Allele-specific alterations in the peptidome underlie the joint association of HLA-A*29:02 and Endoplasmic Reticulum Aminopeptidase 2 (ERAP2) with Birdshot Chorioretinopathy

Virtually all patients of the rare inflammatory eye disease birdshot chorioretinopathy (BSCR) carry the HLA-A*29:02 allele. BSCR is also associated with endoplasmic reticulum aminopeptidase 2 (ERAP2), an enzyme involved in processing HLA class I ligands, thus implicating the A*29:02 peptidome in this disease. To investigate the relationship between both risk factors we employed label-free quantitative mass spectrometry to characterize the effects of ERAP2 on the A*29:02-bound peptidome. An ERAP2-negative cell line was transduced with lentiviral constructs containing GFP-ERAP2 or GFP alone, and the A*29:02 peptidomes from both transduced cells were compared. A similar analysis was performed with two additional A*29:02-positive, ERAP1-concordant, cell lines expressing or not ERAP2. In both comparisons the presence of ERAP2 affected the following features of the A*29:02 peptidome: 1) Length, with increased amounts of peptides >9-mers, and 2) N-terminal residues, with less ERAP2-susceptible and more hydrophobic ones. The paradoxical effects on peptide length suggest that unproductive binding to ERAP2 might protect some peptides from ERAP1 over-trimming. The influence on N-terminal residues can be explained by a direct effect of ERAP2 on trimming, without ruling out and improved processing in concert with ERAP1. The alterations in the A*29:02 peptidome suggest that the association of ERAP2 with BSCR is through its effects on peptide processing. These differ from those on the ankylosing spondylitis-associated HLA-B*27. Thus, ERAP2 alters the peptidome of distinct HLA molecules as a function of their specific binding preferences, influencing different pathological outcomes in an allele-dependent way. These results indicate that ERAP2 functionally active in the transduced cells and its a double influence on the P1 residues of A*29:02 ligands: it of peptides with P1 to trimming by this and increases the abundance of peptides with nonpolar P1 residues.

Immunological components include the presence of CD4+ and CD8+ T-cells in the choroidal lesions and vitreous fluid (3,4), increased levels of IL-17 in the affected eyes (5), higher serum levels of IL-23 and other pro-inflammatory cytokines (6) and IL-17-producing CD8+ T cells in the peripheral blood of the patients (7). A role of killer-cell immunoglobulin-like receptor (KIR) genes has also been proposed (8,9).
BSCR is very strongly associated with HLA-A*29:02 (10). This allele is frequent (about 7%) in European populations, but only a very small percentage of A*29:02-positive individuals develop the disease. This is presumably due to the contribution of multiple genes and perhaps also to the requirement of undefined environmental factors. Yet, with virtually all patients carrying this allele (11), this disease shows the highest association with an HLA gene.
The A*29:02 peptidome is characterized by a prominent binding motif of C-terminal Tyr, present in about 90% of the A*29:02 ligands, and a looser secondary motif of aliphatic/aromatic residues at peptide position (P) 2 and PC-2 (12).
Multiple features of BSCR suggest a central pathogenetic role of HLA-A*29-bound peptides. First, genetic studies narrowed down the HLA association to the A*29:02 subtype and failed to detect, upon conditional analyses, any additional loci in the vicinity of HLA-A that could account for the association of A*29:02 with BSCR through linkage disequilibrium with another risk gene, strongly supporting a direct effect of this allotype (10). Second, the fact that BSCR is essentially an organ-specific disease suggests an involvement of peptide epitopes derived from eye proteins in the initiation or exacerbation of the disease. Given the relatively unrestricted HLA-A*29 binding motif, the number of possible epitopes is quite large. Thus, knowing the peptide features that favor A*29 binding, and how antigen processing can regulate the availability of peptides with such features, may help in the identification of uveitogenic peptides. Third, both the presence of CD8+T cells in the choroidal infiltrates and vitreous fluid of BSCR patients reactive to antigens in the retinal and choroidal lysates (4) provide further evidence of the role of MHC-I-mediated antigen presentation in this disease. Fourth, the involvement of KIR genes in BSCR also points out to the relevance of Major Histocompatibility Complex class I (MHC-I) bound peptides, since the interaction of these receptors with MHC is peptide-dependent (13)(14)(15). In particular, the affinity of peptide/MHC-I interaction is known to affect recognition by KIR receptors (16).
A central role of MHC-bound peptides in the pathogenesis of BSCR is also strongly suggested by the association of a polymorphism determining the expression of endoplasmic reticulum amino peptidase (ERAP) 2 with this disease (10). ERAP2 belongs to the M1 subfamily of Zn-metallopeptidases (17), also including ERAP1, a closely related enzyme that trims peptides in the ER to the proper length for binding to MHC-I proteins (18,19), and insulinregulated aminopeptidase, which is involved in antigen cross-presentation in dendritic cells (20,21).
In contrast to ERAP1, ERAP2 is expressed in only 75% of individuals and shows a very limited functional polymorphism, consisting in a N392K change, known to affect peptide trimming (30). However, due to linkage disequilibrium between the single nucleotide polymorphism encoding the 392N allele and another polymorphism impairing ERAP2 protein expression (31), only the 392K variant is expressed in most individuals.
ERAP2 shows remarkable differences with ERAP1 in specificity and substrate handling, reviewed in (24). For instance, whereas ERAP1 can cleave nearly all residues, albeit with preference for hydrophobic ones (32), ERAP2 cleaves very few N-terminal residues, particularly Arg (22,24,33,34). In addition, whereas ERAP1 is very efficient with relatively long peptides (>9-mers) and virtually unable to cleave 8-mers and shorter peptides (35), ERAP2 is most efficient with short peptides and its trimming capacity quickly decreases with substrate length (36).
Although the structural and enzymatic properties of ERAP2 have been extensively characterized in vitro, the actual role of this enzyme in shaping MHC-I-bound peptidomes in live cells is almost unknown. Only recently, the effects of ERAP2 on the HLA-B*27 peptidome, which is totally different from that of A*29:02, have been reported (37,38). These studies showed, among other alterations, a significant effect of the enzyme on decreasing the expression of peptides with N-terminal basic residues, which is in agreement with its known trimming specificity. They also suggested that ERAP2 and ERAP1 largely act as separate entities in vivo.
The joint association of A*29:02 and ERAP2 with BSCR provides an exceptional opportunity to explore the functional interaction between two major genetic risk factors for this disease. Since the only known function of ERAP2 is in the processing of MHC-I ligands, the effects of this enzyme on the A*29:02 peptidome must constitute the basis for its association with BSCR and directly relate to the pathogenetic role of A*29:02. Thus, in this study we characterized the features of A*29:02 ligands that are modulated by ERAP2 and the magnitude of the alterations induced in the peptidome upon expression of this enzyme, thus providing a basis for the joint contribution of A*29:02 and ERAP2 to the pathogenesis of BSCR.  To determine the effects of ERAP2 expression on the amounts of A*29:02 ligands we compared the A*29:02 peptidomes from PF-GFP and PF-ERAP2 cells, focusing on the shared peptides found in both cell lines. The relative expression of these shared peptides was comparatively analyzed following a strategy described in detail in a previous study (38). The To look for differential features between the A*29:02 ligands showing the largest differences in relative abundance between cell lines, we compared the peptides with IR>1.5 in PF-GFP relative to PF-ERAP2 with those showing IR>1.5 in PF-ERAP2 relative to PF-GFP.

MATERIALS AND METHODS
This was based on the assumption that the quantitative effects of ERAP2 on the A*29:02 peptidome should be best observed among peptides showing larger differences in relative amounts between the two cell lines. As an internal control, the same comparisons were carried out between the peptide subsets with IR>1.0-1.5 from each cell line, assuming that any differences due to ERAP2 should be attenuated or absent among the peptides with similar abundance in both cell lines. Peptides found only in one of the two cell lines were separately compared.
Exactly the same strategy was used for the comparative analysis of the A*29:02 peptidomes from GM19397 and GM19452, except that the MaxQuant Version used was 1.5.8.3 and the Human Uniprot database version was updated (release 20-04-17, 70946 entries).

Classification of amino acid residues according to ERAP1 and ERAP2 susceptibility.
Amino acid residues were classified according to their susceptibility to ERAP1 (32), as ERAP1- Statistical analyses. Differences in peptide length and residue frequencies were assessed by the 2 test with Bonferroni correction, when applicable. Differences in binding affinity and hydropathy of A*29:02 ligands were analyzed by the Mann-Whitney U test. P < 0.05 was considered as statistically significant in all cases.  (Fig 1A). An increased expression of ERAP1 (about 30%) was observed upon lentiviral transduction with either the GFP or the GFP-ERAP2 constructs, with no difference between these two cell lines, indicating that the increased expression of ERAP1 was due to the transduction procedure and unrelated to ERAP2. The expression of this enzyme on PF-ERAP2 cells was significant, albeit lower (36±7%) than in an ERAP2-positive LCL used as control. Both GM19397 and GM19452 expressed the same protein levels of ERAP1 (Fig 1B). Whereas the former cell line did no express ERAP2, the expression level of ERAP2 in GM19452 was similar to that of the control LCL P50, and therefore, over 2-fold the expression level in PF-ERAP2 cells.  (Table S1). These peptides consisted in 1.4% 8-mers, 72.6% 9-mers, 14.6% 10-mers and 11.4% longer peptides, showed a main anchor motif of Cterminal Tyr (92.4%), and aliphatic/aromatic P2 (75.1%) and PC-2 (65.9%) residues (Fig. S2A),

Identification of
with virtually no differences among individual cell lines. These features are similar to those previously reported for A*29:02 ligands from three ERAP2-negative LCL, including PF (12). A total of 3984 peptides were found in both PF-GFP and PF-ERAP2 cells (Table S2). Global residue frequencies were very similar among the peptides identified from mock-or ERAP2transduced PF cells. In particular, no statistically significant differences in the frequency of basic P1 residues were observed between PF-GFP and PF-ERAP2 cells (R+K: 7.5% and 7.0%, respectively).
Identification of A*29:02 ligands from GM19397 and GM19452. A total of 2684 and 4041 peptides of 8 to 14 residues were identified from GM19397 and GM19452 cells, respectively. In order to remove non-A*29 contaminants, including ligands from non-A*29 MHC molecules, two filters were applied. The first one was to select for peptides with either Tyr or Phe at the Cterminus, since this is the major A*29:02 motif. The second one, similarly as done with PFderived peptides, was to select those peptides with a theoretical affinity IC50 of 1471 peptides with the selected length, C-terminal motif and affinity were assigned as A*29:02 ligands, of which 1196 and 1415 were found in GM19397 and GM19452 respectively (Table S3). A total of 1140 peptides were found in both cell lines (Table S4). Both the length and residue distribution of the A*29:02 ligands from these cell lines were very similar as in PF cells (Fig S2B), which confirms that the filtering procedure used selects for A*29:02 ligands with high reliability. The increased amounts of 8-mers suggest that trimming of 9-mers is slightly favored in the presence of ERAP2. Yet, the higher abundance of peptides longer than 9-mers indicates that ERAP2 is not increasing the destruction of long A*29:02 ligands, which might be expected if this enzyme were indirectly favoring ERAP1trimming of such ligands, since peptides longer than 9-mers are the preferred substrates for this latter enzyme. Exactly the same pattern was observed upon comparing the predominant peptides from GM19397 and GM19452 (Fig. 2B),

Quantitative effects of ERAP2 on the
although the lower peptide numbers precluded reaching statistical significance in most cases.
The predicted affinity of the 9-mers was higher than that of longer peptides in all cases (Fig. 3). Therefore, the higher number of peptides longer than 9-mers in ERAP2-positive cells is probably not determined by their affinity, although some bias of theoretical algorithms towards 9-mers cannot be formally ruled out.
The effect of ERAP2 on the length of A*29:02 ligands is opposite to that observed among HLA-B*27 ligands, where 9-mers were increased and longer peptides were decreased in the presence of this enzyme (38).
ERAP2 alters N-terminal residue usage and hydrophobicity. A comparison of the P1 frequencies between the predominant peptides in PF-ERAP2 and PF-GFP revealed that in the presence of ERAP2, Phe, Leu and Trp were increased and Ala, Lys and Gln were decreased in the IR>1.0 and/or IR>1.5 subsets (Fig 4A and S4A). Only Leu was increased in the IR>1.0-1.5 subset in the presence of ERAP2 (Fig. S4A). The significance of these alterations became obvious when residues were grouped either according to their susceptibility to ERAP2 trimming or to their hydrophobicity. The residues most susceptible to ERAP2 trimming were globally decreased and those with the highest hydrophobicity were increased in the presence of ERAP2 among the peptides in the IR>1.0, IR>1.5 and even IR>1.0-1.5 subsets. (Fig. 4A and S4B-C).
No differences were observed when the P1 residues were grouped according to their susceptibility to ERAP1 (Fig. 4A and S4D).
Once again, very similar results were obtained when the predominant peptides (IR>1.0 subsets) of the GM19397 and GM19452 were compared (Fig. 4B). Although, at the level of individual P1 residues, some differences were observed, relative to PF-GFP/PF-ERAP2, most notably in the increased frequency of Tyr in GM19452 and of Gly in GM19397, the pattern of P1 residue usage was similar in both comparisons. Most notably, as in PF-GFP/PF-ERAP2, the joint frequency of ERAP2-suceptible residues was decreased in GM19452 (ERAP2+), relative to GM19397 (ERAP2-), with smaller effect on residue usage according to ERAP1 susceptibility. A tendency towards higher hydropathy of P1 residues was also observed in GM19452.
These results indicate that ERAP2 is functionally active in the transduced cells and its expression has a double influence on the P1 residues of A*29:02 ligands: it diminishes the abundance of peptides with P1 residues susceptible to trimming by this enzyme, and increases the abundance of peptides with nonpolar P1 residues. Whereas ERAP2-susceptible residues at P1 are disfavored for A*29:02 binding, those with high hydropathy are favored (Fig. S5).
Therefore, ERAP2 has an optimizing effect on the P1 residues of A*29:02 ligands.
A similar analysis carried out for the N-terminal flanking (P-1) residues of the peptides in the same subsets from both PF-GFP/PF-ERAP2 and GM19397/GM19452 showed no statistical differences (data not shown), indicating a limited effect of ERAP2 on the generation of A*29:02 ligands as a function of their P-1 residues.
ERAP2 expression increased the hydrophobicity of A*29:02 ligands at positions other than P1 only in transduced cells. We next examined the effects of ERAP2 expression on the frequency and hydrophobicity of residues at positions other than P1 in transduced PF-GFP and PF-ERAP2 cells. This was done because increased hydrophobicity of A*29:02 ligands had been observed in active ERAP1 contexts and correlated with ERAP1 activity in a previous study (12). This analysis was separately carried out for 9-mers and 10-mers to ensure proper alignment of the peptide sequences ( Fig. S6A-B). Increased hydrophobicity at all positions was observed on both the 9-mers and 10-mers predominant in the presence of ERAP2 (IR>1.0 and IR>1.5). The tendency, which was attenuated in the IR>1.0-1.5 subsets, resulted in an increased hydrophobicity of the A*29:02 ligands predominant in the presence of this enzyme, relative to those predominant in its absence (Fig. S6C).
In contrast to the observations in transduced cells, increased hydropathy was not observed, either globally or at the internal positions of A*29:02 ligands, when the peptides predominant in GM19452 (ERAP2+) were compared with those predominant in the ERAP2negative GM19397 LCL (Fig. S7). Actually, the peptides predominant in the former cell line were globally more hydrophilic. Since ERAP2 expression did not show a consistent effect on the hydropathy of the A*29:02 peptidome in different cell lines, our results do not allow to conclude that this enzyme influences the hydrophobicity of internal residues in A*29:02 ligands.
ERAP2 does not alter the global affinity of the A*29:02 peptidome. In spite of the increased hydrophobicity observed in ERAP2-transduced cells, the global affinity of the A*29:02 ligands predominant with or without ERAP2 was virtually the same independently of the IR subsets compared (Fig. 5A). This was surprising, since increased hydrophobicity of A*29:02 ligands correlated with higher affinity in a previous study (12). Since peptides predominant (IR>1.0) in ERAP2-transduced PF cells showed a lower percentage of peptides with C-terminal Tyr , relative to PF-GFP cells (90.8% and 95.9%, respectively), we separately examined the effects of ERAP2 on the affinity of A*29:02 ligands as a function of their Cterminal residues. The results showed that the affinity was higher for peptides with C-terminal Tyr (Fig. 5B). In addition, the affinity of each subpeptidome was higher in ERAP2-transduced cells. Thus, there was a correlation between increased hydrophobicity and affinity, but only among peptides with the same C-terminal residue. The increased abundance of lower affinity peptides without C-terminal Tyr in ERAP2-transduced PF cells compensated the higher affinity of Tyr + ligands, resulting in unaltered affinity of the whole A*29:02 peptidome.
No differences in the global affinity of A*29:02 ligands were observed between the predominant peptides from GM19397 and GM19452 (Fig 5C-D), indicating a lack of effect of ERAP2 expression on the global affinity of the A*29:02 peptidome also in these cells. peptide length distribution (Fig. 6A), 2) decreased frequency of ERAP2-susceptible P1 residues, with little or no influence on their ERAP1 susceptibility (Fig. 6B-C), 3) increased hydrophobicity of P1 residues (Fig. 6D), and 4) global affinity and hydrophobicity (Fig. S8).

Differential features among
These differences did not always reached statistical significance due to the relatively small peptide numbers.
These results indicate that the same effects of ERAP2 expression on modulating peptide amounts in the A*29:02 peptidome apply to the qualitative generation/destruction of specific ligands.

DISCUSSION
Lentiviral transduction of an ERAP2-negative, HLA-A*29:02-positive, LCL expressing highly active ERAP1 variants was used in this study to characterize the effects of ERAP2 on the A*29:02 peptidome. The infection procedure increased ERAP1 levels, but these were the same in the mock-and ERAP2-transduced cells. Since the expression levels of ERAP2 achieved with this procedure were lower than those in LCL, the effects observed cannot be attributed to overexpression of the enzyme. These effects are summarized as follows: 1) on length, with increased amounts of peptides longer than 9-mers, 2) on P1 residue usage, with less ERAP2-susceptible residues and more hydrophobic ones, 3) on internal sequences, with increased hydrophobicity at all positions, 4) on increasing the global affinity of subpeptidomes with the same residue, but not the global affinity of the peptidome. It is important to note that, regardless of the qualitative effects of ERAP2 in the generation or destruction of specific epitopes, a major effect of this enzyme is a quantitative one, on altering the abundance of a significant fraction of the A*29:02 peptidome.
A possible limitation of using transduced cells is that particular subpopulations of both mock-and ERAP2-transduced cells must be selected for use. This selection was based on the high expression of the reporter GFP gene and on significant expression of ERAP2 in the corresponding transduced cells. Although these subpopulations were not monoclonal, neither gene copy number nor their insertion sites are controlled, and these are presumably different in mock-and ERAP2-transduced cells. Therefore, unforeseen differential effects in both populations of transduced cells that might potentially affect antigen processing cannot be ruled out.
For this reason, in order to properly assign the effects observed with transduced cells to ERAP2, we carried out a similar comparative analysis using two non-transduced LCL concordant in the expression of a high activity ERAP1 variant but expressing or not ERAP2.
When the differences observed between these two LCL reproduced those observed in transduced cells they were considered to be ERAP2-dependent effects. These were: 1) the alterations in peptide length distribution, with lower abundance of 9-mers and higher abundance of longer peptides, 2) the diminishment of ERAP2-susceptible P1 residues, and 3) the increased hydrophobicity at P1.
In contrast, the fact that the higher hydrophobicity at internal positions observed on ERAP2-transduced cells was not reproduced in non-transduced LCL implies that this effect cannot be assigned as a consistent one of ERAP2 expression on the A*29:02 peptidome.
The observed alterations in peptide length are not due to a general improvement of peptide processing by ERAP2 because, in the absence of this enzyme, higher ERAP1 activity leads to an increased abundance of 9-mers (12), just the opposite as observed with ERAP2. It is also unlikely due to affinity differences, because 9-mers showed higher affinity than longer peptides for A*29:02. A plausible explanation for our observation is that ERAP2 may, to some extent, protect longer peptides from ERAP1 degradation. ERAP1 is generally more efficient in trimming 10-mers and longer peptides than 9-mers (35), whereas ERAP2 is relatively inefficient with long peptides (36) and its trimming specificity is much more restricted than that of ERAP1 (24). Thus, unproductive binding of peptides longer than 9-mers by ERAP2, namely those with ERAP2-resistant P1 residues, which account for a large majority of the A*29:02 peptidome, might influence their trimming by ERAP1 through a protective effect: peptides unproductively bound to ERAP2 would be less available to ERAP1, slowing down their trimming by this enzyme.
Protection of long peptides from ERAP1 trimming through unproductive binding to ERAP2 may be more relevant for HLA class I molecules with preference for hydrophobic peptides, such as A*29:02, due to the fact that, unlike ERAP1, ERAP2 has a substrate binding site with a largely neutral electrostatic potential (36), which might favor binding of relatively hydrophobic peptides, such as A*29:02 ligands.
The decreased amounts of peptides with ERAP2-susceptible P1 residues strongly suggest a direct effect of ERAP2 in over-trimming and destruction of these ligands. The increased amounts of A*29:02 ligands with hydrophobic residues at P1 observed in the presence of ERAP2 is very reminiscent of the effects observed on ERAP2-negative cells in the presence of highly active ERAP1 variants (12), and might reflect a general improvement of such processing in the presence of ERAP2. However, the limited alteration in the global residue frequencies according to their susceptibility to ERAP1 is unlike that observed across ERAP1 differences in our previous study. Thus, the increased frequency of hydrophobic P1 residues in the presence of ERAP2 may just result from the depletion of peptides with ERAP2-susceptible residues in the ER, which are disfavored for A*29:02 binding, consequently increasing the availability of peptides with ERAP2-resistant hydrophobic P1 residues, which favor A*29:02 binding.
In conclusion, our results suggest that ERAP2 may influence the A*29:02 peptidome in at least two ways: 1) by protecting A*29:02 ligands longer than 9-mers from ERAP1 overtrimming, and 2) by destroying A*29:02 ligands with ERAP2-susceptible residues through direct trimming. A general improvement in the efficiency of peptide processing, in concert with ERAP1, is not ruled out, but it is not essential to account for the observed alterations in P1 residue frequencies.
Although, taken individually, none of the effects assigned in this study to ERAP2 were very prominent, together they alter the relative abundance of a substantial fraction of the A*29:02 peptidome.
Besides the relevance of our study in defining how ERAP2 shapes an MHC-I-bound peptidome, we must address the relationship of these findings to the pathogenetic role of A*29:02 and, more generally, to the pathophysiology of BSCR. As in other autoinflammatory/autoimmune disorders, a simple straightforward answer can hardly be expected due to the complex interplay of innate and adaptive immunity mechanisms likely involved in this disease. However some insights might be offered. ERAP2-mediated alterations of the A*29:02 peptidome could influence ocular inflammation in at least three ways: 1) altering general properties of A*29:02 related to a direct pro-inflammatory capacity, 2) altering specific antigen presentation to T cells, and 3) altering NK receptor recognition.
The first possibility is suggested by studies on HLA-B*27, whose folding properties confer this molecule the capacity to misfold, accumulate in the ER and promote IL23 upregulation (48,49). ligand. In addition, both the histology of choroidal lesions (3,55), their hypopigmented nature, and the epidemiology of melanoma among BSCR patients (56) suggest a potential relevance of melanoma-associated antigens, including those expressed in normal melanocytes (2). One such peptide, derived from the melanocyte differentiation antigen tyrosinase, was identified as an A*29:02 ligand in a previous study (57). The effects of ERAP2 on A*29:02-bound peptides described in our study could be used to identify potential ERAP2-dependent epitopes from melanocyte proteins.
Another effect of the changes induced by ERAP2 on A*29:02 might be through affecting NK cell recognition. KIR receptors are likely to play a role in BSCR (8,9) and their recognition of antigen is sensitive to changes in the nature and affinity (16)         Maximal increase of peptides with small P1 residues. Affinity No alteration Lower