn motif defined by the presence of the proline at position 2. In contrast, expression of the H-2Ab, H-2Ad and H-2Ed MHC class II molecules was the same in wild-type and ERAAP-deficient cells (Fig. 2b). We also obtained similar results with cells from lymph nodes, thymus and bone marrow (data not shown). Furthermore, heterozygous H-2b mice were indistinguishable from wild-type mice, indicating that expression of a single allele is sufficient for its normal function. We conclude that the complete loss of ERAAP expression specifically inhibited surface expression of MHC class I but did not affect the expression of MHC class II molecules.
To confirm that the decrease in MHC class I expression was due to loss of ERAAP expression, we generated stable transfectants expressing mouse ERAAP or vector alone using a tail fibroblast cell line derived from ERAAP-deficient mice. We treated the cells with IFN-γ to upregulate the antigen-processing pathway and MHC molecules. In cells expressing vector alone, expression of H-2Kb as well as H-2Db remained low, but this increased more than 150% in cells expressing mouse ERAAP (Fig. 2c). Thus, ERAAP expression was essential for optimal surface expression of MHC class I in both resting and IFN-γ-treated cells.
ERAAP determines peptide quality for MHC class I
Expression of pMHC class I on the cell surface depends on the quantity as well as quality of the peptide supply. A change in either attribute could account for the decrease in pMHC class I expression in ERAAP-deficient mice. Newly synthesized MHC class I molecules are retained in the ER until they acquire peptides and exit toward the cell surface24. If the quantity of peptides available were compromised in ERAAP-deficient cells, it would cause a lag in the assembly of pMHC class I complexes. We tested this hypothesis by pulse-labeling splenocytes with radiolabeled [35S]methionine. After 'chasing' with unlabeled methionine for various times, we immunoprecipated H-2Kb or H-2Db molecules from the cell lysates (Fig. 3a,b). To distinguish empty MHC class I molecules in the ER from peptide-bound molecules that had left the ER, we treated the immunoprecipitates with endoglycosidase H. This enzyme removes high-mannose oligosaccharides from ER proteins but not from proteins that have traversed to the Golgi and undergone further oligosaccharide modifications. The maturation rate of H-2Kb and H-2Db molecules was similar in wild-type and ERAAP-deficient cells (Fig. 3a,b). Thus, ERAAP did not affect the rate of peptide acquisition or the trafficking of pMHC class I.
To extend that observation to MHC expressed on the cell surface, we treated IFN-γ-stimulated ERAAP-deficient fibroblasts with mild acid to disassociate pre-existing pMHC class I complexes on the cell surface. We then cultured cells in normal medium for various times to allow newly assembled H-2Kb and H-2Db molecules to arrive at the cell surface. Measurement of these MHC molecules by flow cytometry showed that at each time point, cells expressing vector only had a lower mean fluorescence intensity than those expressing ERAAP (data not shown). However, the rate of MHC recovery calculated as a percentage of the maximum expression on vector+ or ERAAP+ cells was identical (Fig. 3c). Thus, the reduction in MHC class I expression due to ERAAP deficiency was not due to a slower rate of assembly of new pMHC class I complexes.
Next we examined the quality of MHC-bound peptides on the cell surface by measuring the dissociation of preformed pMHC class I complexes, which reflects their peptide-dependent stability. We cultured spleen cells in brefeldin A to block the exit of newly assembled pMHC class I from the ER and stained the cells with antibodies to MHC class I after various time intervals. For each MHC molecule tested, pMHC class I disassociation was faster on the surface of ERAAP-deficient cells than on wild-type cells (Fig. 3d). The decrease in MHC expression was rapid within the first 2 h, and the disassociation rate during this period was 10–20% faster in ERAAP-deficient cells, suggesting that a greater fraction of surface MHC class I in ERAAP-deficient cells was bound to suboptimal peptides. Notably, the dissociation of pMHC class I was particularly rapid for H-2Ld molecules. ERAAP-deficient cells lost 40% of surface H-2Ld molecules within the first hour, compared with a loss of 15% for wild-type cells (Fig. 3e). Thus, ERAAP is key in optimizing the peptide repertoire that confers stability to pMHC class I complexes rather than limiting the rate of pMHC class I assembly.
Altered pMHC class I in ERAAP-deficient antigen-presenting cells
To assess the function of ERAAP in generating individual pMHC class I complexes, we used ERAAP-deficient splenocytes as antigen-presenting cells (APCs) for a panel of CD8+ T cell hybridomas. The peptides recognized by these T cells are derived from various endogenous genes expressed in the C57BL/6 (B6) mouse strain. The expression of two peptides presented by H-2Db MHC class I and recognized by the LPAZ (H3a–H-2Db) and 11p9Z (Uty–H-2Db) T cell hybridomas was substantially lower in ERAAP-deficient cells (Fig. 4a). To quantify that decrease and to determine if any structural changes had occurred in the naturally processed peptides, we extracted them from spleen cells and fractionated the peptide mixture by reverse-phase high-performance liquid chromatography (HPLC). We detected antigenic peptides using the cognate T cell hybridoma and appropriate APCs. We did not detect the Uty-encoded 11p9Z-stimulating peptide WI9 in ERAAP-deficient cell extracts (Fig. 4b, top), directly confirming that the generation of naturally processed WI9 peptide was absolutely dependent on ERAAP. In contrast, expression of two unidentified peptides presented by H-2Kb to 18.5Z and 27.5Z hybridomas remained unchanged. Most notably, the peptides recognized by 1AZ (H47–H-2Db) and 30NXZ (H13–H-2Db) T cell hybridomas were increased substantially in ERAAP-deficient splenocytes. HPLC analysis of peptide extracts showed that the H13-encoded 30NXZ peptide SVL9 was increased about 100-fold in ERAAP-deficient splenocytes (Fig. 4b, bottom). Notably, the HPLC elution profile of the SVL9 peptide in ERAAP-deficient cell extracts was identical to that in wild-type cells. Thus, the absence of ERAAP resulted in a large increase in the total quantity of SVL9 peptide rather than a change in its structure. We obtained the same peptide profile when we used bone marrow–derived dendritic cells as APCs (data not shown). We conclude that absence of ERAAP disrupts the pMHC class I repertoire in normal and 'professional' APCs.
To further investigate the function of ERAAP in generating processed peptides from intracellular proteins, we introduced a modified ovalbumin precursor (KOVAK) into ERAAP-deficient splenocytes by retroviral transduction25. The KOVAK precursor is a cytoplasmic protein that contains the SHL8 octamer peptide presented by H-2Kb to the B3Z T cell hybridoma26. Studies of the antigen processing of KOVAK have shown that N-terminally extended forms of the final antigenic peptide are generated in the cytoplasm25. We hypothesized that if the proteolytic intermediates of KOVAK were supplied to the ER, they would require ERAAP for generation of the final SHL8 peptide. In agreement with that hypothesis, B3Z T cells responded strongly to wild-type cells transduced with KOVAK, but their response to similarly transduced ERAAP-deficient cells was one tenth their response to wild-type (Fig. 5a). Analysis of processed peptides in HPLC-fractionated extracts from wild-type cells showed the presence of two distinct peaks corresponding to the SHL8 octamer and the N-terminally extended K-SHL8 nonamer peptides. These peptides represent the SHL8–H-2Kb and K-SHL8–H-2Db complexes, as they contain the appropriate H-2Kb (SIINFEHL) and H-2Db (KSIINFEHL) binding motifs (underlined)27. In contrast, similar analysis of ERAAP-deficient cells showed that the SHL8 peptide presented by H-2Kb was barely detectable, whereas the K-SHL8 peptide presented by H-2Db was still present (Fig. 5a, right). There was about 95% less SHL8 peptide and about 30% less K-SHL8 peptide in ERAAP-deficient versus wild-type cells. As a positive control, we also transduced the spleen cells with a retrovirus encoding SHL8 as a minimal precursor that does not require any processing (discussed below). In contrast to results obtained with the full-length protein, presentation of SHL8–H-2Kb to B3Z T cells and the amount of processed SHL8 peptide in cell extracts was similar in both wild-type and ERAAP-deficient cells expressing the minimal precursor (Fig. 5b). These results provide direct evidence that expression of ERAAP was essential for the generation of pMHC class I from the full-length cytoplasmic KOVAK precursor but not from the minimal SHL8 peptide.
ERAAP is a unique aminopeptidase in the ER
Next we sought to determine if the effects of ERAAP deficiency were due to its proteolytic activity in the ER compartment itself and if it was a unique protease in the antigen-processing pathway. To exclude the possibility of cytoplasmic processing events, we used splenocytes from mice lacking both ERAAP and TAP. We introduced antigenic precursors directly into the ER by appending them to an ER-translocation signal sequence (ES). Because TAP is absent, processed peptides from the cytoplasm cannot be presented by the cells, and therefore the processing of the precursors to their final antigenic form is solely dependent on proteolysis in the ER9. We transduced splenocytes with retroviral vectors expressing one of two precursors: the first, (ES)-SHL8, could yield the final antigenic peptide after cleavage of the ER-translocation signal sequence by signal peptidase alone; the other required removal of seven N-terminal residues ((ES)-AIVMQLK[SHL8]) for generation of the final SHL8 peptide. When used as APCs, ERAAP+ cells expressing the N-terminally extended precursor were at least 1,000 times more efficient than ERAAP-deficient cells in activating B3Z T cells (Fig. 6a). Peptide extracts from ERAAP+ cells contained both K-SHL8 and SHL8 peptides, but neither peptide was detected in extracts from ERAAP-deficient cells (Fig. 6a, right), indicating that trimming of antigenic precursors was completely blocked. Notably, ERAAP deficiency did not affect surface presentation of SHL8–H-2Kb from the (ES)-SHL8 precursor in ERAAP-deficient cells, and peptide extracts from these and ERAAP+ cells contained equivalent amounts of SHL8 peptide (Fig. 6b). We conclude that ERAAP acts on antigenic precursors in the ER itself and is required only when the precursor peptide contains additional residues flanking its N terminus. Most notably, as the absence of ERAAP could not be compensated by another protease, it serves a unique function in the antigen processing pathway.
ERAAP limits presentation of some peptides in the ER
Because ERAAP removed only the extra N-terminal residues from the final SHL8 peptide presented by MHC class I, we sought to determine why its absence resulted in upregulation of other peptides such as the H13-encoded SVL9 peptide (Fig. 4a). We considered two potential explanations. The increase in SVL9 peptide in ERAAP-deficient cells could have resulted from a lack of competition from other peptides that are normally present in wild-type cells but are missing in ERAAP-deficient cells. Alternatively, SVL9 could have been mostly destroyed by ERAAP in wild-type cells and therefore could have accumulated in cells deficient in ERAAP.
To test those hypotheses, we used fibroblasts deficient in both TAP and ERAAP because they lack the normal pool of cytoplasmic peptides transported by TAP and thus provide an ER environment free of most competing peptides. We transfected stable fibroblasts expressing vector alone or ERAAP with the (ES)-SVL9 construct. This construct was similar to the (ES)-SHL8 construct described above and encoded the ER-translocation signal sequence from hen egg lysozyme followed by the nine residues of the SVL9 peptide. Removal of the ER-translocation signal sequence by signal peptidase would generate the final SVL9 peptide in the ER, which could then be loaded onto H-2Db or could be a substrate for ERAAP. After 2 d, we assayed peptide extracts from these and control, untransfected cells for the presence of SVL9 peptide using H-2Db-expressing L cells (H-2Db–L cells) as APCs and SVL9–H-2Db–specific 30NXZ T cells (Fig. 7a). In contrast to untransfected cells, in the extracts of ERAAP-deficient cells expressing (ES)-SVL9 and vector, the SVL9 peptide was readily detected. However, the amount of SVL9 peptide in cells expressing (ES)-SVL9 and ERAAP was one tenth that in ERAAP-deficient cells. This result demonstrates that presence of ERAAP was deleterious to the SVL9 peptide and explains why expression of this peptide was enhanced in its absence.
To further confirm the function of ERAAP in limiting the presentation of the SVL9 peptide, we determined the rate at which the SVL9–H-2Db complex was generated from (ES)-SVL9 in TAP-deficient cells with or without ERAAP. We 'stripped' the pre-existing SVL9–H-2Db complexes from cells with a mild acid wash and then allowed the cells to generate new SVL9–H-2Db complexes. We extracted peptides after various time intervals and assayed the amount of SVL9 as described above. The SVL9 peptide was lost after the acid wash but was regenerated within 6 h in cells expressing the (ES)-SVL9 construct and vector (Fig. 7b). Cells coexpressing ERAAP also generated the SVL9 peptide, but its amount remained low at all time points and it was generated at a rate less than half that of cells expressing vector alone (Fig. 7b). Again, in cells expressing the (ES)-SHL8 construct, the presence or absence of ERAAP had no effect on the amount of SHL8 peptide generated in the cells (Figs. 7c and 6b). Thus, ERAAP inhibited the presentation of only the SVL9 peptide but not the SHL8 peptide when the peptides entered the ER in their final form. We conclude that ERAAP demonstrates substrate preferences in the ER that can limit the generation of some pMHC class I complexes.
ERAAP deficiency disrupts the CD8+ T cell repertoire
Expression of endogenous pMHC class I is essential for the generation and maintenance of the normal CD8+ T cell repertoire28. Because ERAAP-deficient mice displayed an altered peptide repertoire as well as differences in the stability of endogenously generated pMHC class I complexes, we assessed the effect of ERAAP deficiency on the T cell repertoire. The total numbers of B lymphocytes and T lymphocytes as well as those in the CD4 and CD8 subsets were similar in the thymuses, spleens and lymph nodes of ERAAP-deficient and wild-type mice (data not shown). Furthermore, we detected no differences among T cells bearing T cell receptors with specific variable β-chain in thymuses or spleens of wild-type and ERAAP-deficient mice (data not shown). To assess whether the T cell repertoire in ERAAP-deficient mice may nevertheless differ in its antigen-specific responsiveness, we tested responses of female mice to immunogenic peptides derived from the HY antigens encoded on the Y chromosome in male cells29. We immunized female ERAAP-deficient mice on the B6 background with splenocytes from wild-type B6 male mice. After 10 d, we expanded the antigen-specific CD8+ T cell populations in the spleens of responding mice in vitro by culture together with irradiated splenocytes from wild-type B6 male. We counted peptide-specific IFN-γ-secreting CD8+ T cells by intracellular staining and flow cytometry (Fig. 8). In contrast to the robust responses in wild-type mice, ERAAP-deficient mice were notably poor in eliciting CD8+ T cell responses to pMHC class I complexes derived from either Uty (about 70%) or Jarid1d (also called Smcy; 80–90% deficient). We noted the same proportional differences between wild-type and ERAAP-deficient mice ex vivo (data not shown). In contrast, the H-2Ab MHC class II–restricted CD4+ T cell response to the peptide encoded by Ddx3y (also called Dby) was similar in ERAAP-deficient and wild-type mice. Thus, CD8+ but not CD4+ T cell responses to HY antigens were diminished in ERAAP-deficient mice.
Discussion
Study of ERAAP-deficient mice has shown that N-terminal trimming of antigenic peptides in the ER is essential for the generation of the normal repertoire of peptides presented by MHC I molecules and for normal CD8+ T cell responses. Our findings have provided insights into the nature of the processed peptide pool and how it is shaped in the ER by this aminopeptidase in the MHC class I antigen-processing pathway. The ERAAP aminopeptidase was independently identified in rats, humans and mice and has been given a variety of names19,20,21,22. Those studies suggested, based on its ability to trim peptide hormones and its potential function in shedding the tumor necrosis factor receptor, that this aminopeptidase could be involved in the endocrine system and anti-inflammatory responses. In contrast, subsequent studies of mouse and human ERAAP linked it to the MHC class I antigen-processing pathway17,23. Our observations of ERAAP-deficient mice on different mixed and inbred backgrounds have shown that the mice with homozygous deficiency were fertile with no obvious abnormalities and had a lifespan similar to that of their heterozygous and wild-type counterparts. Although the possibility of subtle effects cannot yet be ruled out, it is unlikely that ERAAP has nonimmunological functions essential for viability.
Another aminopeptidase similar to ERAAP has been identified in humans30,31. That enzyme, L-RAP, shares about 50% amino acid sequence identity with ERAAP, is also located in the ER and is upregulated by IFN-γ. In vitro assays have shown that L-RAP is capable of trimming N termini of synthetic substrates, albeit with a distinct specificity. Moreover, ERAAP and L-RAP can form homodimers and heterodimers31. Most notably, when expression of these aminopeptidases is 'knocked down' individually or together by RNA interference, the cells are inhibited in their IFN-γ-induced pMHC class I expression and APC function. Thus, it is possible that in humans, both ERAAP and L-RAP function together in the antigen-processing pathway. The relevance of the existence of two ER aminopeptidases in humans is not yet known, but a mouse ortholog of L-RAP has not been found in the sequence databases. Most notably, our findings here have demonstrated that the absence of ERAAP cannot be compensated for by another enzyme and therefore provide unequivocal evidence that ERAAP serves a unique antigen-processing function in mice.
Cell surface expression of each of the five MHC class I molecules tested was lower in ERAAP-deficient cells. Among those, H-2Ld suffered the most profound loss of surface expression, similar to other observations in cell lines treated with aminopeptidase inhibitors or small interfering RNA directed against ERAAP9,17. Unlike the other MHC I molecules tested, only H-2Ld molecules present peptides with the X-P-Xn motif32. Because X-P-Xn peptides are poorly transported by TAP, it has been suggested that their precursors would arrive in the ER as N-terminally extended analogs and would only then be trimmed to their final X-P-Xn form10,11. The profound requirement for ERAAP for the expression of H-2Ld MHC class I in mice has now directly confirmed that original hypothesis and has established ERAAP as the unique aminopeptidase that serves this function. Notably, about 20% of human MHC class I molecules are also known to present X-P-Xn peptides12. As discussed above, humans express both ERAAP and L-RAP as N-terminal trimming enzymes in the ER, and it remains to be determined which one or both can generate the unique pools of X-P-Xn peptides.
Expression of MHC class I on the cell surface depends on the assembly of stable pMHC class I complexes and their exit from the ER33. When peptide supply is blocked by inhibition of cytoplasmic proteases or by TAP activity, pMHC class I expression on the cell surface is strongly inhibited. Notably, MHC class I expression is also reduced in cells lacking tapasin because of its function in stabilizing TAP and optimizing the loading of peptides onto the MHC class I in the ER34,35,36. In ERAAP-deficient cells, the reduction in MHC class I expression was due to faster dissociation of pMHC class I from the cell surface rather than to a slower rate of pMHC class I assembly in the ER. Therefore, cells lacking ERAAP are not deficient in peptide supply but instead are deficient in the composition of the peptide pool. Our results suggest that ERAAP trimming generates optimal peptides that yield stable pMHC class I complexes. It is likely that in the absence of ERAAP, some pMHC class I molecules on the cell surface contained weakly bound, N-terminally extended peptides.
Analysis of individual pMHC class I complexes in ERAAP-deficient cells has provided new insights regarding the pool of naturally processed peptides. We noted three distinct sets of peptides among the pMHC class I generated in ERAAP-deficient cells. Expression of one set of peptides was profoundly decreased, whereas expression of another set remained unchanged in wild-type and ERAAP-deficient cells. We established the characteristics of these two peptide sets by analyzing their processing in cells deficient in both TAP and ERAAP. In the ER of TAP-deficient cells, precursors with extra N-terminal residues flanking the final peptides required ERAAP for trimming, whereas those that did not contain any extra residues were independent of ERAAP. A similar ERAAP requirement is also evident in TAP+ cells for the processing of full-length KOVAK protein, which has been shown to yield N-terminally extended intermediates in the cytoplasm25. Thus, in addition to the X-P-Xn peptide pool, other peptides arriving in the ER with N-terminal extensions also require ERAAP for generation of the final pMHC class I complexes, whereas those without N-terminal flanking residues continue to be presented in absence of ERAAP. This also indicates that some peptides arriving in the ER would have been generated in their final form in the cytoplasm by the proteasome, tripeptidyl peptidase II or other proteases7. Unfortunately, the identity of the final peptides or their precursor proteins that were unchanged in ERAAP-deficient cells is not known. Therefore, the nature of their cytoplasmic intermediates and the proteolytic mechanisms that generate such peptides remain to be defined.
Notably, we also identified a third set of peptides whose abundance and presentation was enhanced considerably in the absence of ERAAP (H13 and H47). Analysis of the fate of the final peptides in the ER of TAP-deficient cells showed that ERAAP destroyed the H13-encoded SVL9 peptide, whereas it had no effect on the SHL8 peptide. This apparent substrate preference of ERAAP was not due to differences in the N-terminal amino acid, as both peptides begin with a serine residue. Furthermore, the difference in the size of the SHL8 octamer peptide versus the SVL9 nonamer peptide is also unlikely to account for their different fates, because the K-SHL8 nonamer peptide presented by the same H-2Db MHC class I was not destroyed by ERAAP. Thus, the proposal that ERAAP acts as a molecular 'ruler' to generate the set of final peptides of the correct length does not apply to these peptides18,23. How ERAAP chooses its substrates and determines their final fate therefore remain open issues.
Our findings have shown that absence of ERAAP resulted in disruption of the normal peptide repertoire displayed by MHC class I and that only a fraction of the peptide pool was unaffected by ERAAP deficiency. Even given the small number of individual peptides examined, it is clear that some peptides are missing, whereas others have much higher expression. If the random set of peptides analyzed here can be considered to be representative of all naturally processed peptides, it follows that ERAAP deficiency results in a pMHC class I repertoire that is skewed toward over-representation of a few peptides, which could mask the absence of many others.
The development, maintenance and responsiveness of CD8+ T cells depends on the expression of endogenous pMHC class I complexes28. Although the total number of CD8+ T cells and the distribution of T cell receptors with specific variable β-chain in ERAAP-deficient mice were similar to those of wild-type mice, their responses to the pMHC class I derived from HY antigens was inhibited considerably. Whether this deficiency in CD8+ T cell response was due to a defect in generating appropriate pMHC class I or a 'hole' in the CD8+ T cell repertoire is unclear. Although the cells from wild-type B6 male mice used as the immunogen do express the HY-derived pMHC class I, it is conceivable that the APCs in the ERAAP-deficient host may have been unable to cross-present the donor-derived HY antigens. That hypothesis is consistent with findings regarding the function of ER in the presentation of exogenous antigens37. Alternatively, the disrupted self peptide–MHC class I repertoire could have altered the CD8+ T cell repertoire and thus could have diminished the ability of ERAAP-deficient mice to elicit HY-specific CD8+ T cell responses. In conclusion, we have established that ERAAP is the aminopeptidase that makes the 'final cut' in antigenic precursors in the ER. This event is key to shaping the normal pMHC class I repertoire and to eliciting normal CD8+ T cell responses.
Methods
Mice.
Use of all mice was done with approval of the Animal Care and Use Committee of the University of California at Berkeley (Berkeley, California). ERAAP-deficient mice were generated by homologous recombination. Genomic clones of the mouse ERAAP were isolated from an RPCI-22 mouse bacterial artificial chromosome library (ResGen), with mouse cDNA encoding ERAAP as a probe17. The targeting vector (Fig. 1a) was constructed by amplification of DNA for the 5′ arm by PCR with the following primers: 5′ arm forward, 5′-CCCTTCACAAACTGTTCTCACCC-3′, and reverse, 5′-GCAGCATCCAGAGCATAATCGG-3′. The PCR product was digested with EcoRI and the resulting 2.8-kb fragment was cloned into pKO915 Scrambler (Lexicon Genetics). Similarly, the 3′ arm was derived by amplification of a region containing two internal HindIII sites with primers 5′-CAGGCTCAGTAAGATGGCTGTCTG-3′ (forward) and 5′-AGTTTGCGTGTCCAATGGGC-3′ (reverse). The product was digested with HindIII and the resulting 2.5-kb fragment was cloned downstream of the 5′ arm. For negative selection, a diptheria toxin A fragment (Lexicon Genetics) was cloned upstream of the 5′ arm into an RsrII site. For positive selection, loxpNeo (Institut fur Genetik, Cologne, Germany) was inserted into the BamHI site between the 5′ and 3′ arms. The construct was then made linear and was electroporated into TC1 embryonic stem cells at the University of California at Berkeley Pathogen-free Transgene facility as described38. Of 108 clones screened, two contained the correctly targeted Arts1 locus (encoding ERAAP) and were used to generate ERAAP-deficient mice using standard procedures. Both mouse lines with homozygous ERAAP deficiency had identical genotypes and one (line number 30) was used for further breeding. Wild-type B6, B6.129S2-Abcb2tm1Arp (TAP1-deficient) and B10.D2-Hc1H2dH2-T18c/nSnJ mice were obtained from the Jackson Laboratory. ERAAP-deficient mice on the B6 background were obtained by backcrosses with wild-type B6 mice for at least five generations. H-2b mice deficient in both TAP and ERAAP or H-2d mice deficient in ERAAP alone were obtained by breeding of ERAAP-deficient mice with TAP-deficient or B10.D2 mice, respectively.
Antibodies and flow cytometry.
Unless noted otherwise, all antibodies were purchased from BD Biosciences. Spleen cells were first blocked with antibody to Fc receptor (anti–Fc receptor; 2.4G2) and then were stained with fluorescein isothiocyanate–conjugated antibodies to MHC specific for H-2Kb (AF6-88.5), H-2Db (KH95), H-2Kd (SF1-1-1), H-2Dd (34-5-8), H-2Ld (28-14-8), H-2Ab and H-2Ad (25-9-17), or H-2Ed (14-4-4). For detection of H-2Ld expression, cell suspensions were first depleted of B lymphocytes with anti-mouse IgG Dynabeads (Dynal Biotech) and then were stained with the peptide-dependent B22.249 monoclonal antibody to H-2Ld. Fluorescein isothiocyanate–conjugated anti-mouse IgG (Cappel) was used as the secondary antibody. Expression of H-2Kb on fibroblasts was detected with monoclonal antibodies Y3, 5F1 and EH.144. H-2Db was detected with monoclonal antibodies B22.249, KH95 and CTDb (Caltag Laboratories). Phycoerythrin-conjugated anti-mouse IgG (Caltag Laboratories) was used as the secondary antibody. Cells were analyzed by flow cytometry with a FACScan (Coulter) and data were analyzed with FlowJo software (TreeStar).
Cell lines and T cell activation assays.
Immortal fibroblasts cell lines were generated as described39 from wild-type mice, ERAAP-deficient mice and mice deficient in both TAP and ERAAP. Fibroblasts stably expressing cDNA encoding ERAAP or vector DNA were generated by retroviral transduction with murine stem cell virus (MSCV; described below) and were sorted for equivalent expression of green fluorescent protein (GFP). Culture conditions and the generation of β-galactosidase (lacZ)–inducible T cell hybridomas B3Z, 11p9Z, 30NXZ, 1AZ, LPAZ, 27.5Z and 18.5Z and H-2Kb- or H-2Db-expressing L cells have been described9. The pMHC class I complexes on splenocytes or peptides in cell extracts were detected by overnight culture of H-2Db–L cells or H-2Kb–L cells as APCs together with lacZ-inducible T cell hybridomas. Intracellular lacZ was measured with the substrate chlorophenolred-β-D-galacto-pyrannoside (Sigma) by its absorbance at 595 nm and 655 nm as reference.
Immunoblots, pMHC class I assembly and surface stability.
Expression of ERAAP, LMP7 and H-2Kb in wild-type or ERAAP-deficient fibroblasts was determined by immunoblot as described17. The rate of pMHC class I assembly was determined by immunoprecipitation analysis of metabolically labeled cells. Spleen cells (5 × 107) from mice with homozygous or heterozygous ERAAP deficiency were pulse-labeled for 7 min at 37 °C with 65.5 MBq/ml of [35S]methionine in methionine-free RPMI medium (Sigma) plus 5% dialyzed FCS (Hyclone). Cells were then diluted in 10 ml normal medium containing 10 mM unlabeled methionine, were pelleted and were resuspended at a density of 6 × 105 cells/ml. At each time point, 6 × 106 cells were placed on ice, and at the final time point, cells were lysed for 30 min at 4 °C with constant rotation in 1 ml Tris-buffered saline (150 mM NaCl and 10 mM Tris, pH 8.0) plus 0.5% Triton X-100 (Roche) containing protease inhibitors (5 mM iodoacetimide, 10 μM MG132, 0.5 mM phenylmethylsulphonylfluoride, 5 mM 1-10 phenanthroline, 2 μg/ml of aprotonin and 10 μM pepstatin A). All subsequent steps were done at 4 °C. After pelleting of debris and preclearing of the supernatant for 5 h, MHC class I molecules were immunoprecipitated for 1 h with protein G–Sepharose beads preconjugated to anti-H-2Kb (Y3) or anti-H-2Db (B22.249). After immunoprecipitation, beads were washed once with 1 ml 0.5% Triton X-100 in Tris-buffered saline, twice with 0.1% Triton X-100 in Tris-buffered saline and once with water. Bead pellets were resuspended in 40 μl of 4 × SDS loading buffer and were heated for 10 min at 70 °C before beads were 'spun out', followed by digestion of the supernatant with endoglycosidase H (New England Biolabs) and separation by 8% SDS-PAGE. After electrophoresis, gels were fixed, treated with Amplify (Amersham Biosciences) and dried before overnight exposure to a Storage Phosphor Screen (Amersham Biosciences) and subsequent exposure to film. Bands corresponding to molecular weights of H-2Kb or H-2Db molecules were quantified with ImageQuant Software version 1.2 (Molecular Dynamics). For measurement of the rate of pMHC class I expression at the cell surface, ERAAP-deficient fibroblasts stably expressing ERAAP or vector alone were treated with IFN-γ as described above. After 48 h, plate-bound cells were treated with 0.131 M sodium citrate and 0.066 M sodium phosphate, pH 3.1, to denature surface pMHC class I and then were cultured for various times in normal medium9. For measurement of the disassociation of pMHC class I at the cell surface, splenocytes were cultured for various times at 37 °C in medium containing 5 μg/ml of brefeldin A (Sigma), were stained with anti–MHC class I and were analyzed as described above. Dead cells were excluded from analysis by propium iodide staining.
Peptides, peptide extracts and HPLC analysis.
Peptides of various sequences were synthesized by D. King (University of California at Berkeley) and their structures were confirmed by mass spectrometry. Endogenous peptides were extracted from 1 × 108 spleen cells cultured overnight in medium containing 10 μg/ml of lipopolysaccharide (LPS; Sigma) and 5 μg/ml of concanavalin A (Sigma). These or other transduced cells were lysed in 10% acetic acid supplemented with 10 μM irrelevant peptide, were boiled for 5 min and were passed through a 10-kilodalton-cutoff filter (Millipore). The filtrate was then fractionated on a 2.1-mm × 250-mm C18 column (Vydac) over a gradient of 5–40% acetonitrile. Five-drop fractions were collected in 96-well plates, were dried and were analyzed with T cell hybridomas and H-2Kb–L cells or H-2Db–L cells as APCs. Synthetic peptides and buffer alone were analyzed in identical conditions to establish their retention times and the absence of cross-contamination between samples.
DNA constructs, transfections and retroviral transduction.
The retroviral MSCV-IRES GFP vector was a gift from W. Sha (University of California at Berkeley)40. KOVAK, cDNA encoding ERAAP, and the (ES)-SHL8 and (ES)-AIVMQLK[SHL8] constructs have been described17,25 and were cloned into the BamHI, XhoI-NotI or XhoI-ClaI sites of the MSCV vector. The (ES)-SVL9 construct was generated from oligonucleotides inserted into the BamHI-NotI sites of pcDNAI vector and encodes the ER-translocation signal sequence from chicken egg lysozyme with some modifications. This construct encodes the amino acid sequence MLGKNDPMCLVLVLLAATAIASSVVGVWYL. Bosc23 cells, the generation of MSCV retrovirus and splenocyte infection were done as described40. Bosc23 cells were transfected with 10 μg DNA using ExGen 500 transfection reagent (MBI Fermentas) and supernatants were collected at 48 h. LPS-stimulated cells were 'spin-infected' at 1,000g for 2 h in 2 ml viral supernatant (plus 8 μg/ml of polybrene) per 2 × 106 cells. Cells were then cultured for 48 h in medium containing 30 μg/ml of LPS (Sigma). The percent of cells that were transduced (about 10%), identified by measurement of GFP by flow cytometry, did not vary much for spleen cell samples from different mice. For each population, the same number of GFP+ cells were used as APCs or for peptide extracts and HPLC as described above. Fibroblasts deficient in both TAP and ERAAP and stably expressing ERAAP or vector alone were transfected with 3 μg DNA using Fugene transfection reagent (Roche). The transfection efficiency was consistently 30% and did not vary between cell types. After 48 h, peptides were extracted as described above, were dried overnight and were analyzed with T cell hybridomas and H-2Kb–L cells or H-2Db–L cells as APCs.
Immunizations and IFN-γ assay.
ERAAP-deficient mice on a B6 background (N5; fifth back-cross) and B6 female mice were immunized intraperitoneally with 20 × 106 splenocytes from B6 male mice. At day 10, spleens were collected and were restimulated with irradiated splenocytes from B6 male mice in medium containing 20 U/ml of recombinant human interleukin 2 (BD Biosciences). After 7 d, live cells were recovered and 1 × 106 cells were incubated in medium containing brefeldin A and 0.1 μM Smcy peptide (KCSRNRQYL) or Uty peptide (WMHHNMDLI) or 10 μM Dby peptide (NAGFNSNRANSSRS) as described for intracellular cytokine staining (BD Biosciences). Cells were stained with phycoerythrin-conjugated anti-CD8 or anti-CD4 and were fixed and permeabilized before being stained with fluorescein isothiocyanate–conjugated anti–mouse IFN-γ. Cells were then analyzed by flow cytometry for IFN-γ+ T cell populations.
Accession codes.
GenBank: Arts1 (ERAAP) locus, AF227511; H3a, AF060243; H13, AF017785; H47, NM024439; Uty, NM_009484; Jarid1d (Smcy), NM_011419 XM_489864; Dbx3y (Dby), NM_012008.
References
Shastri, N., Schwab, S. & Serwold, T. Producing nature's gene-chips. The generation of peptides for display by MHC class I molecules. Annu. Rev. Immunol. 20, 463–493 (2002).
Falk, K., Rotzschke, O., Stevanovic, S., Jung, G. & Rammensee, H-G. Allele-specific motifs revealed by sequencing of self-peptides eluted from MHC molecules. Nature 351, 290–296 (1991).
Townsend, A.R., Bastin, J., Gould, K. & Brownlee, G.G. Cytotoxic T lymphocytes recognize influenza haemagglutinin that lacks a signal sequence. Nature 324, 575–577 (1986).
Attaya, M. et al. Ham-2 corrects the class I antigen-processing defect in RMA-S cells. Nature 355, 647–649 (1992).
Yewdell, J.W., Reits, E. & Neefjes, J. Making sense of mass destruction: quantitating MHC class I antigen presentation. Nat. Rev. Immunol. 3, 952–961 (2003).
Rock, K.L., York, I.A., Saric, T. & Goldberg, A.L. Protein degradation and the generation of MHC class I-presented peptides. Adv. Immunol. 80, 1–70 (2002).
Kloetzel, P.M. Generation of major histocompatibility complex class I antigens: functional interplay between proteasomes and TPPII. Nat. Immunol. 5, 661–669 (2004).
Falk, K., Rötzschke, O. & Rammensee, H.G. Cellular peptide composition governed by major histocompatibility complex class I molecules. Nature 348, 248–251 (1990).
Serwold, T., Gaw, S. & Shastri, N. ER aminopeptidases generate a unique pool of peptides for MHC class I molecules. Nat. Immunol. 2, 644–651 (2001).
Neisig, A. et al. Major differences in transporter associated with antigen presentation (TAP)-dependent translocation of MHC class I-presentable peptides and the effect of flanking sequences. J. Immunol. 154, 1273–1279 (1995).
Van Endert, P.M. et al. The peptide-binding motif for the human transporter associated with antigen processing. J. Exp. Med. 182, 1883–1895 (1995).
Rammensee, H.G., Bachmann, J. & Stevanovic, S. in MHC Ligands and Peptide Motifs, 1–457 (Landes Bioscience, Austin, Texas, 1997).
Lauvau, G. et al. Human transporters associated with antigen processing (TAPs) select epitope precursor peptides for processing in the endoplasmic reticulum and presentation to T cells. J. Exp. Med. 190, 1227–1240 (1999).
Fruci, D., Niedermann, G., Butler, R.H. & van Endert, P.M. Efficient MHC class I-independent amino-terminal trimming of epitope precursor peptides in the endoplasmic reticulum. Immunity 15, 467–476 (2001).
Brouwenstijn, N., Serwold, T. & Shastri, N. MHC class I molecules can direct proteolytic cleavage of antigenic precursors in the endoplasmic reticulum. Immunity 15, 95–104 (2001).
Mo, X.Y., Cascio, P., Lemerise, K., Goldberg, A.L. & Rock, K. Distinct proteolytic processes generate the C and N termini of MHC class I-binding peptides. J. Immunol. 163, 5851–5859 (1999).
Serwold, T., Gonzalez, F., Kim, J., Jacob, R. & Shastri, N. ERAAP customizes peptides for MHC class I molecules in the endoplasmic reticulum. Nature 419, 480–483 (2002).
Saric, T. et al. An IFN-γ-induced aminopeptidase in the ER, ERAP1, trims precursors to MHC class I-presented peptides. Nat. Immunol. 3, 1169–1176 (2002).
Hattori, A., Matsumoto, H., Mizutani, S. & Tsujimoto, M. Molecular cloning of adipocyte-derived leucine aminopeptidase highly related to placental leucine aminopeptidase/oxytocinase. J. Biochem. 125, 931–938 (1999).
Schomburg, L., Kollmus, H., Friedrichsen, S. & Bauer, K. Molecular characterization of a puromycin-insensitive leucyl-specific aminopeptidase, PILS-AP. Eur. J. Biochem. 267, 3198–3207 (2000).
Miyashita, H. et al. A mouse orthologue of puromycin-insensitive leucyl-specific aminopeptidase is expressed in endothelial cells and plays an important role in angiogenesis. Blood 99, 3241–3249 (2002).
Cui, X. et al. Identification of ARTS-1 as a novel TNFR1-binding protein that promotes TNFR1 ectodomain shedding. J. Clin. Invest. 110, 515–526 (2002).
York, I.A. et al. The ER aminopeptidase ERAP1 enhances or limits antigen presentation by trimming epitopes to 8–9 residues. Nat. Immunol. 3, 1177–1184 (2002).
Townsend, A. et al. Association of class I major histocompatibility heavy and light chains induced by viral peptides. Nature 340, 443–448 (1989).
Kunisawa, J. & Shastri, N. The group II chaperonin TRiC protects proteolytic intermediates from degradation in the MHC class I antigen processing pathway. Mol. Cell 12, 565–576 (2003).
Karttunen, J., Sanderson, S. & Shastri, N. Detection of rare antigen presenting cells by the lacZ T-cell activation assay suggests an expression cloning strategy for T-cell antigens. Proc. Natl. Acad. Sci. USA 89, 6020–6024 (1992).
Malarkannan, S., Goth, S., Buchholz, D.R. & Shastri, N. The role of MHC class I molecules in the generation of endogenous peptide/MHC complexes. J. Immunol. 154, 585–598 (1995).
Goldrath, A.W. & Bevan, M.J. Selecting and maintaining a diverse T-cell repertoire. Nature 402, 255–262 (1999).
Simpson, E., Scott, D. & Chandler, P. The male-specific histocompatibility antigen, H-Y: A history of transplantation, immune response genes, sex determination and expression cloning. Annu. Rev. Immunol. 15, 39–61 (1997).
Tanioka, T. et al. Human leukocyte-derived arginine aminopeptidase. The third member of the oxytocinase subfamily of aminopeptidases. J. Biol. Chem. 278, 32275–32283 (2003).
Saveanu, L. et al. Concerted peptide trimming by human ERAP1 and ERAP2 aminopeptidase complexes in the endoplasmic reticulum. Nat. Immunol. 6, 689–697 (2005).
Falk, K. et al. Identification of naturally processed viral nonapeptides allows their quantification in infected cells and suggests an allele specific T cell epitope forecast. J. Exp. Med. 174, 425–434 (1991).
Cresswell, P., Bangia, N., Dick, T. & Diedrich, G. The nature of the MHC class I peptide loading complex. Immunol. Rev. 172, 21–28 (1999).
Grandea, A.G. et al. Impaired assembly yet normal trafficking of MHC class I molecules in tapasin mutant mice. Immunity 13, 213–222 (2000).
Garbi, N. et al. Impaired immune responses and altered peptide repertoire in tapasin-deficient mice. Nat. Immunol. 1, 234–238 (2000).
Williams, A.P., Peh, C.A., Purcell, A.W., McCluskey, J. & Elliott, T. Optimization of the MHC class I peptide cargo is dependent on tapasin. Immunity 16, 509–520 (2002).
Ackerman, A.L., Kyritsis, C., Tampe, R. & Cresswell, P. Access of soluble antigens to the endoplasmic reticulum can explain cross-presentation by dendritic cells. Nat. Immunol. 6, 107–113 (2005).
Ramirez-Solis, R., Davis, A.C. & Bradley, A. Gene targeting in embryonic stem cells. Methods Enzymol. 225, 855–878 (1993).
Lander, M.R., Moll, B. & Rowe, W.P. A procedure for culture of cells from mouse tail biopsies: brief communication. J. Natl. Cancer Inst. 60, 477–478 (1978).
Bhattacharya, D., Logue, E.C., Bakkour, S., DeGregori, J. & Sha, W.C. Identification of gene function by cyclical packaging rescue of retroviral cDNA libraries. Proc. Natl. Acad. Sci. USA 99, 8838–8843 (2002).
Acknowledgements
We thank E. James, H. Nolla and D. King for assistance; and N. Blanchard and J. Shugart for comments on the manuscript. Supported by the National Institutes of Health (N.S.).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing financial interests.
Rights and permissions
About this article
Cite this article
Hammer, G., Gonzalez, F., Champsaur, M. et al. The aminopeptidase ERAAP shapes the peptide repertoire displayed by major histocompatibility complex class I molecules. Nat Immunol 7, 103–112 (2006). https://doi.org/10.1038/ni1286
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/ni1286
This article is cited by
-
Down-regulation of ERAP1 mRNA expression in non-small cell lung cancer
BMC Cancer (2023)
-
ERAP1-ERAP2 dimers trim MHC I-bound precursor peptides; implications for understanding peptide editing
Scientific Reports (2016)
-
Differences in genetic variation in antigen-processing machinery components and association with cervical carcinoma risk in two Indonesian populations
Immunogenetics (2015)
-
ERAP1 in the pathogenesis of ankylosing spondylitis
Immunologic Research (2014)
-
ERAP1 genetic variations associated with HLA-B27 interaction and disease severity of syndesmophytes formation in Taiwanese ankylosing spondylitis
Arthritis Research & Therapy (2012)