ERAP1 Controls the Autoimmune Response against Melanocytes in Psoriasis by Generating the Melanocyte Autoantigen and Regulating Its Amount for HLA-C*06:02 Presentation

This article is featured in Top Reads, p.

Common autoimmune diseases arise from a polygenic predisposition, in which combinations of diverse protective and risk gene variants induce an autoimmune response and ultimately lead to a clinical phenotype (1). The association with certain HLA alleles characterizes virtually any human autoimmune disease. In particular, three of the >19,000 HLA class I alleles currently registered in the IMGT/HLA Database (2) pose a high risk for immune-mediated diseases: HLA-C*06:02 is associated with psoriasis, HLA-B*27 with ankylosing spondylitis, and HLA-B*51 with Behçet disease. This is interpreted to indicate that the pathomechanisms in the three diseases are directly related to these HLA class I alleles. In view of the unexplained autoimmune pathogenesis, the diseases were grouped under the term MHC-I-opathies (3). As another shared feature, genome-wide association studies (GWASs) have revealed that epistasis—that is, nonadditive gene–gene interaction between the HLA class I risk alleles and certain variants of the endoplasmic reticulum aminopeptidase 1 (ERAP1)—controls the risk of contracting these diseases (4–8). The functional interaction between HLA class I alleles and ERAP1 may thus play a crucial role in the development of autoimmune diseases, the elucidation of which should provide essential insights into disease pathogenesis and enable the development of innovative therapeutic strategies that benefit patients without causing broad immunosuppression.

HLA class I molecules and ERAP1 cooperate in Ag processing and presentation. HLA class I molecules primarily present peptide Ags originating from cytosolic proteins to the TCRs of CD8⁺ T cells. Thus, CD8⁺ T cell–mediated immune responses are specifically directed against target cells that express and process these proteins into antigenic peptides (9). The peptide Ags are generated by the multicatalytic cellular proteasome. It creates mainly NH₂-extended peptides with a defined C terminus that are translocated by the TAP into the endoplasmic reticulum for loading onto HLA class I molecules. HLA class I molecules limit the size of peptide Ags typically to 8–10 aa because the peptide-binding groove is closed at both sides (10). ERAP1 is a peptidase that may create the appropriate peptide length for HLA class I binding by N-terminal trimming of Ag precursors, but ERAP1 can also destroy HLA class I ligands by overtrimming (11, 12). Approximately one third of peptides presented by HLA class I result from processing by ERAP1, whereas the major part of the HLA class I peptidome is generated independently from ERAP1 activity (13). Various nonsynonymous coding ERAP1 variants modulate the catalytic aminopeptidase activity (14, 15). This may lead to altered peptide repertoires, HLA misfolding, and HLA homodimer formation, all of which have been discussed as possible causes of abnormal CD8⁺ T cell or NK cell activation or autoinflammation (16). However, it is ultimately unresolved how ERAP1 variants, in conjunction with the disease-associated HLA class I alleles, determine the development of immune-mediated inflammatory diseases.

Psoriasis is a common T cell–mediated autoimmune skin disease. HLA-C*06:02 is the main risk allele (17, 18). Development of psoriasis lesions depends on the epidermal recruitment, activation, and clonal expansion of CD8⁺ T cells (19–21). Through unbiased screening of a TCR rearranging the Vα3S1 gene segment in the α-chain and the Vβ13S1 gene segment in the β-chain (Vα3S1/Vβ13S1 TCR) CD8⁺ T cell clone from the pathogenic psoriatic T cell infiltrate of an HLA-C*06:02–positive donor, we had discovered that HLA-C*06:02 directs an autoimmune CD8⁺ T cell response against melanocytes as an underlying pathomechanism of psoriasis (22). We further had identified a peptide from ADAMTS-like protein 5 (ADAMTSL5) as the causative melanocyte autoantigen, which is naturally processed in melanocytes, presented by HLA-C*06:02, and recognized by the Vα3S1/Vβ13S1 TCR. ADAMTSL5 is highly expressed in psoriasis lesions, especially melanocytes (22, 23). ADAMTSL5-specific CD8⁺ T cells are clearly detectable in psoriasis patients and produce psoriasis key cytokines, IL-17A and IFN-γ, upon antigenic stimulation (22).

These insights enabled us to investigate the functional mechanisms underlying the statistically defined gene–gene interaction between particular ERAP1 haplotypes and the main HLA class I risk allele, HLA-C*06:02. For our experiments, we had established a TCR activation assay that reproduces the in vivo psoriatic autoimmune response against melanocytes and ADAMTSL5 as the causative psoriatic autoantigen. It employs the autoreactive HLA-C*06:02–restricted ADAMTSL5-specific Vα3S1/Vβ13S1 TCR expressed in a CD8⁺ mouse T hybridoma reporter cell line (22), which indicates TCR stimulation by the induction of super GFP (sGFP) under the control of the promoter of NFAT (22, 24). Together with HLA-C*06:02, melanocytes as the autoimmune target cells of the Vα3S1/Vβ13S1 TCR, the melanocyte autoantigen ADAMTSL5, and different risk-associated ERAP1 haplotypes, this assay combines the core components of the psoriatic autoimmune response. Such complex experimental conditions of a human autoimmune disease can hardly be simulated in experimental animal models.

Our results demonstrate that ERAP1 elicits melanocyte immunogenicity for the psoriatic autoimmune response by generating the autoantigenic ADAMTSL5 peptide for presentation by HLA-C*06:02. Based on this, different ERAP1 haplotypes control the extent of an autoimmune response against melanocytes and thus probably also the risk of HLA-C*06:02 carriers for psoriasis by different autoantigen yields. Interestingly, cell surface HLA-C expression proved to be much more affected by ERAP activity than that of the overall HLA class I expression, which comprises HLA-A, HLA-B, and HLA-C molecules, resembling the ERAP1 dependence of the expression of the mouse major histocompatibility Ag molecule H-2L^d (25). These findings reveal how the interaction between HLA-C*06:02 and ERAP1 drives CD8⁺ T cell–mediated autoimmune pathogenesis in psoriasis, and they propose a pathomechanism by which ERAP1 risk haplotypes enhance autoantigen presentation to induce CD8⁺ T cell autoimmune responses in other HLA class I–associated diseases as well. This may open new avenues for the treatment of HLA class I–associated autoimmune diseases, which target ERAP1 function.

Generation of the Vα3S1/Vβ13S1–TCR CD8⁺ reporter T hybridoma from the αβ-TCR chains of a lesional psoriatic CD8⁺ T cell clone and its culture conditions have been described (20, 22). WM793 (HLA-A*01:01, HLA-A*29:01, HLA-B*35:01, HLA-B*57:01, HLA-C*04:01, and HLA-C*06:02) and WM278 cells (HLA-A*02:01, HLA-A*26:01, HLA-B*13:02, HLA-B*38:01, HLA-C*06:02, and HLA-C*12:03) (National Center for Biotechnology Information Biosample accession: https://www.ncbi.nlm.nih.gov/biosample/?term=SAMN03471796) were originally obtained from the Wistar Institute. They were cultured in TU2% medium containing MCDB 153, 20% Leibovitz’s L15 medium, 5 µg/ml insulin (Sigma-Aldrich), 2% FCS, and 1.68 mM CaCl₂. WM793 and WM278 cells were tested for DNA short tandem repeat analysis. HEK293T cells (HLA-A*02:01, HLA-A*03:01, HLA-B*07:02, HLA-B*07:02, HLA-C*07:02, and HLA-C*07:02) were maintained in DMEM supplemented with geneticin (500 µg/ml). All cell lines were negative for mycoplasma contamination.

Single-guide RNAs targeting ERAP1 inserted into the CRISPR/Cas9–encoding px330 plasmid and generation of ERAP1^−/− HEK293T cells have been reported (26). WM793 or WM278 cells were transfected with CRISPR plasmid three times and cloned at densities of 0.3 cells per well. Successful gene editing and deletion of protein expression were confirmed by Western blotting (Fig. 1A, 1B) and by sequencing of targeted sites of genomic DNA.

Parental and ERAP1^−/− WM793, WM278, and HEK293T cells were seeded at growth-adjusted cell numbers prior to HLA staining in 24-well plates. HEK293T cells were incubated with IFN-γ (1 ng/ml) to induce sufficient HLA-C expression. On the next day, cells at 80–90% density were detached using 0.2% EDTA/PBS and stained with HLA-C Ab (DT-9, no. 566372; BD Biosciences), HLA-ABC Ab (W6/32, no. 311406; BioLegend), or corresponding isotype IgG (IgG2b, no. 401208; IgG2a no. 400214; BioLegend), all conjugated with PE. Data were acquired by flow cytometry and analyzed by FlowJo (889; Tree Star). The gating strategy is given in Supplemental Fig. 1A. DT-9 Ab is considered the most specific HLA-C Ab, although it also detects HLA-A*23:01, HLA-A*80:01, HLA-B*13:01, HLA-B*35:01, HLA-B*40:06, HLA-B*73:01, and HLA-E (27). Only WM793 cells express HLA-B*35:01, which is recognized by DT-9 with low affinity. HLA-E (no. ab11821; Abcam) and corresponding isotype (no. 556648; BD Pharmingen) stainings were detected by goat anti-mouse IgG (H+L) cross-absorbed Ab Alexa Fluor 488 (no. A11001; Invitrogen).

To stain adherent cells, chamber glass slides were coated with 0.5 mg/ml poly-d-lysine (no. P7280; Sigma-Aldrich) at 4°C overnight and seeded with parental cell lines or ERAP1^−/− WM793 and WM278 cell clones adjusted to yield comparable cell density on analysis. After 2 d of culture, cells were reacted with Abs for HLA-C, HLA-ABC, or corresponding isotype control (no. 401216, no. 400264, no. 311441, and DT-9, all without azide; BioLegend) and anti-mouse IgG (H+L) Alexa Fluor 488 or 594 secondary Ab (no. A11001 or no. A11005; Invitrogen). After washing, the cells were mounted with Fluorescence Mounting Medium (no. S3203; DAKO).

Quantitative evaluation of TCR stimulation was validated using TCR cross-linking by increasing concentrations of plate-bound CD3 Ab (no. 14-0032-82; eBioscience) or by stimulation with serially diluted synthetic 9mer ADAMTSL5 peptide presented by stably HLA-C*06:02–transfected COS-7 cells (Supplemental Fig. 2A–C). After 24 h of stimulation, the degree of hybridoma activation was determined with respect to the percentage and mean fluorescence intensity of sGFP-positive (sGFP⁺) hybridoma cells. The percentage of sGFP⁺ hybridoma cells showed a more direct association with the degree of TCR ligation than mean fluorescence intensity of sGFP and was therefore used for quantification of TCR stimulation in this study. The maximal possible yield of 50–70% of sGFP⁺ hybridoma cells reflects their maximum activatability also observed in former experiments that had established this technology (24, 28).

To determine hybridoma stimulation by either parental or ERAP1^−/− WM793 or WM278 cells, cells were seeded in 48-well plates. WM278 cells required incubation with IFN-γ (1 ng/ml) for TCR stimulation. Vα3S1/Vβ13S1–TCR hybridoma cells were added 24 h later. As in all TCR stimulation experiments, sGFP induction in Vα3S1/Vβ13S1–TCR hybridoma cells was examined after 24 h coculture by flow cytometry. As negative/positive controls in all stimulation experiments, Vα3S1/Vβ13S1–TCR hybridoma cells were incubated in culture plates either untreated or precoated with CD3 Ab (17A2, 2 µg/ml in PBS; eBioscience).

To compare stimulation of the Vα3S1/Vβ13S1–TCR hybridoma by ADAMTSL5 8mer and 9mer, stably HLA-C*06:02–transfected COS-7 cells or IFN-γ–treated (1 ng/ml) WM278 cells were incubated with synthetic ADAMTSL5 peptides (10 μg/ml), and sGFP induction of Vα3S1/Vβ13S1–TCR hybridoma cells was measured after 24 h coculture (Supplemental Fig. 3B, 3C). ADAMTSL5 peptides were synthesized with purity >95% by Thermo Fisher Scientific.

Cloning of ERAP1 haplotypes has been described (26). For the expression of short antigenic peptides, forward and reverse oligonucleotides were annealed and ligated into pcDNA3.1D/V5-His-TOPO vector using the Directional TOPO Expression Kit (Invitrogen) as described (22).

To compare ERAP1 Hap2 and Hap10 activity, ERAP1^−/− WM793 clones were transfected with pcDNA–ERAP1 Hap2 or Hap10 (100 ng) at 60–80% cell density using FuGENE HD (Promega). Twenty-four hours after transfection, HLA expression was evaluated, or Vα3S1/Vβ13S1–TCR hybridoma cells were added and assessed for sGFP induction after 24 h of coculture by flow cytometry. Parental or ERAP1^−/− HEK293T cells were seeded on 48-well plates and cotransfected with pRSV-HLA-C*06:02 (75 ng), plasmid-encoded ADAMTSL5 peptides (75 ng), and either pcDNA–ERAP1 Hap2, Hap10, or pcDNA vector (75 ng) using FuGENE HD. Twenty-four hours later, Vα3S1/Vβ13S1–TCR hybridoma cells were added. Staining with CD8–PerCP-Cy5.5 (no. 344710; BioLegend) differentiated hybridoma cells from ERAP1^−/− HEK293T cells. The gating strategy to examine TCR hybridoma stimulation is given in Supplemental Fig. 2A.

RNA was isolated from cell lines (RNeasy Mini Kit; QIAGEN). The same amounts of total RNA were reverse transcribed into cDNA with random primers using SuperScript III (Invitrogen) according to the manufacture’s protocol. Quantitative PCR (qPCR) was performed in triplicate using Light Cycler 2.0 (Roche Diagnostics). Porphobilinogen deaminase was used as an internal standard for quantifying ADAMTSL5 mRNA.

Adherent cells were detached using 0.2% EDTA/PBS, washed twice with PBS, and lysed in Cold Spring Harbor Buffer: 50 mM Tris (pH 7.4), 0.25 M NaCl, 1 mM EDTA, and 0.1% Triton X-100, with phosphatase inhibitors (PhosSTOP; Roche Diagnostics) and protease inhibitors (Complete, Mini, EDTA-free; Roche Diagnostics). Protein concentration was measured using the BCA Protein Assay Kit (Pierce). Protein gel electrophoresis and immunoblotting were performed with the same amounts of 5–15 μg of denatured protein lysate by using the XCell SureLock Mini-Cell system with 4–12% gels in MES SDS buffer and PVDF membranes in each experiment (all from Invitrogen). After blocking (no. 11921681001; Roche Diagnostics), blots were incubated with primary Abs (no. AF2334, 1/2000; R&D Systems; no. sc-166088, 1/500; Santa Cruz Biotechnology; no. SAB3500142, 1/1000; Sigma-Aldrich) overnight at 4°C, washed with 0.1% Tween 20 in PBS, and incubated with HRP-conjugated secondary Abs (no. 6885; Abcam; no. 7076S; Cell Signaling; no. 6877; Abcam; diluted as recommended) for 1 h. Subsequently, blots were washed and visualized by chemiluminescence (no. 89168-782; GE Healthcare). Protein levels of β-actin served as a control for constant loading and transfer.

ADAMTSL5 11mer peptide (20 nM) was incubated with purified recombinant ERAP1 Hap2 or Hap10 (2 µg) at 37°C in 50 mM Tris-HCl (pH 8) for up to 2 h. Reactions were stopped by the addition of 0.5% trifluoroacetic acid. Each sample was analyzed for peptide digestion by reversed-phase HPLC (Shimadzu) on a 2.1 × 250 mm C18 column (Vydac) over a gradient of 18–34% acetonitrile and a flow rate of 0.25 ml/min. Peptide peaks were analyzed using LC Solutions software (Shimadzu). Synthetic peptides (20 nM) and buffer-only samples were run and analyzed in identical conditions to establish the ADAMTSL5 peptide series retention times and the absence of cross-contamination (Fig. 5A).

Lesional skin biopsies were obtained from patients with clinically and histologically verified chronic plaque psoriasis, and normal skin specimens were obtained from donors undergoing plastic surgery. Patients and healthy individuals participated voluntarily and gave written informed consent. The study of human material was performed in accordance with the Helsinki Declaration and approved by the Ethics Committee of the Ludwig-Maximilian-University, Munich, Germany. Immunofluorescence stainings were performed as previously described with several modifications (22). Heat-induced Ag retrieval used Tris-EDTA buffer for MART1/ERAP1 and citrate buffer (10 mM citric acid, 0.05% Tween 20 [pH 6]) for HLA-C/c-Kit or HLA-C/ERAP1 staining. Slides were incubated with Abs against ERAP1 (no. AF2334, goat, 1/100; R&D Systems), MART1 (A103, 1/100; Leica), HLA-C (no. sc-166088, 1/100; Santa Cruz Biotechnology), or c-Kit (no. A4502, 1/400; DAKO) and washed with PBS, 0.1% Triton, or 0.1% saponin-containing PBS for ERAP1 staining. For HLA-C staining, slides were blocked with 5% goat serum at 4°C overnight. Ab reactivity was detected using biotinylated anti-goat IgG Ab and streptavidin–Alexa 488, donkey highly cross-adsorbed anti-mouse IgG (H+L) secondary Ab Alexa Fluor 594 (no. A21207; Invitrogen), goat anti-mouse IgG (H+L) cross-absorbed Ab Alexa 488 (no. A11001; Invitrogen), or goat anti-rabbit IgG (H+L) highly cross-adsorbed Ab Alexa 594 (no. A11037; Invitrogen). DAPI counterstained cell nucleoli. MART1 and c-Kit Abs marked the same cell population in the epidermis of healthy skin and psoriatic lesions, and ERAP1 reactivity of the goat Ab overlapped with that of the rabbit ERAP1 Ab (no. ab124669; Abcam), validating the detection methods (Supplemental Fig. 4A).

Epidermal areas of five randomly and continuously selected view fields were photodocumented using an Axio Observer microscope (ZEISS, Visitron). Isotype staining with the same concentration of corresponding isotype Ab in each experiment determined background levels and positivity thresholds. Fluorescence intensity was determined using ImageJ software. The selection gate for whole epidermis or basal layers was defined on DAPI staining and basal layers by one cell width from the basal membrane. MART1⁺ or c-Kit⁺ gate excluded the background staining in keratinized layer and c-Kit⁺ cells in the dermis. The selection gate was transferred onto HLA-C or ERAP1 staining, and fluorescence intensity was quantified for each selection (Supplemental Fig. 4B, 4C). Results represent the median values of five sequential view fields from each individual.

Statistical analyses were performed using GraphPad Prism v.7 (GraphPad Software). Microsoft Excel was used to store data. Kruskal–Wallis H test was used for multiple comparisons, and Bonferroni correction was applied. When the p value of the Kruskal–Wallis H test was significant for comparing multiple groups, two-group comparison was performed using a parametric t test for hybridoma experiments or HLA staining, nonparametric Mann–Whitney U test for unpaired continuous variables, or Wilcoxon signed-rank test for paired clinical samples between two groups. Two-tailed p < 0.05 was considered significant. Although some groups were assessed to have nonnormal distributions in hybridoma stimulation experiments or HLA staining data (Shapiro–Wilk W test), significance outcomes based on the results from Mann–Whitney U test were similar to those from a parametric t test. In hybridoma stimulation experiments, or HLA staining, group sizes were determined based on the results of preliminary experiments, considering the variation and mean of the samples. In principle, no data were excluded from analyses. Some stimulation experiments were excluded when the positive control samples in the experiments or GFP transfection had failed. Investigators were not blinded for samples. All data are not subjective but are based on quantifications.

To investigate the role of ERAP1 in the immunogenicity of melanocytes, we generated ERAP1^−/− clones from two HLA-C*06:02–positive melanoma cell lines, WM793 and WM278, by CRISPR/Cas9 gene editing (Fig. 1A, 1B). These cell lines are potent stimulators of the HLA-C*06:02–restricted autoreactive psoriatic Vα3S1/Vβ13S1 TCR and can replace primary melanocytes for analyzing melanocyte-specific autoreactivity (22). In our experiments, we used each two knockout clones to avoid possible clone-specific effects.

Expression of HLA class I molecules on the cell surface requires binding of peptides (29). The significance of ERAP1 for cell surface HLA class I expression by human cells is still debated. ERAP1 knockout in WM793 and WM278 cells reduced HLA-C cell surface expression measured by the reactivity of an HLA-C Ab (DT-9) by >50%, whereas the overall HLA-ABC expression decreased only moderately (Fig. 1C–H). Although the HLA-C Ab also reacts against HLA-E, HLA-E Ab staining revealed only minimal HLA-E expression in these cell lines, which remained essentially unchanged by ERAP1 knockout and thus had no significant influence on the overall result (Fig. 1I, 1J). Furthermore, DT-9 recognizes HLA-B*35:01 with low affinity (27), which is expressed by WM793 cells. Analysis of WM278 verified that ERAP1 knockout primarily decreased the cell surface expression of HLA-C rather than that of HLA-A or HLA-B, as this cell line does not express HLA-A or HLA-B alleles recognized by DT-9. Western blot analysis revealed unchanged total cellular expression of HLA-C upon ERAP1 knockout (Fig. 1A, 1B), indicating that the lack of ERAP1 activity had not affected the synthesis of HLA-C but caused intracellular HLA-C retention.

To validate these results, we examined the effect of ERAP1 knockout on the HLA class I expression in another cell line, HEK293T cells (Supplemental Fig. 1B–E). These cells require IFN-γ to induce HLA-C expression. As in WM278 and WM793 cells, HLA-C expression was significantly decreased on the cell surface but not intracellularly in ERAP1^−/− HEK293T cells, whereas the overall HLA-ABC cell surface expression as well as HLA-E expression remained largely unaffected. Thus, ERAP1 activity strongly affects HLA-C cell surface expression, presumably by generating self-peptides for presentation.

GWASs had revealed that certain ERAP1 single-nucleotide polymorphisms are associated with the risk of psoriasis in carriers of HLA-C*06:02 (6–8). The different coding ERAP1 single-nucleotide polymorphisms are not transmitted individually but are inherited in different combinations as sets of DNA variations in the ERAP1 gene that form 10 main haplotypes (Hap1–10) (30). Hap2 is a risk haplotype in epistasis with HLA-C*06:02 in psoriasis, whereas Hap10 protects against the disease. To determine the role of the two ERAP1 haplotypes with opposite effects on psoriasis risk in our experimental system of the psoriatic autoimmune response, we restored ERAP1 expression in ERAP1^−/− WM793 clones 1 and 2 with Hap2 or Hap10. Western blot showed comparable protein expression of both ERAP1 haplotypes in each clone (Fig. 2A). Reconstitution of the ERAP1^−/− WM793 clones with Hap2 induced a significantly higher HLA-C surface expression compared with that by Hap10 (Fig. 2B). Thus, the psoriasis risk haplotype, Hap2, mediates higher HLA-C expression than Hap10.

HLA class I expression levels determine target-cell immunogenicity in malignant melanoma and other tumors, and reduced cell membrane HLA expression is thought to contribute to tumor escape from immune recognition (31). To assess the relevance of ERAP1 for the HLA-C*06:02–restricted immunogenicity of melanocytes, we performed stimulation experiments in which we cocultured WM278 and WM793 cells with the Vα3S1/Vβ13S1 TCR hybridoma. We measured the extent of stimulation by the percentage of hybridoma cells induced to express sGFP, which shows a direct association with the degree of TCR ligation (Supplemental Fig. 2A–C) (32). As observed in former experiments that had established this technology (28), depending on the mode of stimulation, it may reach a maximal possible yield of 50–70% activatable cells.

In parallel with the reduced HLA-C expression, ERAP1 knockout greatly reduced the ability of WM793 and WM278 cells to stimulate the HLA-C*06:02–restricted Vα3S1/Vβ13S1 TCR, as measured by the percentage of sGFP⁺ hybridoma cells (Fig. 3A, 3B). As shown by comparison with untransfected ERAP1^−/− cells, reconstitution with the psoriasis risk haplotype Hap2 restored the ability of the ERAP1^−/− melanoma cell clones to stimulate the Vα3S1/Vβ13S1 TCR to a significantly greater extent than reconstitution with Hap10 (Fig. 3C, 3D) at comparable protein expression of both haplotypes (Fig. 2A). Thus, in our in vitro system of the psoriatic autoimmune response, ERAP1 is essential for melanocyte immunogenicity, with the psoriasis risk haplotype, Hap2, conferring higher immunogenicity and cell surface HLA-C expression to melanocytes than the protective haplotype, Hap10.

Stimulation of the Vα3S1/Vβ13S1 TCR by melanocytic cells reflects the autoimmune response against a peptide from the psoriatic autoantigen, ADAMTSL5 (22). Western blot and qPCR analyses showed comparable transcription and expression of ADAMTSL5 in ERAP1^−/− WM278 and WM793 clones and in the parental cell lines (Fig. 3E–G). This rules out that the loss of immunogenicity of ERAP1^−/− WM278 and WM793 cells is due to decreased expression of the source protein and suggests instead that the ADAMTSL5 epitope must initially be produced as longer precursors that are subsequently trimmed in the endoplasmic reticulum.

We therefore investigated the role of ERAP1 in the generation of the ADAMTSL5 epitope and cloned elongated ADAMTSL5 precursors or shortened ADAMTSL5 peptides into pcDNA3.1 expression vector. For presentation, we coexpressed them together with HLA-C*06:02 in HEK293T cells, which are suitable for investigating the immunogenicity of these self-peptides because they lack endogenous antigenicity for the Vα3S1/Vβ13S1 TCR. Peptide stimulation was measured by coculture-induced sGFP expression in Vα3S1/Vβ13S1–TCR hybridoma cells. Using this experimental approach, recombinant expression of ADAMTSL5 peptides with N-terminal extensions up to a length of 12 aa in wild-type HEK293T cells induced a substantial activation of the Vα3S1/Vβ13S1 TCR, although the magnitude of TCR activation gradually decreased with increasing peptide length (Fig. 4A). N-terminal extensions beyond a length of 12 aa strongly reduced the ability of ADAMTSL5 peptides to stimulate the TCR, whereas C-terminally extended peptides did not ligate the Vα3S1/Vβ13S1 TCR (Supplemental Fig. 3A). This is consistent with findings that the proteasome generates the C terminus of antigenic peptides (33). Antigenicity was also lost for ADAMTSL5 peptides truncated at the NH₂ or C terminus, as this leads to a loss of anchor residues for HLA-C*06:02 (Fig. 4A, Supplemental Fig. 3A) (34, 35).

When presented by ERAP1^−/− HEK293T cells, only the plasmid-encoded ADAMTSL5 8mer retained its ability to strongly stimulate the Vα3S1/Vβ13S1 TCR. The ability of ADAMTSL5 peptides 9–11 aa in length for TCR stimulation was much reduced, whereas longer peptides completely lost their ability to stimulate the Vα3S1/Vβ13S1 TCR in the absence of ERAP1 (Fig. 4A). Reconstitution of ERAP1^−/− HEK293T cells with ERAP1 Hap2 or Hap10 restored the antigenicity of the ADAMTSL5 peptide precursors in a stimulation pattern corresponding to the parental cell lines, with Hap2 producing stronger stimulation by the ADAMTSL5 peptides than Hap10. We thus conclude that ERAP1 is involved in the generation of the autoantigenic ADAMTSL5 epitope from precursors.

We then directly compared the activity of the two ERAP1 haplotypes with opposite effects on psoriasis risk with respect to their ability to restore the stimulatory capacity of NH₂-terminally elongated ADAMTSL5 peptides, focusing on the 12mer to 10mer peptides. With similar protein expression (Fig. 4B), reconstitution of ERAP1^−/− HEK293T cells with Hap2 mediated significantly higher stimulation of the Vα3S1/Vβ13S1 TCR by the NH₂-elongated ADAMTSL5 10mer and 11mer than reconstitution with Hap10. Although we observed stronger stimulatory effects of Hap2 also for the ADAMTSL5 12mer, the difference from Hap10 was not statistically significant (Fig. 4A, 4C).

These experiments clearly demonstrate that ADAMTSL5 precursor peptides require N-terminal trimming by ERAP1 to become immunogenic. Stimulation with the plasmid-encoded peptides presented by the ERAP1^−/− cells, however, may not reflect the actual length of the ADAMTSL5 epitope. The plasmid-encoded peptides are produced with an N-terminal methionine that should normally be removed by methionine aminopeptidases or ERAP1. Methionine cleavage by the methionine aminopeptidases, however, may be specifically hindered by valine and arginine at the N terminus of the ADAMTSL5 9mer and 8mer (36, 37) so that the plasmid-encoded 9mer may not be adequately presented by HLA-C*06:02. The plasmid-encoded 8mer might instead become a 9mer through the N-terminal methionine and thus stimulate the Vα3S1/Vβ13S1 TCR. As synthetic peptides, both the ADAMTSL5 9mer and 8mer activate the Vα3S1/Vβ13S1 TCR, whereas consistent with the predicted binding affinity for HLA-C*06:02 by NetPanMHC 4.0, the 9mer stimulates the TCR hybridoma more strongly than the 8mer (Supplemental Fig. 3B, 3C).

We finally investigated the ability of the ERAP1 haplotypes to trim NH₂-elongated ADAMTSL5 precursor peptides by in vitro digestion experiments. We selected the NH₂-elongated ADAMTSL5 11mer for this analysis because the difference in digestion by ERAP1 Hap2 and Hap10 was most pronounced in the coculture experiments for peptides of this length (Fig. 4C). Direct exposure of synthetic ADAMTSL5 11mer to recombinant ERAP1 haplotypes confirmed that ERAP1 trims NH₂-terminally elongated ADAMTSL5 peptides to the length that stimulates the Vα3S1/Vβ13S1 TCR, although HPLC analysis of the digested products was limited in distinguishing 9mer from 8mer (Fig. 5A, 5B). With repeated N-terminal excisions at the same peptide, ERAP1 showed de facto ability for processivity, as formerly suggested by crystal structure analysis (38). The trimming experiments further clearly revealed that Hap2 cleaved the precursor peptides much more efficiently and, in a given time period, produced more of the mature ADAMTSL5 epitope from NH₂-elongated precursors than did Hap10 (Fig. 5B, 5C). After 1 h, Hap2 had trimmed 89% of the ADAMTSL5 11mer to generate shorter peptides, whereas 69% remained untrimmed by Hap10. Compared with Hap10, the psoriasis risk haplotype, Hap2, thus provides a higher supply of the psoriatic autoantigen from NH₂-elongated precursor peptides.

Together, these results reveal that the ADAMTSL5 peptide belongs to the ERAP1-dependent peptidome. The ERAP1 risk haplotype generates more ADAMTSL5 peptide and thus likely increases the immunogenicity of melanocytes by creating a higher density of self-peptide/HLA-C*06:02 complexes on the cell surface of these autoimmune target cells for stimulation of ADAMTSL5-specific CD8⁺ T cells.

Our results from the in vitro model of the psoriatic autoimmune response reveal a mechanism by which ERAP1 and HLA-C determine the autoimmunogenicity of melanocytes and control the development of an autoimmune response in psoriasis in a mutually dependent manner. To examine the relevance of this mechanism directly in patients, we compared the expression of ERAP1 and HLA-C in healthy skin and psoriasis lesions by immunofluorescence staining. Fluorescence intensity was quantified for the whole epidermis, the basal epidermal cell layer, and specifically melanocytes, which we distinguished by staining with MART1 (melan-A) or c-kit (CD117) Abs (Supplemental Fig. 4A).

By immunohistologic staining and analysis of the National Center for Biotechnology Information GDS4602 data (https://www.ncbi.nlm.nih.gov/sites/GDSbrowser?acc=GDS4602) (39), we found that transcription and protein expression of both ERAP1 and HLA-C are dramatically upregulated in psoriasis lesions as compared with those in healthy skin (Fig. 6A–D). More specifically, we observed that in the epidermis of normal skin, ERAP1 expression was more pronounced in the basal epidermal layer (Fig. 6A, 6B, 6D, Supplemental Fig. 4B–D), which consists of germinative keratinocytes and melanocytes. Compared with normal skin, the inflammatory psoriasis lesions showed an overall increase in ERAP1, especially in basal layer cells, where it was most pronounced in melanocytes (Fig. 6A, 6B, 6D, Supplemental Fig. 4B–D). This indicates a high enzymatic activity of ERAP1 in psoriatic melanocytes, as it is related to the ERAP1 expression level (40). In healthy skin, melanocytes showed the highest HLA-C expression, whereas otherwise, HLA-C had a mainly suprabasal distribution, as reported (41). In psoriasis lesions, melanocytes exhibited a significant increase in HLA-C (Fig. 6A, 6B).

Thus, melanocytes in psoriatic lesions are characterized by a combined upregulation of the proteins encoded by the two most important genes for psoriasis susceptibility, HLA-C and ERAP1 (Fig. 6D). This is suggestive of an enhanced functional interaction of ERAP1 and HLA-C, ultimately leading to higher melanocyte autoimmunogenicity and potentiating the autoimmune CD8⁺ T cell response in psoriasis.

The results of our study reveal that in psoriasis, the HLA-C*06:02–restricted autoimmune response against melanocytes is controlled by ERAP1 because the generation of a melanocyte autoantigen for presentation by HLA-C*06:02 is ERAP1 dependent. Our data furthermore show that ERAP1 variants determine the level of melanocyte autoimmunogenicity in psoriasis by generating different amounts of the autoantigenic ADAMTSL5 epitope. This mechanism functionally explains how epistasis between HLA-C*06:02 and ERAP1 variants, which is statistically defined by GWAS, affects the risk for psoriasis. The density of presented Ags on the cell surface crucially determines priming and recall responses of CD8⁺ T cells (42, 43). The generation of a greater amount of ADAMTSL5 peptides by the ERAP1 risk haplotype should cause a higher density of self-peptide/HLA-C*06:02 complexes on the cell surface of the autoimmune target cells and thus increase the activation of autoreactive CD8⁺ T cells, as reflected by the direct relationship between Ag concentration and TCR hybridoma activation (Supplemental Fig. 2C). Consequently, different trimming efficacies of different ERAP1 haplotypes for the autoantigen translate directly into different degrees of HLA class I–restricted immunogenicity of the autoimmune target cells. This may explain why the increased production of the melanocyte autoantigen by the psoriasis risk haplotype, Hap2, mediates a greater risk of psoriasis in HLA-C*06:02 carriers as opposed to Hap10. These findings propose that ERAP1 has a true gatekeeper function for HLA class I–restricted autoimmune responses, and they define the enzymatic activity of ERAP1 haplotypes as a central checkpoint in the development of psoriasis.

Both stimulation experiments with NH₂-elongated ADAMTSL5 peptides and the ERAP1 digestion data of ADAMTSL5 precursor peptides clearly demonstrate the ability of ERAP1 to successively remove the N-terminal residues of ADAMTSL5 precursors, shortening the peptide to a length required for HLA-C*06:02 binding, respectively. Our results thus reveal how the HLA class I–restricted immunogenicity of the target cells of the psoriatic autoimmune response emerges. The findings from psoriasis propose that the yet-unknown (auto)antigens of ankylosing spondylitis and Behçet disease, which show strong statistical epistasis between ERAP1 variants and the HLA class I risk alleles (4, 5), also belong to the ERAP1-dependent peptidome and become presented by the disease-associated HLA class I molecules to elicit pathogenic CD8⁺ T cell responses. Clonal CD8⁺ T cell expansions in tissue lesions of ankylosing spondylitis indicate extensive Ag-driven activation and support a CD8⁺ T cell–driven autoimmune pathogenesis (44), similar to what we have shown in psoriasis patients (20, 22, 45). In skin lesions of Behçet disease, we find that Ag-activated CD8⁺ T cells appear to drive neutrophil recruitment and vasculitis by a strong IL-17A production (46). Other diseases also involving epistasis between HLA class I alleles and ERAP1 variants, such as birdshot chorioretinopathy, type I diabetes, multiple sclerosis, and Crohn disease (reviewed in Ref. 47) may also involve this ERAP1-dependent pathogenic pathway. As such, our experimental work on psoriasis may provide insights into the still-controversial role of ERAP1 and HLA class I risk alleles in other autoimmune diseases.

HLA class I molecules are transported from the endoplasmic reticulum to the cell surface only if they have bound peptides for presentation. HLA-C cell surface expression is strongly influenced by the diversity of peptides available for HLA-C binding (48). In our experimental system, the cell surface expression of HLA-C on melanocytes was significantly reduced by ERAP1 knockout compared with the total HLA class I expression comprising HLA-A, HLA-B, and HLA-C. Thus, HLA-C expression in humans resembles the selective ERAP1 dependence of H-2L^d in mice, which decreased by ∼70% by ERAP1 knockout (25). We therefore conclude that ERAP1 may generate a certain proportion of the immunopeptidome required for HLA-C expression and thus increase the availability of peptides of suitable length and sequence for HLA-C binding. This strongly emphasizes that ERAP1 activity may be of much greater importance for HLA-C– than HLA-A– or HLA-B–restricted immune responses. Furthermore, the different autoantigen densities generated by ERAP1 Hap2 and Hap10 for presentation may not only influence the probability and strength of the autoimmune response but also have possible effects on the cytokine pattern of the stimulated T cells, as has already been shown for CD4⁺ T cell responses (49, 50).

Cell surface expression of HLA-C molecules is very low compared with that of HLA-A and HLA-B molecules and requires IFN-γ or other proinflammatory signals for a substantial upregulation (51, 52). Peptides that can bind to HLA-C may therefore be largely sequestered from presentation on the cell surface under immunologically quiet conditions and thus be ignored by T cells. They may, however, become immunogenic under circumstances upregulating HLA-C. By inducing ERAP1, IFN-γ increases the amount of ERAP1-generated self-peptides (53), and by inducing HLA-C (51), IFN-γ promotes their presentation. Upregulation of HLA class II molecule expression likely plays an important role in initiating the autoimmune response in HLA class II–associated autoimmune diseases by enhancing Ag presentation to T cells (54). In a similar fashion, the coordinated upregulation of the two functionally linked molecules by IFN-γ in psoriasis, which is evident from our immunohistologic study of psoriatic skin sections, should enhance the immunogenicity of melanocytes. Our data thus propose that an increase in the density of HLA class I complexed autoantigens on the cell surface by IFN-γ is an important mechanism amplifying autoreactive CD8⁺ T cell activation in psoriasis and presumably also other HLA class I–associated diseases.

Overall, using melanocytic cells as well-defined target cells and ADAMTSL5 as causative autoantigen of the HLA-C*06:02–restricted autoimmune response in psoriasis, our study unravels essential aspects of the functional mechanisms underlying the statistical epistasis between ERAP1 variants and particular HLA class I alleles in the risk of HLA class I–associated autoimmune diseases. Our data thus establish a plausible pathogenic concept for these diseases in which ERAP1 generates the autoantigenic epitopes for presentation by the respective disease-associated HLA class I molecule to CD8⁺ T cells. Different ERAP1 haplotypes then control the likelihood and strength of the autoimmune response through generating different amounts of autoantigen for HLA class I presentation. In this way, ERAP1 determines the HLA class I–restricted immunogenicity of the autoimmune target cells and the subsequent risk of autoimmune CD8⁺ T cell activation. Specific generation of certain autoantigens by ERAP1 may thus serve as an essential checkpoint in HLA class I–restricted CD8⁺ T cell–mediated autoimmune diseases, suggesting ERAP1 activity as a promising therapeutic target for the downregulation of autoimmune pathology in psoriasis and potentially other diseases with a similar genetic predisposition. Thus, our study uncovers underlying functional mechanisms of the gene–gene interaction between ERAP1 variants and HLA class I risk alleles identified by GWASs (4–8), providing important, and to our knowledge, novel insights into the pathomechanisms of common human autoimmune diseases.

We thank Ulla Knaus, Takashi K. Satoh, Seçil Vural, and Noriaki Sasai for critical reading. We thank C. Kammerbauer, U. Puchta, and A. Yanagida for technical advice. We acknowledge Life Science Editor Helen Pickersgill for editorial assistance.

The authors have no financial conflicts of interest.