ER-aminopeptidase 1 determines the processing and presentation of an immunotherapy-relevant melanoma epitope

Dissecting the different steps of the processing and presentation of tumor-associated antigens is a key aspect of immunotherapies enabling to tackle the immune response evasion attempts of cancer cells. The immunodominant glycoprotein gp100 209-217 epitope, which is liberated from the melanoma differentiation antigen gp100 PMEL17 , is part of immunotherapy trials. By analyzing different human melanoma cell lines, we here demonstrate that a pool of N-terminal extended peptides sharing the common minimal epitope is generated by melanoma proteasome subtypes. In vitro and in cellulo experiments indicate that ER-resident aminopeptidase 1 (ERAP1)—but not ERAP2— deﬁnes the processing of this peptide pool thereby modulating the T-cell recognition of melanoma cells. By combining the outcomes of our studies and others, we can sketch the complex processing and endogenous presentation pathway of the gp100 209-217 -containing epitope/peptides, which are produced by proteasomes and are translocated to the vesic-ular compartment through different pathways, where the precursor peptides that reach the endoplasmic reticulum are further processed by ERAP1. The latter step enhances the activation of epitope-speciﬁc T lymphocytes, which might be a target to improve the efﬁciency of anti-melanoma immunotherapy. cells expressing HLA-A*02:01 · LC-MS : liquid-chromatography–mass spectrometry · MECL-1 : multicatalytic endopeptidase complex like 1 · MS : mass spectrometry · PMEL/Pmel17 : premelanosomeprotein · QME :quantiﬁcationwithminimaleffort · SCS : site-speciﬁc cleavage strength · TPP2 : tripeptidyl peptidase 2


Introduction
Recognition and destruction of malignant cells by CD8 + cytotoxic T lymphocytes (CTLs) are the underlying mechanisms of the three leading approaches in cancer immunotherapy. At first, injection of antibodies blocking the immune checkpoints CTLA-4 or PD-1, which causes activation of patient's T cells, has emerged as a breakthrough in antitumor therapy. Secondly and thirdly, adoptive transfer of tumor-specific CTLs and therapeutic vaccination are under preclinical as well as clinical investigation. A key factor in these latter immunotherapies is the selection of optimal epitopes that are presented by major histocompatibility complex (MHC) class I molecules to CTLs [1][2][3].
Tumor-expressed T cell epitopes are generated, transported, and presented at the cell surface by MHC class I molecules via the antigen processing and presentation pathway (APP). The majority of these epitopes are produced by proteasome complexes, which are the central proteolytic unities of the ubiquitin-proteasomesystem. Proteasomes generate MHC class I-restricted epitopes by canonical peptide-bond hydrolysis as well as by post-translational peptide splicing [4]. The proteasome's 20S catalytic core particle is arranged as four staggered rings, each containing seven nonidentical subunits. The outer rings contain the α subunits (α1-α7), which form the 'gates' through which substrates enter and products are released. Each of the two inner rings contains the β subunits (β1-β7), three of which (β1, β2, and β5) harbor the six active sites. In some hematopoietic cell types and in the presence of type I and/or type II interferons, the β1, β2, and β5 standard subunits are replaced by their alternative variants, the immunosubunits β1i/LMP2 (low molecular weight protein), β2i/MECL-1 (multicatalytic endopeptidase complex like 1), and β5i/LMP7, thereby forming de-novo synthesized immunoproteasomes [5,6]. Besides standard-and immunoproteasome complexes, expression of intermediate-or mixed-type proteasomes containing both standard-and immunoproteasome subunits have been detected in various tumor cells as well as in healthy tissue [7][8][9]. These proteasome isoforms quantitatively differ in their capability to produce peptides although the existence of proteasome-specific substrate cleavage sites and peptide products is still a matter of discussion [7,8,[10][11][12][13][14][15].
Following generation by proteasomes, the epitopes and their precursor-peptides can be destroyed or further processed in the cytosol by aminopeptidases such as tripeptidyl peptidase 2 (TPP2) [16]. Peptides that survive this APP step are often translocated into the ER lumen via TAP, where peptides can be further trimmed by the IFN-γ inducible ER-resident aminopeptidases ERAP1 and/or ERAP2 [17,18]. In addition, ERAP1/2 has been shown to be constitutively expressed in different tissues and tumor cells, where the ERAP1 level correlated with MHC class I expression [19][20][21]. Whereas the generation of various, mainly viral, epitopes is enhanced by ERAP1 [22], some tumor epitopes are destroyed by ERAP1 activity [23,24], thereby suggesting that N-terminal trimming in the ER can be a key step of MHC class I antigen presentation and recognition of both infected and cancer cells [25]. So far, no precise role of ERAP2 in MHC class I epitope generation could be defined, although its specificity has been characterized to be distinct from ERAP1 activity [18,26]. In general, the specific mechanism of ERAP1 and/or ERAP2 function could not be clarified so far. Although ERAP1 and ERAP2 have been shown to trim peptides in solution separately and also completing each other's activity, ERAP1/2 heterodimer formation has been proposed to exist in cells. However, the exact nature of the in vivo heterodimer could not be demonstrated so far. An artificially created ERAP1-ERAP2 heterodimer was shown to change the enzymatic parameters of ERAP1 leading to increased peptide trimming efficacy of the ERAP1/2 complex in vitro [27]. Furthermore, ERAP1-ERAP2 heterodimer constructs have been shown to trim precursor peptides while they are bound to MHC class I molecules, albeit no direct ERAP/MHC interaction could be detected [28]. Peptides are further handled by the peptide loading complex, which inserts the peptide into the binding groove of the MHC class I molecules according to their sequence affinity [29,30].
Initially identified by the Rosenberg lab, the melanomaassociated glycoprotein 100 (gp100) 209-217 epitope has been frequently targeted in clinical anti-melanoma trials, mainly in its anchor-modified form (gp100 209-217/(T210M) ) to enhance HLA-A*02:01 binding affinity [31][32][33][34]. The T210M substitution not only improves the binding of the gp100 209-217/T210M epitope to HLA-A*02:01 but also alters significantly the proteasomal processing of that antigenic sequence [35]. The gp100 209-217 epitope is derived from the melanocyte differentiation protein gp100 (also called PMEL or Pmel17), which belongs to the group of differentiation antigens expressed by tumor cells and healthy tissue of origin. PMEL can be frequently detected in metastatic melanomas making it a suitable target for cancer immunotherapy [36].
The gp100 209-217/T210M epitope can be recognized by CTLs not only in its minimal version but also in its N-terminal extended versions gp100 207-217/T210M and gp100 208-217/T210M . The latter peptides bind the HLA-A*02:01 complex with lower affinity than the gp100 209-217/T210M epitope. Any N-extended version of gp100 209-217 (from gp100 205-217 to gp100 208-217 ) produced by proteasomes can be trimmed in vitro by ERAP1 to the length of the minimal gp100 209-217 epitope, whereas further degradation seems to be prevented [35]. According to these results, ERAPmediated trimming of this pool of peptides in melanoma cells should determine the CTL response and therefore affect the efficacy of immunotherapies targeting this epitope.
To test this hypothesis, we investigated in vitro and in cellulo how ERAP1 and ERAP2 control the activation of the gp100 209-217/(T210M) -specific CTL clone, which recognizes both WT gp100 209-217 and mutant gp100 209-217/(T210M) epitopes. The experimental outcomes provided the missing information to construct the APP model of this critical melanoma antigen.

Cleavage preferences of proteasome subtypes expressed in human melanoma cell lines
Since the generation and presentation of the pool of gp100 209-217containing peptides is influenced by the proteasome isoform content of the cell [10,12,13,35], we investigated the proteasome catalytic-subunit composition in two human melanoma cell lines, i.e. the gp100-expressing Ma-Mel-63a cells and the gp100negative Ma-Mel-86a cells (Supporting Information Fig. 1A). Both melanoma cell lines expressed a mixed population of proteasome isoforms including what we inferred being the intermediatetype proteasome carrying the β1/delta, β5i/LMP7, and β2/Z or the β1i/LMP2, β5i/LMP7, and β2/Z subunits (Fig. 1A). Accordingly, analysis of purified proteasomes by two-dimensional gel electrophoresis revealed the assembly of intermediate-type proteasomes containing the two immunosubunits ß1i/LMP2 and ß5i/LMP7 in the absence of ß2i/MECL-1 (Supporting Information Fig. 1B). However, compared to isolated human spleen, ß1i and ß5i expression was considerably less pronounced in both melanoma cell lines. Based on these observations, we conclude that the melanoma cell lines Ma-Mel-63a and Ma-Mel-86a express intermediate-type proteasomes containing ß1i/LMP2, ß5i/LMP7 and ß2/Z.
Exposure of melanoma cell lines to INF-γ, which could be a frequent situation in the tumor and microenvironment, resulted in the up-regulation of all three immunosubunits including the β2i/MECL-1 subunit (Fig. 1B), thereby suggesting that the and Ma-Mel-86a cells upon IFN-γ treatment analyzed by immunoblotting using antibodies, as indicated. Melanoma cells were exposed to 200 U/mL IFN-γ for 48 h and compared to HeLa cells (two biological replicates, one out of two independent experiments is shown). GAPDH served as loading control. modification of the proteasome isoform content is modifiable in these cell lines.
To investigate the specific proteolytic activity of melanomaderived 20S proteasomes we performed in vitro processing experiments of the synthetic peptide gp100 201-230 using 20S proteasomes purified from the two melanoma cell lines. The substrate and peptide products were measured by mass spectrometry (MS) and the MS outcomes were analyzed by the quantification with minimal effort (QME) method, which allowed to quantify the amount of each peptide product and to compute the substrate cleavage-site usage, i.e. the substrate site-specific cleavage strength (SCS) [37]. Correlation between in vitro experiments carried out with purified 20S proteasomes and in cellulo and in vivo experiments has been demonstrated in various studies [7,8,12,13,[38][39][40][41][42][43][44][45][46][47]. Proteasomes were purified from cell lines grown in the absence of inflammatory stimuli to analyze their baseline activity. Consistently with previous studies [10,12,13,35], the minimal gp100 209-217 epitope was not detectable in melanoma proteasome digestions, whereas its N-terminal extended versions could be quantified ( Fig. 2A). In terms of substrate degradation and preferences of cleavage within the synthetic gp100 201-230 peptide, the proteasomes purified from the two melanoma cell lines showed a similar degradation rate and a similar cleavage pattern, which differed from that  of proteasomes purified from human erythrocytes (standard proteasome) and spleen (immunoproteasome-enriched) ( Fig. 2A and B). 20S melanoma proteasomes produced the gp100 205-217 peptide in considerably larger amounts than the shorter N-extended versions of the gp100 209-217 epitope ( Fig. 2A and C). This was mediated by a strong cut between S 204 and S 205 at the gp100 205-217 N-terminus (Fig. 2B). All substrate cleavage sites used by erythrocyte and spleen proteasomes were used also by melanoma proteasomes, thereby confirming the hypothesis that proteasome isoforms largely cleave after the same residues although with significant quantitative differences [10,37]. The SCSs of erythrocyte and spleen proteasomes were similar to those of a previous study carried out with 20S proteasome purified from these specimens [10], thereby confirming the robustness of this type of assay and analysis.

ERAP1 provokes degradation of the gp100 205-217 substrate as well as gp100 209-217 epitope generation
As already anticipated, 20S proteasomes obtained from melanoma cell lines did not produce the minimal gp100 209-217 epitope in detectable amounts, confirming the result of previous studies carried out with various proteasomes and conditions [10,12,13,35]. www.eji-journal.eu Table 1. Kcat/KM of the trimming of N-extended versions of the minimal gp100 209-217 epitope by recombinant ERAP1. The Michaelis-Menten constants were computed from the in vitro digestions of the 50 µM synthetic peptides gp100 205-217 , gp100 206-217 , gp100 207-217 , gp100 208-217 , gp100 209-217 by 2.5 ng recombinant ERAP1 over time. The MS-measured results of the digestions were published previously [35]. Means and standard deviations of two independent experiments are shown. Kinetic parameters were estimated in a Bayesian framework resulting in parameter distributions and therefore providing confidence estimates (see also Supporting Information Fig. 2 Two N-extended versions -i.e. gp100 208-217 and gp100 207-217have been shown to induce CTL activation in vitro [35]. They are produced in significant smaller amounts than the gp100 205-217 peptide by 20S proteasome purified from melanoma cell lines ( Fig. 2A) and other cell types [35], though. These quantitative observations suggest that the trimming of the N-extended versions of the minimal gp100 209-217 epitope could play a significant role in the presentation of the gp100 209-217 epitope by melanoma cells to CTLs. Therefore, we computed the in vitro efficiency in trimming N-extended versions of the minimal gp100 209-217 epitope by recombinant ERAP1. To this end we made use of a mathematical model of ERAP1 activity and inferred the kinetic parameters based on the quantitative substrate degradation measurements of the digestions of the gp100 205-217 , gp100 206-217 , gp100 207-217 , gp100 208-217 , and gp100 209-217 synthetic peptides previously published [35] (Supporting Information Fig. 2A). The degradation kinetics of the substrates recapitulated a Michaelis-Menten-like reaction. Thus, we set out to estimate the Michaelis-Menten parameters kcat, KM, and kcat/KM (Supporting Information Fig.  2B), and reported the kcat/KM values (Table 1), which gave the best confidence in our model. The estimated kcat/KM value of gp100 209-217 peptide was significantly lower compared to the parameters of the N-extended versions, thereby reflecting the fact that the epitope was not trimmed by ERAP1 in those conditions. The ERAP1-mediated trimming efficiency for the substrates gp100 208-217 , gp100 207-217 , and gp100 205-217 was comparable. On the contrary, the substrate gp100 206-217 was processed with highest efficiency, most likely due to a higher kcat (Table 1 and Supporting Information Fig. 2B).
Normal concentration of ERAP2 could generate only a small amount of the gp100 207-217 epitope ( Fig. 3A and C), which could be enhanced by using high ERAP2 enzyme concentrations (Supporting Information Fig. 3B and C). No smaller peptide products could be clearly identified when we used the synthetic peptides gp100 207-217 , gp100 208-217 , and gp100 209-217 as substrates (Supporting Information Fig. 3B and C and data not shown). This limited trimming activity of ERAP2 is substrate specific since the cleavage of the HIV-gp160 313-327 precursor peptide was efficiently carried out as expected (Supporting Information Fig. 4A and B) [17]. In ERAP1/2 digestions, the gp100 205-217 substrate degradation rate directly correlated with the ERAP1:ERAP2 relative ratio. Accordingly, more products were detectable by increasing the ERAP1:ERAP2 ratio, although in those assays where ERAP2 was present, the gp100 206-217 peptide product could not be identified (Fig. 3A-F).
To further test our hypothesis that ERAP1 promotes the generation of immunogenic gp100 209-217 -containing epitopes, we performed in vitro and in cellulo experiments by using the metalloprotease inhibitor leucinethiol, which targets ERAP1. Consistently, addition of leucinethiol to the in vitro digestion resulted in a blockade of gp100 205-217 degradation and stabilization of the precursor peptide ( Fig. 4A and B). Furthermore, activation of a gp100 209-217 -specific CTL clone was significantly reduced by ERAP1 inhibition in Ma-Mel-63a (Fig. 4C) and UKRV-Mel-15a melanoma cells (Fig. 4D), which both express the gp100 antigen and HLA-A*02:01. As activity control, the ability of leucinethiol treated cell lysates to trim the H-Leu-AMC or the H-Arg-AMC substrate was analyzed. As expected, the turnover of the H-Leu-AMC-representing ERAP1 activity-was significantly reduced in the presence of leucinethiol, whereas the H-Arg-AMC substraterepresenting ERAP2 activity-was not affected (Fig. 4E).

ERAP1, but not ERAP2, promotes activation of gp100 209-217 -specific CTL clones by melanoma cell lines
These results suggested a beneficial impact of ERAP1 rather than ERAP2 in defining the presentation of the gp100 209-217containing peptides/epitope by melanoma cells to CTLs. We tested this hypothesis and the ERAP1/2 content in human melanoma cell lines Ma-Mel-63a, Ma-Mel-86a, and UKRV-Mel-15a. They all expressed both ERAP1 and ERAP2, although ERAP2 expression was lower than ERAP1 with different ERAP1:ERAP2 ratio among the melanoma cell lines (Fig. 5A-C). Upon exposure to IFN-γ, ERAP1 expression was enhanced in both Ma-Mel-86a cells and Ma-Mel-63a cells (Fig. 5D) and also in UKRV-Mel-15a cells as shown before [24]. Since ERAP activity could be influenced by the various polymorphisms that these enzymes have, we analyzed the ERAP1 alleles present in the investigated cell lines. Amongst others, we observed two described polymorphisms associated with the risk of cancer development [48,49]. The R127P polymorphism in HeLa, UKRV-Mel-15a and Ma-Mel-63a and the Q730E mutation in Ma-Mel-86a cells suggest an altered peptide trimming activity in the latter [50] (Fig. 5E). To test whether ERAP1 and ERAP2 could impinge upon the presentation of the gp100 209-217 -containing peptides/epitope by melanoma cell lines to CTLs, we performed siRNA knock down of either ERAP1 or ERAP2 in Ma-Mel-63a cells. In agreement with our in vitro results, knocking down ERAP1 significantly reduced the gp100 209-217 -specific CTL activation, whereas silencing ERAP2 displayed no significant alterations of the amount of IFN-γ released by the CTL clone ( Fig. 6A-C). Similar results were obtained when we used UKRV-Mel-15a (Supporting Information Fig. 5A and B) and gp100-transfected HeLa A2+ (HeLa cells expressing HLA-A*02:01) cell line clones (Fig. 6D-F). This hints toward the hypothesis that ERAP1, rather than ERAP2, promotes the activation of gp100 209-217 -specific CTLs by increasing the presentation of gp100 209-217 epitopes by melanoma cell lines as well as by other cancer cell lines.

Discussion
The modified gp100 209-217/(T210M) melanoma epitope has been part of successful phase 2/3 clinical trials combining the gp100 209-217/(T210M) peptide with interleukin-2 application [34]. However, combined administration of gp100 peptides together with the CTLA-4-antagonist ipilimumab in patients with metastatic melanoma displayed no improvement in disease progression compared to patients treated with ipilimumab alone, thereby suggest-ing an insufficient presentation of the gp100 209-217 epitope at least in parts of the melanoma [51,52].
Our results provide further information about the APP of the gp100 209-217 -containing epitopes and specifically the role that ERAP1 rather than ERAP2 plays in promoting epitope presentation. Both in vitro kinetic analyses and in cellulo inhibitor and siRNA experiments demonstrated that ERAP1 generates and stabilizes the minimal gp100 209-217 epitope. This is in contrast to the MART-1 [26][27][28][29][30][31][32][33][34][35] (melanoma antigen recognized by T cells) epitope being destroyed by ERAP1 [24]. ERAP1 has been shown to be constitutively expressed in tumor cells, even in the absence of IFN-γ [19,24]. Analyzing three different melanoma cell lines, we found differences in ERAP1 expression (Fig. 5). We could detect two ERAP1 polymorphisms (R127P and Q730E), which are correlated with an increased risk of cancer development [49,53]. However, only the Q730E mutation has been shown to display reduced peptide trimming activity suggesting an impaired gp100 209-217 epitope generation in melanoma cells expressing the Q730E variant [50]. To note, we did not observe any effect of ERAP2 on gp100 209-217 epitope processing, neither in in vitro digestionsusing ERAP2 alone or in combination with ERAP1-nor in cellular antigen presentation assays analyzing ERAP1/2 activity based on ERAP natural expression in melanoma cells.
By combining the outcome of this and other studies [12,13,35,54], we can sketch how gp100 209-217 -containing epitopes are presented to CTLs: in the cytosol, proteasomes produce two potential epitopes, i.e. gp100 207-217/T210M and gp100 208-217/T210M  -15a cells were exposed to 30 µM leucinethiol for 16 h or to DTT as solvent control and were co-incubated with gp100 209-217 -specific T lymphocytes for 16 h. IFN-γ release by gp100-specific CTLs was measured by IFN-γ ELISA. T2 cells without and with gp100 209-217 peptide were used as control for T cell specificity. Data are displayed as mean + SD of three technical replicates. One independent experiment out of two is shown (Student's t-test; *p < 0.05; **p < 0.01). (E) Specific activity of ERAPs in cell lysates of leucinethiol treated Ma-Mel-63a and UKRV-Mel-15a cells. Specific activity was measured through incubation of the cell lysates with the H-Leu-AMC-or the H-Arg-AMC-substrate, which display ERAP1 and ERAP2 activity, respectively. DTT-treated cells served as control. Data are displayed as mean and SD of two technical replicates (one out of two independent experiments).

that
share the gp100 209-217/T210M minimal sequence IT/MDQVPFSV. The WT gp100 207-217 and gp100 208-217 peptides bind with low affinity to the HLA-A*02:01 complex, therefore their in vivo immunogenicity is disputable, and these are most likely only sources for the gp100 209-217 generation in the ER. In contrast, the T210M gp100 207-217 and gp100 208-217 peptides bind to the HLA-A*02:01 complex with higher affinity and trigger a stronger CTL response in vitro than the WT gp100 209-217 epitope [35]. Furthermore, melanoma proteasomes, which contain intermediate-type proteasomes, as shown for both Ma-Mel-63a and Ma-Mel-86a cells here as well as in previously published literature [7,13], abundantly generate the N-extended version of the gp100 209-217 -epitope, i.e. the gp100 205-217 precursor. In the cytosol, the epitopes (and likely also their N-extended precursors) can be destroyed by TPP2 [54]. The peptides that survive this step can be translocated into the ER lumen via TAPs and to the endosomal compartment by another unknown carrier. In the ER, the peptides gp100 205-217 , gp100 206-217 , gp100 207-217 , and gp100 208-217 can be trimmed by ERAP1-but not by ERAP2-thereby augmenting the amount of the gp100 209-217 -containing epitopes, as shown in the present study and in ref. [ 35]. For those gp100 209-217/T210M -containing epitopes that followed the canonical TAP-dependent pathway, the allocation into the HLA-A*02:01 binding groove is likely mediated by the peptide loading complex. For the portion of the gp100 209-217 epitope that is translocated into the endosomal  pathway via an alternative route, the function of tapasin is not mandatory to efficiently present the gp100 209-217 epitope at the cell surface [54] (Fig. 7).
The APP that triggers a response of gp100 209-217/T210M -specific CTLs is therefore branched and can be carried out with success independently of several proteins/enzymes. This supports the use of this pool of epitopes/peptides in immunotherapy because a diversified APP can be a potent tool to tackle the immune-response escape attempts implemented by cancer. For example, we can speculate that even in case that cancer compromises either TAP or tapasin, the presentation of the gp100 209-217/T210M -containing epitopes could be enhanced in melanoma by inhibiting TPP2 and/or stimulating ERAP1.
To separate the subunits of the 20S proteasome complex, isoelectric focusing by carrier ampholytes was combined with SDS-PAGE. Proteasomes were applied to a carrier ampholyte isoelectric focusing gel. In the second dimension, proteins were loaded onto SDS-PAGE and stained with Coomassie brilliant blue G-250. Proteasome subunits could be identified based on their migration behavior in comparison to reference electrophoreses of selected proteasomes [59].

20S proteasome and recombinant ERAP1/2
20S proteasomes were purified from frozen melanoma cell lines and from human spleen and erythrocytes as described [60]. Recombinant human ERAP1 and ERAP2 were purchased from R&D Systems.
The fluorescence was measured by Tecan fluorometer using excitation 360 nm and emission 460 nm wavelength in a kinetics experiment (0-20 min). The free AMC calibration curve was done by measuring the fluorescence of free AMC at various concentrations (0-1 µM). The synthetic peptide gp100 201-230 (40 µM) was digested by 3 µg of purified 20S proteasomes in 100 µL TEAD buffer (20 mM Tris, 1 mM EDTA, 1 mM NaN 3 , 1 mM DTT, pH 7.2) in kinetic experiments (0-8 h) at 37°C. For the assays performed with melanoma proteasomes we carried out two biological replicates, each of them measured 2-3 times by MS. For the assays performed with erythrocyte and spleen proteasomes we carried out one assay, measured twice by MS since the outcome recapitulated the results described previously [10]. Liquid chromatography-mass spectrometry (LC-MS) analyses of polypeptide digestion products were performed using the ESI-ion trap instrument DECA XP MAX (ThermoFisher Scientific, USA) as previously described [61]. Database searching was performed using SpliceMet's ProteaJ, which allowed the identification of spliced and non-spliced peptide products [61]. Quantification of proteasome-generated peptides and computation of the substrate SCS was carried out by applying the QME (Quantification with Minimal Effort) method to the LC-MS analyses [37]. QME estimated the absolute content of peptide products based on their MS ion peak area measured in the digestion probe. The QME algorithm parameters were empirically computed in our previous study [37] and here applied. SCS describes the relative frequencies of proteasome cleavage after any given residue of the synthetic polypeptide substrates [37]. The SCS values shown in this study are the average of SCS measured over time [37].

Analysis of in vitro digestions of precursor peptides with ERAPs
Fifty micromolar of peptides were digested in vitro with 3 ng recombinant ERAP1 or recombinant ERAP2 in 20 µL assay buffer (25 mM Tris, pH 7.5, 150 mM NaCl, 0.5 µg/mL albumin) at 37°C. For high enzyme concentrations, 25 µM of peptides were digested with 2 µg/mL ERAP1 or ERAP2 [18]. To inhibit ERAP1 activity, 3 ng recombinant enzyme were pre-incubated with 30 µM leucinethiol at room temperature for 20 min and then applied to in vitro processing experiments. For the analysis of the digestion products obtained from the gp100 205-217 precursor peptide, different ERAP1/ERAP2 ratios (pre-incubation 30 min at room temperature) were used, total enzyme concentration (3 ng/20 µL digestion volume) and peptide concentration (50 µM) were constant, recombinant ERAP1/ERAP2 ratio is indicated in Fig. 3. The digestion period was 4 h. Experiments were performed at least twice. Peptide products were identified by LC-MS/MS, as described for in vitro proteasome experiments (see above). Substrates' abundance was quantified through the titration of synthetic peptides.
To analyze the substrate trimming dynamics of ERAP1, we applied a mathematical model describing Michaelis-Menten kinetics. We estimated the kinetic parameters of the model Km and kcat of the trimming of the synthetic substrates gp100 205-217 , gp100 206-217 , gp100 207-217 , gp100 208-217 , gp100 209-217 by purified ERAP1 in two of the kinetics experiments described elsewhere [35].
The model reactions can be depicted as follows: S + E ↔ [SE] → P + E, resulting in the ordinary differential equation model describing substrate degradation over time t: dS/dt = -E*kcat*S / (KM+S), where S is the substrate, P is the sum of all products, E is the enzyme ERAP1; KM is the Michaelis-Menten constant; kcat describes the trimming of the substrate to products P (also maximal velocity). According to our model, although a large range of KM and kcat values could result in a good model fit with the experimental data, the KM and kcat were strongly correlated (Supporting Information Fig. 2B). Hence, we could compute ERAP1 efficiency as the ratio kcat/KM and then derivate the parameter confidence distribution for KM and kcat (Supporting Information  Fig. 2B). The parameters kcat, KM, and kcat/KM were estimated using exact Bayesian inference in a Markov Chain Monte Carlo scheme. The latter resulted in posterior parameter distributions (rather than single point estimates) and therefore we could also provide with an estimate of the parameter uncertainty. The medians and standard deviations (SD) of the marginal posterior parameter distributions are reported in Table 1.
Identification and quantification of ERAP1 in the lysates of different cell lines were performed as described before [62]. LC-MS runs were conducted as follows: samples were trapped and then analyzed by nanoscale LC-MS/MS measurements using a Q Exactive Plus mass spectrometer coupled with an Ultimate 3000 RSLCnano (ThermoFisher Scientific). Protein identification and relative label-free quantification were performed using MaxQuant software version 1.6.0.1 [62] and Andromeda label-free quantification parameters were set to default. Spectra were matched to a human database (20 244 reviewed entries, downloaded from swissprot.org), a contaminant, and decoy database. In addition, protein identifications were calculated with FDR = 1% and proteins with one razor peptide per protein were used for identification. Lysates were digested twice and analyzed three times.

Identification of ERAP1 allelic variants
To identify the expressed ERAP1 alleles the full length ERAP1 construct derived from total RNA of HeLa A2+ cells, Ma-Mel-63a, Ma-Mel-86a and UKRV-Mel-15a cells was cloned into pCR 2.1 plasmid via TA cloning Kit (Thermo Fisher Scientific) and subsequently sequenced and analyzed for previously described allelic variants according to Yao et al. [64].