Discovery of an orally available potent ER aminopeptidase 1 (ERAP1) inhibitor that enhances anti-tumor responses and limits inflammatory autoimmunity in vivo

Endoplasmic reticulum aminopeptidase 1 (ERAP1) is an intracellular enzyme that can regulate immune responses primarily by proteolytically processing peptides before loading and presentation on the cell surface by m ajor histocompatibility class I molecules (MHC -I). ERAP1 activity can either reduce the immunogenicity of cancer cells by over-trimming cancer-associated antigenic peptides or contribute to autoimmunity by generating self -antigenic peptides. As a result, ERAP1 inhibition has emerged as a tractable approach for cancer immunotherapy and specific classes of autoimmunity. Here, we describe the discovery, after hit -to-lead optimization, of a potent and selective ERAP1 inhibitor based on the pyrrolidine 3 -carboxylic acid scaffold that ta rgets the regulatory allosteric site. The compound has favourable in vivo pharmacokinetics, including oral bioavailability, and can regulate the immunopeptidome of cancer cells and enhance cancer cell antigenicity in vivo in a dose- dependent manner, controlling tumor growth. In addition, when administered in the murine collagen- induced arthritis model, it does not induce any exacerbation of autoimmune responses but rather results in a dose-dependent therapeutic benefit. Our results demonstrate that ERAP1 inhibition can constitute a tractable approach to modulating immune responses for therapeutic applications, providing mechanistic insight and a valuable lead and in vivo tool for further drug development efforts and for interrogating ERAP1 biology. preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 2

T cells of t he adaptive immune system recognize, via the T cell receptor (TCR), infected or aberrant cells on the basis of small peptides bound onto major histocompatibility complex (MHC) molecules on the cell surface. MHC class I molecules (MHC-I) fold inside the endoplasmic reticulum (ER) by binding peptides that originate from intracellular proteins and then translocat e to the cell surface. 1 TCRs recognize these MHC -I/peptide complexes thus gaining insight on the proteomic content of the cell. 2 The presence of peptides that belong to pathogens or aberrantly expressed proteins elicit s recognition by T cells and signifies infection or cancerous transformation, inducing molecular responses that lead to the killing of the target cell. The MHC-presented peptides, collectively named the immunopeptidome, are generated by complex proteolytic cascades that often include the proteasome or its immune system counterpart, the immunoproteasome. 3 Peptide fragments from intracellular proteins are then translocated into the ER by a specialized transporter called t ransporter associated with antigen presentation (TAP).4 However, since MHC-I molecules bind peptides with strong selectivity for length, many ER imported peptides require additional processing to enable binding . To achieve this, ER - resident aminopeptidases trim those peptides , optimizing their length for binding onto nascent MHC -I molecules (on average 9 amino acids length). 5 One such aminopeptidase, ER aminopeptidase 1 (ERAP1) is highly specialized for this task and can efficiently prepare correct-length peptides for MHC- I binding, although it can also over -trim some peptides to lengths that preclude MHC -I binding, essentially destroying their antigenic potential. 6,7 Although ERAP1 is typically retained into the ER, it can be recruited to endosomes in dendritic cells where it can process peptides from internalized proteins for cross -presentation, a process whereby peptides from extracellular proteins are presented in association with MHC -I molecules.8 Other functions have also been proposed for ERAP1, although these are less explored. For instance, under certain inflammatory conditions, ERAP1 can be secreted by immune cells, enhancing innate immune responses through the activation of the inflammasome complex.9,10 Several studies have shown that genetic down- regulation of ERAP1 in cells , or its mouse orthologue endoplasmic reticulum aminopeptidase associated with antigen processing ( ERAAP), can strongly enhance the antigenicity of cells.7 These effects have been primarily attributed to changes in the cellular immunopeptidome, the sum of presented peptides by MHC-I molecules. Several molecular and cellular mechanisms have been proposed that may synergize depending on the inflammatory and cellular context: i) presentation of longer length, unstable peptides that are recognized as foreign by TCRs 7 or fail to productively engage inhibitory receptors on natural killer ( NK) cells,11,12 ii) presentation of a different repertoire of good -binding peptides that is recognized by existing T cells or can lead to the maturation of new effector T cells 13 and iii) reduced presentation of peptides derived from the MHC -I signal sequence by HLA-E.14,15 ERAP1 expression levels regulated by expression quantitative trait loci have been reported to regulate cancer immunotherapy responses, 16. Additionally, ERAP1 activity has been correlated with CD8+ T cell tumor infiltration17 and ERAP1 knockdown has been shown to synergize with other agents in promoting the efficacy of cancer immunotherapy.12,15 Thus, an increasing amount of evidence suggests that ERAP1 may be a tractable target for cancer immunotherapy approaches. MHC (HLA in humans) are amongst the most polymorphic genes in humans with over 20, 000 different alleles identified to date. 18 This polymorphic variation is primarily focused on the antigenic peptide binding groove and determines the breadth of possible peptides that can be presented by each individual, defining the context for variable immune responses within human populations. A subset of inflammatory diseases of autoimmune aetiology, called MHC-I-opathies, is very strongly correlated with specific HLA class I alleles, likely due to the presentation of self -antigenic peptides or to changes in folding of MHC-I/peptide complexes.19 ERAP1 is also polymorphic, and ERAP1 coding single nucleotide polymorphisms ( SNPs) have been correlated, often in epistasis with MHC -I haplotypes, with predisposition to MHC -I-opathies, most notably Ankylosing Spondylitis. 20 ERAP1 coding SNPs affect enzymatic activity21 and are found in human populations in allotypes with variable penetrance and highly variable enzymatic properties.22,23 Several inhibitors of ERAP1 have been reported, discovered either by rational design or screening campaigns. 24,25 A group of sub-micromolar ERAP1 inhibitors reported target the Zn(II)-containing active site of the enzyme and utilize phosphinic pseudopeptide, diaminobenzoic acid, or α ‐hydroxy-β-amino preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 3 acid scaffolds, but present limitations in selectivity possibly due to the highly conserved active site among M1 aminopeptidases. 26-28 Allosteric sites that allow for superior selectivity have been also explored, leading to n anomolar potency inhibitors targeting one of two allosteric sites found to be occupied by buffer components in a high- resolution structure of ERAP1. 29-32 The malate allosteric site has attracted significant interest due to its functional significance in recognizing the C -terminus of ERAP1’s peptidic substrates and regulating length selectivity. 33 Inhibitors targeting the active site or allosteric sites have also been shown to be able to modulate the immunopeptidome of cancer cell lines, albeit in distinct patterns. 34-36 Development of selective small molecules targeting ERAP1 has significantly accelerated, with the first clinical candidate GRWD5769 entering clinical trials in 2023. 37 Here, we report the development of an orally available, potent ERAP1 inhibitor targeting the regulatory site of the enzyme 33 optimized from a series of 3- pyrrolidine carboxylic acids originating from a previously described high-throughput screen (HTS) .31 This compound can regulate the immunopeptidome of cancer cells and regulate cellular immunogenicity in two distinct in vivo models with favourable therapeutic outcomes. Our results provide proof-of-concept that targeting the regulatory site of ERAP1 has significant therapeutic potential and report the first potent in vivo tool for interrogating ERAP1 biology and guide further drug development efforts.

Following a previously described HTS campaign, (3S,4S)-1-(3-cyano-6-methylpyridin-2-yl)-4- isopropylpyrrolidine-3-carboxylic acid 1 was identified as a ligand efficient hit molecule with micromolar inhibition of ERAP1, as measured using a previously described assay monitoring the cleavage of 9-mer antigenic peptide YTAFTIPSI to 8- mer TAFTIPSI (Table 1).31,38 The corresponding enantiomer acid 2 was prepared, which was observed to have approximately equivalent binding potency. Both compounds were found to have high kinetic solubility, consistent with the carboxylic acid moiety and low lipophilicity as measured by ChromlogD7.4. Table 1. Profiling of hit enantiomers 1 and 2 Compound (S,S)-1 (R,R)-2 ERAP1 biochemical pIC50 5.61 5.62 ChromlogD7.4 2.04 2.04 MW / HAC 273 / 20 273 / 20 ERAP1 LE / LLE1 0.38 / 3.6 0.39 / 3.6 Kinetic Solubility (µg/mL) ≥109 ≥86 1LLE based on ChromlogD7.4 A co-crystal structure for both compounds 1 and 2 with ERAP1 was generated and is presented in Figure 1. This confirmed that the hit molecules target the ERAP1 regulatory site, as previously observed for a natural product modulator identified within our laboratories .31 In both co -crystal structures, the carboxylic acid makes charge interactions with Tyr684/Lys685/Arg807, while the isopropyl substituent occupies a proximal hydrophobic pocket. The relative configuration of the two pyrrolidine substituents

in similar spatial arrangement of the acid and isopropyl groups, accounting for the similar potency observed between the two enantiomers. The nitrile group from the aminopyridine forms a hydrogen bonding interaction with Gln881. preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 4 Figure 1. Co-crystal structure of hit pyrrolidine 3-carboxylic acid enantiomers 1 (left) and 2 (right) with ERAP1 The co- crystal structures of compounds 1 and 2 indicated that the 6- methylpyridine was in close proximity to both and Asn678/Gln730. It was thought that inclusion of hydrogen bond donors or acceptors in this region could gain productive hydrogen bonding interactions with either of these sidechains to further increase binding potency, as outlined in Table 2. Based on the similar binding potency of the two enantiomers, compounds were prepared as racemates, with enantiomeric separation or chiral synthesis carried out for key analogues.39 For in-line comparison, compound 3 was prepared, a racemic mixture of compounds 1 and 2. Introduction of a hydrogen bond acceptor in 2- methoxypyridine 4 led to no improvement in potency compared to compound 3 . However, introduction of a hydrogen bond donor in 2-aminopyridine 5 led to a significant improvement in biochemical potency, ligand efficiency (LE) and lipophilic ligand efficiency (LLE). Cellular activity of compound 5 was profiled using a previously described assay based on detection of ERAP1-dependent antigenic epitope SIINFEKL in HeLa cells,31 and was found to inhibit antigen presentation at micromolar level (Table 2). The co-crystal structures of compounds 1 and 2 indicated potential close proximity of the appended amine to the proximal hydrophobic pocket. Isopropyl amine 6 afforded the desired biochemical potency improvements, while also improving cellular activity (Table 2). Cyclisation of the isopropyl to cyclopropyl in compound 7 retained binding activity of isopropylamine 6 with improved LLE. Cyclopropyl amine 7 was separated to give both (R,R)-enantiomer 8 and (S,S)-enantiomer 9, which revealed a minor biochemical preference of the ( R,R)-enantiomer. A co -crystal structure of compound 8 with ERAP1 supports that the 6- aminopyridyl group forms the desired hydrogen bonding interaction with Asn678 (Figure 2). preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 5 Table 2. SAR of 6-pyridyl substituents to probe proximal hydrogen bonding interactions Compound R ERAP1 pIC50 HeLa antigen presentation pIC50 ChromlogD7.4 LE LLE Kinetic Solubility (µg/mL) (±)-3 Me 5.671 nt 2.05 0.39 3.6 ≥102 (±)-4 OMe 5.53 nt 2.39 0.36 3.1 ≥118 (±)-5 NH2 6.31 5.481 0.94 0.43 5.4 ≥118 (±)-6 NHiPr 7.45 6.68 2.68 0.44 4.8 ≥109 (±)-7 NHcPr 7.59 6.86 2.38 0.45 5.2 ≥145 (R,R)-8 NHcPr 7.43 6.99 2.52 0.44 4.9 ≥106 (S,S)-9 NHcPr 7.13 6.80 2.44 0.43 4.7 ≥96 1n=2. nt = not tested Figure 2. Co-crystal structure of cyclopropylamine derivative 8 with ERAP1. Further lipophilic analogues occupying the hydrophobic pocket were explored, outlined in Table 3. Cyclopentyl and cyclohexyl analogues 10 and 11 resulted in further increases in binding potency and cellular pIC50>7, though with no LLE inc rease compared to cyclopropyl analogue 9. Introduction of polarity into the more potent cyclopentyl analogues through 3-aminotetrahydrofurans 12 and 13 demonstrated a significant LLE benefit for the preferred ( S)-isomer 13 however cellular activity was reduced. Cyclohexyl analogues such as difluorocyclohexyl compound 14 increased LLE with retained cellular activity, while the low lipophilicity of tetrahydropyran 15 proved detrimental to cellular potency. A compromise in biochemical LLE and cellular potency was observed with aminomethyl cycloalkanes, where cyclopropyl methylamine 16 demonstrated high LE /LLE values and retained cellular potency , while reduced LE/LLE was observed with cyclobutylmethylamine 17 . preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 6 Table 3. SAR of aminopyridine substituents to occupy proximal hydrophobic pocket Compound R ERAP1 pIC50 HeLa antigen presentation pIC50 ChromlogD7.4 LE LLE Kinetic Solubility (µg/mL) (±)-10 8.27 7.44 3.42 0.45 4.9 ≥129 (±)-11 8.52 7.63 3.93 0.45 4.6 ≥112 (±)-12 6.57 5.65 1.50 0.36 5.1 ≥96 (±)-13 8.17 6.66 1.67 0.45 6.5 ≥136 (±)-14 8.34 7.37 3.42 0.41 4.9 ≥167 (±)-15 7.93 5.86 1.86 0.42 6.1 ≥140 (±)-16 8.05 7.54 2.90 0.50 5.2 ≥101 (±)-17 8.15 7.27 3.52 0.45 4.6 ≥116 Based on the crystal structure of compound 9 (Figure 2) a further hydrophobic region at the ortho- position to the nitrile was identified for growth and potential to further balance physicochemical properties. While introduction of a methyl group in this region with compound 18 gave little benefit to biochemical potency when compared to compound 14, cellular potency increased to pIC 50 >8 (Table 4). Separation of the enantiomers of compound 18 again demonstrated a binding preference for (R,R)- enantiomer 19 over ( S,S)-enantiomer 20. F urther variation of the amine group identified 4- methyltetrahydropyran 21 as a potent and cell active binder with reduced lipophilicity compared to compound 19. Profiling in vitro clearance from human hepatocytes for compounds 19 and 21 indicated improved unbound intrinsic clearance ( CLint,u) for compound 21 ( Supplementary Table 1). Crystallography of compound 21 bound to ERAP1 indicated that the key interactions from compound 9 were retained, while an additional interaction was observed between the THP oxygen atom and Gln730 (Figure 3). preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 7 Table 4. SAR following nitrile ortho-methylation with variable amine substituents Compound R ERAP1 pIC50 HeLa antigen presentation pIC50 ChromlogD7.4 LE LLE Kinetic Solubility (µg/mL) Human hepatocyte CLint,u (mL/min/kg) (±)-18 8.42 8.11 3.64 0.41 4.8 ≥108 nt (R,R)-19 8.552 8.09 3.67 0.40 4.9 ≥126 68 (S,S)-20 8.16 7.39 3.72 0.39 4.4 ≥112 nt (R,R)-21 (GSK235) 8.45 7.251 2.44 0.41 6.0 ≥201 39 1n=2. nt = not tested Figure 3. Co-crystal structure of ERAP1 and compound 21. Due to their favourable cellular potency and physicochemical properties, compounds 19 and 21 were progressed to mouse oral pharmacokinetic ( PK) studies to assess their potential as in vivo tool molecules. Despite lower cellular activity, compound 21 was found to have improved unbound oral exposure in mouse compared to compound 19 (Table 5). Therefore compound 21, otherwise known as GSK235, was progressed to further in vivo studies. preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 8 Table 5. Oral PK of compounds 19 and 21 following administration to naïve BALB/c mice (n=3). Compound Dose (mg/kg) Cmax (ng/mL) AUC0-t, (ng.h/mL) fu,blood DNAUC0-t,u (ng.h/mL/mg/kg) 19 10 950 2989 0.002 0.602 21 (GSK235) 10 8516 15440 0.005 7.43 AUC = Area under the curve. DNAUC = Dose-normalised AUC Extended profiling and developability screening of compound 21 is described in Table 6. Compound 21 is a potent ERAP1 inhibitor with high LE/LLE which retains inhibitory activity against mouse ERAAP. Selectivity of over 1000x for the two most homologous enzymes ERAP2 and IRAP was observed for compound 21 based on previously reported assays .31 During the course of optimisation, the original HeLa antigen presentation assay protocol was adjusted such that the data was normalized against a pharmacological tool compound (see Materials and Methods).38 Based on this the cellular potency of compound 21 was adjusted to pIC50 = 7.05, with unbound cellular pIC50 = 7.26, accounting for binding to the cell assay media. Consistent with the low lipophilicity of compound 21, solubility was high based on amorphous solubility and crystalline FaSSIF solubility , with moderate in vitro permeability. Compound 21 was found to be inactive in Ames and MLA assays ; while extended cross-screening across a panel of known liability targets indicated no major flags, other than weak inhibitory activity against OATPB1 and BSEP transporters. No activity against the hERG channel was observed. Table 6. Profile of compound 21 (GSK235) Compound 21 (GSK235) ERAP1 biochemical pIC50 (human, YTAFTIPSI) 8.45 ERAAP biochemical pIC50 (mouse, YTAFTIPSI) 7.59 ERAP2 / IRAP biochemical pIC50 4.02 / 4.28 HeLa antigen presentation pIC50 total / unbound 7.05 / 7.26 MW / ChromlogD7.4 / Acidic pKa 387 / 2.44 / 4.01 Solubility (µg/mL): CAD (amorphous) / FaSSIF (crystalline) ≥201 / 675 AMP / MDCK Papp (nm/s) 28 / 46 Hepatocyte CLint m/r/h (mL/min/g tissue) 0.86 / 2.20 / 1.43 Fraction unbound: media; blood (m/r/h) 0.61; 0.005 / 0.002 / 0.006 hERG pIC50 / Ames / MLA <4 / Negative / Negative Secondary pharmacology cross-screening panel OATP1B1 pIC50 = 5.0, BSEP pIC50 = 4.9 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 9 Profiling of Compound 21 (GSK235) Compound 21 was progressed to profiling in two in vivo mechanistic models with the purpose of investigating the linkage between target engagement and downstream pharmacology whilst also providing insight into potential liabilities. Firstly, the impact of modified tumor immuno-visibility resulting from modulation of MHC proteins was evaluated with compound 21 in an oncology setting through (i) evaluation of tumor -relevant peptide substrates, such as MART1 40 (ii) comparison of inhibition with knock-out in the CT26 mouse tumor cell line (iii) inhibition of in vivo tumor growth with the syngeneic CT26 model. Secondly, compound 21 was evaluated in an inflammatory autoimmune setting, based on the strong genetic associations between ERAP1 polymorphisms and immune- mediated diseases such as ankylosing spondylitis and psoriasis. These associations imply: (1) ERAP1 may participate in the generation of antigenic peptides recognised by pathogenic T cells in these diseases and (2) inhibition of ERAP1 may interfere with the generation of these pathogenic peptides and provide benefit in disease. However, alteration of the immunopeptidome by ERAP1 inhibition could have the unintended effect of generating new self-antigen epitopes with the potential of generating or exacerbating an autoimmune response. Such a risk has been implied by published studies of ERAAP knock-out.7,41 Mouse collagen-induced arthritis (CIA) was selected as an in vivo model for this purpose due to the presence of both an autoimmune T and B cell response and inflammation. Such a setting would be expected to provide an environment rich in co-stimulatory signals capable of enabling a T cell response that might be triggered by ERAP1-inhibition driven new self-peptides. Standard pathology and scoring endpoints for CIA were supplemented with several measures of immune activation to investigate how ERAP1 inhibition impacted autoimmunity and inflammation. Doses of 30, 90 and 270mg/kg BID were selected based on preliminary mouse PK studies ( Table 5), targeting maximal inhibition and a dynamic concentration range. Herein we describe the PK, estimation of target engagement based on in vitro findings, in vivo pharmacology and impact on pathological endpoints. a) Estimation of unbound drug concentration in vivo. The most comprehensive and representative longitudinal PK data were measured in the CIA study and are presented in Figure 4. PK sampling was undertaken on days 18, 26 and 35 at 1h, 3h and 12h, immediately before the second daily dose, and at 24h, immediately before the first dose on the following day. The unbound concentration of compound 21 was calculated from the total blood concentration and fraction unbound data in Table 6. The concentrations observed were largely dose- proportional, with some variation across the study, for example lower concentrations were noted in some groups on day 26. The ‘peak’ and ‘trough’ concentrations were assessed to be at 1h and 24h and varied from approximately 3-100nM for the 30mg/kg group, 20-700nM for the 90mg/kg group and 200-2000nM for the 270mg/kg group. Figure 4. Unbound concentration of GSK235 based on the CIA study PK on days 18, 26 and 35 following oral BID dosing at (A) 30mg/kg, (B) 90mg/kg and (C) 270mg/kg. (A) (B) (C) preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 10 b) In vitro target engagement and estimation of in vivo target engagement The murine ERAAP potency of compound 21 against YTAFTIPSI and tumor-relevant MART1 EAAGIGILTV40 substrates was measured and is presented in Table 7.42 It is noteworthy that the depth of response and IC 50 are substrate dependent, illustrating the challenge and complexity of assessing the degree of target engagement. Table 7. Mouse ERAAP potencies of compound 21 for distinct peptide substrates. Substrate ERAAP pIC50 Max. asymptote YTAFTIPSI 7.59 (25.7nM) 86 EAAGIGILTV (MART1) 8.21 (6.2nM) 99 The ERAAP potency and maximum asymptote (2 5.7nM, 86%; YTAFTIPSI substrate) were combined with unbound concentrations at 1 and 24 hours from the CIA study to provide preliminary estimations of peak and trough target engagement (Table 8). The predicted ERAAP inhibition varied from 53-73% (peak) declining to 14-31% (trough) at the 30mg/kg dose group escalating to 80 -83% (peak, Cmax) to 43-79% (trough, Cmin) for the 270mg/kg group. It is important to note that the YTAFTIPSI is the most conservative choice of substrate potency and the maximum inhibition achievable in this prediction would be 86%. Table 8. Peak and trough target engagement (TE) estimations derived from CIA PK data and ERAAP potency (YTAFTIPSI substrate) Mouse oral dose Estimated TE 30mg/kg BID 90mg/kg BID 270mg/kg BID Cmax 53-73% 69-80% 80-83% Cmin 14-31% 29-55% 43-79% The impact of compound 21 against a wider variety of immunopeptides was assessed by 30- day treatment of CT26 tumor cells with 1µM compound 21 and compared with CT26 ERAAP knock-out cells (Figure 5). The unbound concentration was estimated to be 610nM in the study based on the fraction unbound measured in cell assay media (Table 6). Substantial evidence of peptide lengthening was observed; for example , significant increases in 10- mer and 11- mer pepti des were accompanied by decreases in 9-mer and 8-mer peptides. Furthermore, the 10-mer and 11-mer peptides were typically extensions of 9-mer peptides (blue circles) indicating a clear effect of ERAAP inhibition or knock -out. The number of peptides lengthened by inhibitor treatment is apparently 80-90% of the total affected by knock-out, indicating a substantial pharmacological effect at an unbound concentration of 610nM, 23 and 100-fold higher than the biochemical inhibition measured using YTAFTIPSI and MART substrates respectively. Figure 6 compares the magnitude of effect for compound 21 (y-axis) versus knock-out (x- axis) and the variance of the correlation line (blue) from the line of unity implies substantial but incomplete inhibition compared with the knock -out condition. Taken together these data provide substantial evidence that compound 21 is likely to induce significant modulation of the immunopeptidome in the 270mg/kg BID dose group given the unbound concentrations of 200-2000nM achieved (see Figure 4). preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 11 Figure 5. Impact on the CT26 tumor cell immunopeptidome in vitro by compound 21 (1µM total, 0.61µM unbound) and ERAAP knock-out compared to wild-type. Figure 6. Comparison of compound 21 treatment and ERAAP knock-out on the immunopeptidome (CT26 cells) preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 12 c) Impact of compound 21 on pathological endpoints in vivo. Tumor growth inhibition was evaluated in mice using the CT26 syngeneic model. Dosing of compound 21 was initiated concurrently with CT26 tumor cell inoculation (day 1) and treatment continued at 30, 90 and 270mg/kg BID with periodic monitoring of tumor growth. A clear dose response (270 > 90 > 30mg/kg) was observed with significant reduction in tumor volume (<50 mm 3) by termination of the study, day 17, at the highest dose ( Figure 7, pane A) and no significant dose- dependent body-weight changes were observed ( Figure 7 , pane B). The findings are not considered directly clinically translatable given the prophylactic nature of the experiment ; however, the result motivated further tumor-bearing studies, the results of which will be reported in due course. Nevertheless, the data provide evidence for biologically meaningful modulation of tumor cell immunogenicity by compound 21. Furthermore, the impact of 610nM compound 21 (unbound) on the immunopeptidome in vitro (vide supra) provides robust mechanistic support given that the unbound concentration in the 270mg/kg group was expected to be 200-2000nM. Figure 7. Tumor growth (A) and body weight change (B) with escalating oral doses of 21. (A) (B) Having established that compound 21 could drive a substantial response in tumor -bearing mice, we sought to explore potential risks by investigating the effects of ERAP1 inhibition in an inflammatory context. The pathology results from the CIA study are presented in Figure 9, with compound 21 eliciting a dose-proportional reduction of arthritis score that was at least as strong as the positive control , the TNF inhibitor E nbrel, used clinically to treat rheumatoid arthritis, in the 270mg/kg group. M inimal changes in body weight were observed compared to vehicle or positive control at the lowest and highest doses respectively (Supplementary Figure 1) and the findings clearly contradicted the hypothesis that MHC-presented peptide modulation could exacerbate an autoimmune response in an inflammatory setting. preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 13 Figure 8. (A) AUC Arthritis Score and (B) Arthritis Score Profile comparing naïve and positive control (Enbrel) compared to escalating dose of compound 21 (A) (B) Multiple measures of paw pathology, including inflammation, pannus formation, cartilage damage, bone resorption and periosteal bone formation ( Figure 9) closely reflected the arthritis score and confirmed a dose-dependent beneficial effect of ERAAP inhibition, again achieving comparable protection to the positive control at the higher dose. Knee pathology also reflected these findings ( Supplementary Figure 2). Figure 9. Paw histopathology with escalating dose of compound 21 d) In vivo pharmacology measurements. A suite of immunological endpoints was measured in the CIA study and indicated that the beneficial effects of compound 21 on arthritis readouts were linked to reductions in autoimmunity and inflammation. This included lower levels of anti- type II collagen antibodies in 21-treated mice (Figure 10, pane A) as well as dose- dependent inhibition of IL -12p40 and IL- 6 (Figure 10, pane B) whereas several other cytokines including KC, MCP-1 and TNF were unaffected by compound 21. Notably, dose- dependent reductions in the cell surface expression of MHC -I protein were observed on B cell and conventional dendritic cell (cDC) populations in lymph nodes and spleen ( Figure 10 , pane C), consistent with modulation of ERAAP function by treatment with compound 21. preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 14 Figure 10. (A) Collagen antibody levels (B) IL-12p40 and IL-6 levels (C) MHC levels in B cells and cDC cells resulting from dose-escalation of compound 21. (A) (B) (C) In addition to assessing CIA- specific pathology, broader t issue histopathology was evaluated to determine whether ERAAP inhibition might lead to non-specific inflammation and to investigate potential additional compound-related effects such as direct toxicity , metabolites or engagement of additional pharmacology. No significant effects attributable to the compound were noted in any of the tissues assessed, including numerous endpoints in the skin, muscle, liver inflammation or cellular infiltrate, gall bladder and small intestine. In summary, the integrated data provides robust correlation of unbound drug levels with pharmacology across an oral dose range of 30mg/kg to 270mg/kg. U nbound drug levels of 200-2000nM were maintained for several weeks at 270mg/k g and clear linkage between proximal pharmacodynamic effects and downstream pharmacology is evident in both systems tested (1) In vitro treatment of CT26 cells with 610nM unbound concentration of compound 21 resulted in significant impact on the immunopeptide repertoire and CT26 tumor-growth regression was observed in vivo at the 270mg/kg dose, expected to achieve or exceed this concentration (2) Dose-dependent down-regulation of MHC proteins translated to significant impact on immunological and pathological end-points in the CIA study across the same dose range, achieving effects comparable to the positive control at 270mg/kg. Undesirable effects were not observed in either study, yet all available data indicates that substantial, chronic inhibition of ERAAP was achieved in both studies. preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 15

Synthetic Procedures Chemical synthesis and characterisation are provided in Supplementary Information Crystallography Methods The ERAP1 (1-941) ∆486-513 GSG insert, N70Q, N154Q, N414Q, N760Q protein was generated as described previously29 with a final protein buffer 50 mM Hepes pH 7, 150 mM NaCl. Crystallisation was carried out using sitting-drop vapour diffusion at 20°C, with 100 nL + 100nL and 100nL + 50nL (protein + well) drops. Co-crystallisations were carried out for each ligand using 100 mM DMSO stock solutions with final ligand concentrations of ~2 mM, using protein at ~8 mg/ml. The crystallisation conditions were initially identified from PACT screen (Qiagen) and optimised in each case, with the conditions used to grow the harvested crystals as follows. Compounds 1 and 2: 25% w/v PEG1500, 0.1M SPG buffer pH=5.5. Compound 8: 25% w/v PEG1500, 0.1M SPG buffer pH=5.0. Compound 21: 20% w/v PEG1500, 0.1M SPG buffer pH=5.5. Crystals were harvested into a 20% ethylene glycol cryoprotectant for a few seconds and then flash cooled in liquid nitrogen. X-ray diffraction data was collected at 100K at Diamond Light Source beamline I03 (Compounds 1 and 2), APS beamline 22-ID (Compound 8) and ESRF beamline ID30B (Compound 21). The datasets were processed and scaled using AUTOPROC (Compounds 1 and 21),43 DIALS (Compound 2), 44 or HKL2000 (Compound 8), 45 utilising XDS, 46 AIMLESS47 and the CCP4 suite of programs. 48 The structures were determined using the coordinates of an isomorphous unliganded model of ERAP1 (unpublished). Preliminary refinement was carried out using BUSTER. 49 The primary ligands were clearly visible in the resulting Fo-Fc electron density maps (Supplementary Figure 3). Model building was carried out with COOT,50 using a ligand dictionary generated from GRADE.51 Final refinement was carried out using BUSTER. Data collection statistics and refinement details for the final models are given in Supplementary Table 2. The coordinates and structure factors have been deposited in the Protein Data Bank under the accession codes XXX, XXX, XXX, and XXX. Biochemical assays Biochemical activity monitoring the cleavage of 9-mer antigenic peptide YTAFTIPSI to 8-mer TAFTIPSI was performed as described previously,31 all data n≥3 unless otherwise specified. An identical method was used to determine ERAAP biochemical activity with YFATIPSI and EAAGIGILTV (MART1) peptide substrates. ERAP2 and IRAP biochemical assays were conducted as previously reported. 31 Cellular antigen presentation assay The cellular antigen presentation assay data described in Tables 1- 4 for compounds 1 -21 was performed as described previously,31 all data n≥3 unless otherwise specified. Adjustments were made to the data normalisation method for data presented for compound 21 in Table 6 , as described previously.38 Physicochemical property data Data was acquired using published protocols for ChromLogD7.4,52 kinetic solubility measured by CAD53 and FaSSIF54 solubility. Fraction Unbound in Blood, Microsomes, and Cellular Assay Media Blood was obtained on the day of experimentation from in house GSK stock animals. Human liver microsomes were obtained from BioIVT (Westbury, NY) and diluted to 0.5mg protein/mL in potassium phosphate buffer. Cellular assay media was provided by GSK Biology. The fraction unbound in blood, human liver microsomes, and cellular assay media (DMEM/F-12 media with 10% FBS, 2 mM Glutamax, and 50 U/mL Pen-Strep) was determined using rapid equilibrium dialysis technology (RED plate (Linden Bioscience, Woburn, MA) at nominal concentrations of 0.5 to 5 µM. Matrices were dialyzed against phosphate buffered saline solution by incubating the dialysis units at 37°C for 4 h with constant shaking preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 16 at 200 rpm. Following incubation aliquots of blood and buffer were matrix matched prior to analysis by LC−MS/MS. The unbound fraction was determined using the peak area ratios in buffer and in blood. Immunopeptidomics i) MHC Class I Peptide Enrichment Immunoprecipitation (IP) of MHC -I complexes was performed from CT26 cell lysates using a mouse monoclonal antibody (US Biological, M3885-73R) specific for H2-Kd/Dd. The protocol was adapted from Bassani Sternberg, 55 with modifications to accommodate experimental conditions. Three biological replicates were processed per condition: ERA AP knock- out, compound 21 treatment, and vehicle control. All steps involving antibody binding and elution were conducted at 4 °C to preserve complex integrity. CT26 cells were lysed according to the referenced protocol, and lysates were applied to affinity columns containing 2 mg of antibody per 400 µL bead matrix. Sequential washes were performed using buffers of increasing ionic strength (150 mM and 400 mM NaCl in 20 mM Tris-HCl, pH 8.0). Bound complexes were eluted with 0.1 M acetic acid. The presence of MHC complexes was confirmed by SDS -PAGE and silver staining. Elution fractions were pooled, dried via SpeedVac, and subjected to C18- based peptide cleanup prior to mass spectrometry analysis. ii) Mass Spectrometry Analysis (MHC-I Peptide Enrichment) Lyophilized peptide samples were resuspended in 0.05% trifluoroacetic acid and analysed using an Ultimate3000 nanoRLSC system (Dionex) coupled to an Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific). Peptides were separated on custom- packed 50 cm × 100 µm ID reversed- phase columns (C18, 1.9 µm, Reprosil -Pur, Dr. Maisch) maintained at 55 °C. Gradient elution was performed from 2% to 40% acetonitrile in 0.1% formic acid and 3.5% DMSO over 120 minutes . 56 Raw data were processed using FragPipe v22.0. MSFragger 57 was configured with the default “Nonspecific-HLA” workflow. Spectra were searched against a SwissProt-based Mus musculus protein database (downloaded January 11, 2018) supplemented with a custom contaminant list. Search parameters included: Precursor and fragment mass tolerance: ±20 ppm, Isotopic error: 0/1, Cleavage specificity: nonspecific, Peptide length: 7 –25 amino acids, Variable modifications (max 3): Oxidation (M), Acetylation (Protein N-term), Pyro-glutamate formation from Q (nQ) and E (nE), Peptide-spectrum matches (PSMs) were re -scored using MSBooster with DIA -NN (v1.8.1) deep learning- based fragment. 58 Validation was performed using Percolator. Quantification was conducted using IonQuant with MaxLFQ -based label-free quantification, including match-between-runs (MBR).59 Ion-level false discovery rate (FDR) was controlled at 1%, followed by intensity-based normalization. iii) Statistical Analysis Statistical analysis was performed in R (http://www.r -project.org) using peptide intensities as a proxy for relative abundance. Variance stabilization normalization (VSN) was applied to derive log2- transformed intensities suitable for differential analysis. Only peptides quantified across all conditions were considered for further analysis. Differential abundance was assessed using the LIMMA modified t-test. 60 P-values were adjusted for multiple testing using the Benjamini -Hochberg method. Peptides were considered significantly regulated if they exhibited an adjusted p- value < 0.05 and a fold- change ≥ 1.5. In Vivo Studies All animal studies were ethically reviewed and carried out in accordance with Animals Scientific Procedures Act 1986 and the GSK Policy on the Care, Welfare and Treatment of Animals. i) M ouse oral PK Female BALB/c mice were orally dosed with compound at a target dose of 10mg/kg, formulated as suspensions in 1% aq. methylcellulose (w/v) at 1 mg/mL. Serial blood samples were collected up to 24 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 17 h post oral administration. Following collection, aliquots of blood were diluted with an equal volume of water, mixed, prior to freezing on cardice and storage at - 80°C until bioanalysis. Aliquots of diluted whole blood (50:50) were analysed using quantitative high -performance liquid chromatography with tandem mass spectrometric detection (LC-MS/MS) following protein precipitation. PK parameters were estimated from the blood concentration–time profiles using noncompartmental analysis with WinNonlin (Pharsight, Mountain View, CA). ii) Mouse CT26 Tumor Model Female BALB/c mice (6-8 weeks old) were acquired from Envigo and housed under standard laboratory conditions with ad libitum access to food and water. On study day one, animals were randomized by body weight into study groups and inoculated subcutaneously in the right hind flank with 5x10^4 CT26 cells. Concurrently, treatment administration began with either compound 21 ( GSK235) or vehicle control and dosing was performed twice daily (BID) for the duration of the study. iii) Mouse Collagen-Induced Arthritis (CIA) Model A total of 70 male DBA/1OlaHsd mice (6– 7 weeks old) were obtained from Envigo. Animals were housed under standard conditions with food and water provided ad libitum. On Day 0 (D0), mice received an intradermal (i.d.) injection of 0.2 mg bovine collagen type II emulsified in 100 µL Freund’s Complete Adjuvant (FCA). From Day 18 to Day 36, mice were administered either vehicle (1% methyl cellulose), compound 21 (GSK235) at doses of 30, 90, or 270mg/kg, or the positive control Enbrel at 10mg/kg. Vehicle and compound 21 (GSK235) were administered twice daily via oral gavage, while Enbrel was administered intraperitoneally every other day. On Day 21, a booster i.d. injection of 0.2 mg bovine collagen type II in 100 µL FCA was given. Clinical arthritis scores were assessed in a blinded manner for all four paws (Supplementary Table 3). PK blood samples were collected on Days 18, 19, 27, 35, and 36. On Day 36, animals were euthanised, and serum was collected for quantification of anti-collagen type II IgG and pro-inflammatory cytokines. Tissues including joints, liver, small intestine, and skin were harvested for histological analysis. Spleen and lymph nodes (popliteal and inguinal) were collected for flow cytometry. Anti-Collagen Type II IgG Quantification Serum samples were analysed in duplicate using a Mouse Anti -Collagen Type II IgG ELISA kit (Chondrex, Inc., Cat #2036T), following the manufacturer’s instructions. Multiplex Cytokine Assays Levels of IL- 12p40 and IL- 6 were quantified using the U -PLEX multiplex platform (Meso Scale Discovery), according to the manufacturer’s protocol. Flow Cytometry Single cell suspensions from spleen and pooled lymph nodes were stained using antibody panels detailed in Supplementary Table 4. Samples were processed with the True- Nuclear™ Transcription Factor Buffer Set (BioLegend, Cat #424401) and analysed on a CytoFLEX Flow Cytometry Analyzer (Beckman Coulter, Inc.). preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 18

Accumulating evidence support s the tractability of inhibiting ERAP1 for regulating human adaptive immune responses. Although ERAP1 has been reported to participate in several biological functions, the most established one is a highly specialized role in regulating which antigenic peptides can be presented by MHC -I molecules. 61 A secondary and possibly related role in specific cellular and inflammatory context includes an autoinflammatory function when secreted and operates in the interface of innate and adaptive immunity. 10 Both functions have been shown to be regulated by small molecule inhibitors and may synergize in regulating immune responses. 34,62,63 As a result, down- regulation of ERAP1 activity may induce synergistic pleiotropic effects on T cell and NK cell responses as well as inflammatory signals. Such effects show promise in reprogramming the immune system to fight tumors since the upregulation of ERAP1 may act as an immune evasion mechanism. In addition, since antigen presentation is central to tumor responses, ERAP1 mediated immune effects may synergize with other immune modulating approaches in cancer immunotherapy efforts including immune checkpoint blockade, cancer vaccines, adoptive cell therapies, oncolytic viruses, epigenetic modulators, proteasome inhibitors, radiotherapy, NK cell therapies, Stimulator of interferon genes (STING) agonists, and cytokine therapies. 12,15,64-68 Despite significant effort towards the development of ERAP1 inhibitors, suitable in vivo tools have been lacking and reported inhibitors suffer from limitations such as low potency, limited selectivity or low cellular potency.24,25 To address these issues, we pursued a library hit that targets the unique regulatory site of the enzyme, which was optimised to an in vivo tool ERAP1 inhibitor, compound 21 (GSK235). This compound has excellent potency and selectivity both in vitro and in cells, with favourable oral PK and target engagement profile. Furthermore, it induces relevant in vivo effects in two murine models suggesting that it is suitable as a tool for further in vivo investigations. The mechanism of action of compound 21 focuses on the regulatory site of ERAP1 that normally accommodates the C-termini of peptidic substrates and self-activates the enzyme so that it can preferentially cleave elongated peptides that can concurrently occupy both the active and regulatory site. 33 Thus, although this compound acts as a non-competitive activator of small substrates, it is a competitive inhibitor of longer peptides.31 Small molecules targeting this site have been proposed to induce a large conformational change in ERAP1, part of its normal catalytic cycle, that closes of external access to the active site and blocks further catalytic site. 69 Thus, although the allosteric site is mechanistically distinct, the effect on binding of small molecules can be functionally equivalent to active site blocking. This is supported by both cell -based assays for a single antigen and immunopeptidomics analysis that demonstrates functional equivalence of the reported compound to genetic knock -out. Thus, it appears that GSK235 can combine high selectivity by targeting a non- conserved site while retaining the pan- substrate effect of an active site inhibitor. This comes as a contrast to a previously described natural-product that targets the same site, which presented distinct effects on the cellular immunopeptidome. 35 However, the potency of that compound was approximately 3 orders of magnitude weaker and intra-molecular competition between the C-termini of substrate peptides and the weak inhibition may underlie such effects. Testing the efficacy of GSK235 in a murine cancer model demonstrated effective and dose- dependent retardation of tumor growth, leading to tumor control at the higher dosages. The pharmacodynamic effect corresponds well to immunopeptidome shifts identified in cellulo , consistent with the main therapeutic hypothesis that ERAP1 inhibition leads to immunopeptidome shifts that induce more potent or novel cellular adaptive immune responses. Although we did not proceed to detailed immunological analysis using this tool, observed effects parallel the described efficacy of ERAP1 genetic knockdown in other murine models that underlie T cell and NK cell mechanisms.11-13,15,17 However, the interpretation of immune effects after genetic downregulation of ERAP1 may have important limitations for translating to therapeutic interventions due to the dynamic and spatial nature of immune responses. Specifically, murine knock -out models may adapt to lack of ERAP1 activity pre- tolerance and genetic down- regulation in implanted cell lines simulates perfect cellular targeting that is not easily achievable clinically. Thus, GSK235 constitutes a valuable tool for interrogating the in vivo role of ERAP1 in different cancer immunotherapy settings and more importantly in combinatorial treatments. Combinatorial cancer immunotherapy is actively being pursued due to its promise in avoiding cancer evasion mechanisms. 70 Although GSK235 was found to be surprisingly effective as a monotherapy in preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 19 our murine model, future investigations will require investigation into the interplay of ERAP1 inhibition with other immune system interventions in cancer. The significant effects of ERAP1 inhibition on the cellular immunopeptidome have raised concerns for potential autoimmune effects due to the presentation of self -antigens that are normally repressed. 71 Self-mimicry due to presentation of self-antigenic peptides has been proposed to be a major mechanism behind MHC-I-opathies a class of inflammatory diseases of autoimmune aetiology that correlate with specific HLA alleles and ERAP1 allotypes, often in epi stasis.19,72-74 In addition, potential effects of ERAP1 on ER homeostasis, stress and the unfolded protein responses have been proposed as separate or complementary mechanisms. 75,76 These mechanisms generate concerns of immune toxicity of ERAP1 inhibitors that could complicate therapeutic approaches. On the other hand, enhanced ERAP1 activity has been correlated mechanistically with predisposition to these diseases, suggesting that ERAP1 inhibition may hold therapeutic promise. 77 To gain further insight into these concerns, we evaluated the in vivo effects of GSK235 in the mouse collagen induced arthritis model. ERAP1 inhibition did not exacerbate disease but had a strong, dose- dependent, therapeutic effect that equalled or surpassed the effect of Enbrel , a TNF inhibitor used clinically to treat rheumatoid arthritis. These effects closely correlated with reductions in anti-collagen IgG and B cell MHC-I levels as well as with IL-12p40 and IL-6 production. These results add to previous reports indicating that ERAP1 knockdown or inhibition can dampen T cell and NK cell activation and cytokine signalling, while also having effects on dendritic cell and macrophage function, limiting Th1 and Th17 polarization.63,78 The effects observed here on autoantibody production could be mediated by down-regulated generation of unknown self -antigenic epitopes and subsequent reduced antigen presentation to T helper cells, either during initial T cell priming on dendritic cells or in the context of T cell – B cell help . Since the T cell – antigen-presenting cell interactions driving antibody responses typically involve MHC class II peptide presentation, such a mechanism would need to invoke a (direct or indirect) role for ERAP1 in the MHC II pathway; a secondary role in shaping the peptide load of MHC class II molecules has been previously proposed.79 Conversely, reduced T and B cell responses may be secondary to a reduced overall inflammatory environment mediated by ERAP1 inhibition. Thus , the effects of ERAP1 function on innate immune responses may also play a role and could include inhibition of secreted ERAP1 by B cells or macrophages or the effect of ERAP1 in cellular homeostasis leading to reduction of innate immunity responses.10,75,80 The molecular details of these complex and potentially synergistic effects will have to be thoroughly analysed in future studies and GSK235 will undoubtably be a valuable tool since it can perturb ERAP1 function dynamically , circumventing limitations of pre- tolerance genetic manipulations. Still, our results suggest that ERAP1 inhibition may have a significant therapeutic potential not only for HLA-associated autoimmunity but also in other forms of inflammatory autoimmunity that should be explored. 81

In summary, we report the discovery of the first potent in vivo tool for interrogating ERAP1 biology that also constitutes a potential lead molecule for drug- discovery efforts. Hit compound 1 was identified via high-throughput screening and early generation of X-ray crystallography allowed efficient optimisation to lead compound 21 (GSK235) through introduction of productive interactions. Compound 21 was progressed firstly to a pilot prophylactic study in tumor-bearing mice, confirming dose-dependent tumor growth inhibition, then to a comprehensive CIA model to assess risks implied by literature studies with knock-out mice. Administration of compound 21 across a substantial dose range resulted in significant protection in this model with integration of in vitro and in vivo data strongly implying 50- 80% inhibition of ERAAP through the study. The CIA model findings do not directly support a protective role for ERAP1 inhibition in human disease; however, the data substantially allay concerns of disease exacerbation resulting from ERAP1 inhibition on an inflammatory background and, encouragingly, no substantial pathological consequences were evident in the study. Our results help establish the tractability of ERAP1 inhibition for cancer immunotherapy and expand options to certain forms of inflammatory autoimmune disease, not limited to MHC -I-opathies. The public availability of GSK235 will empower mechanistic studies that will help progress the therapeutic value of ERAP1 inhibition and facilitate the exploration of combinatorial strategies aiming to reprogram the immune system. preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 20

The authors thank Heather Barnett, Stephen Besley, Shenaz Bunally, Esme Clarke, Yanan He, Justyna Korczynska, Stefan Maehringer, Ferdausi Mazumder, Richard Upton and Bob Watson. Additional Information Supplementary Information contains supplementary figures & tables, synthetic procedures and X- ray crystallography methods Corresponding Authors Correspondence to Simon Peace [email protected] and Efstratios Stratikos [email protected] Competing interests All authors except D. Koumantou, I. Temponeras and E. Stratikos were employees at GSK at time of this work. preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 21

1 York, I. A. & Rock, K. L. Antigen processing and presentation by the class I major histocompatibility complex. Annu. Rev. Immunol. 14, 369-396 (1996). https://doi.org/10.1146/annurev.immunol.14.1.369 2 Pishesha, N., Harmand, T. J. & Ploegh, H. L. A guide to antigen processing and presentation. Nat. Rev. Immunol. 22, 751-764 (2022). https://doi.org/10.1038/s41577-022-00707-2 3 Rock, K. L. & Goldberg, A. L. Degradation of cell proteins and the generation of MHC class I - presented peptides. Annu. Rev. Immunol. 17, 739-779 (1999). https://doi.org/10.1146/annurev.immunol.17.1.739 4 Lauvau, G. et al. Human transporters associated with antigen processing (TAPs) select epitope precursor peptides for processing in the endoplasmic reticulum and presentation to T cells. J. Exp. Med. 190, 1227-1240 (1999). https://doi.org/10.1084/jem.190.9.1227 5 Falk, K. & Rötzschke, O. The final cut: how ERAP1 trims MHC ligands to size. Nat. Immunol. 3, 1121-1122 (2002). https://doi.org/10.1038/ni1202-1121 6 York, I. A. et al. The ER aminopeptidase ERAP1 enhances or limits antigen presentation by trimming epitopes to 8-9 residues. Nat. Immunol. 3, 1177-1184 (2002). https://doi.org/10.1038/ni860 7 Hammer, G. E., Gonzalez, F., James, E., Nolla, H. & Shastri, N. In the absence of aminopeptidase ERAAP, MHC class I molecules present many unstable and highly immunogenic peptides. Nat. Immunol. 8, 101-108 (2007). https://doi.org/10.1038/ni1409 8 Firat, E. et al. The role of endoplasmic reticulum-associated aminopeptidase 1 in immunity to infection and in cross-presentation. J. Immunol. 178, 2241-2248 (2007). https://doi.org/10.4049/jimmunol.178.4.2241 9 Goto, Y., Ogawa, K., Hattori, A. & Tsujimoto, M. Secretion of endoplasmic reticulum aminopeptidase 1 is involved in the activation of macrophages induced by lipopolysaccharide and interferon-gamma. J. Biol. Chem. 286, 21906-21914 (2011). https://doi.org/10.1074/jbc.M111.239111 10 Blake, M. K. et al. ERAP1 is a critical regulator of inflammasome-mediated proinflammatory and ER stress responses. BMC Immunol. 23, 9 (2022). https://doi.org/10.1186/s12865-022- 00481-9 11 Cifaldi, L. et al. ERAP1 regulates natural killer cell function by controlling the engagement of inhibitory receptors. Cancer Res. 75, 824-834 (2015). https://doi.org/10.1158/0008-5472.Can- 14-1643 12 Tempora, P. et al. Combining ERAP1 silencing and entinostat therapy to overcome resistance to cancer immunotherapy in neuroblastoma. J Exp Clin Cancer Res 43, 292 (2024). https://doi.org/10.1186/s13046-024-03180-y 13 James, E., Bailey, I., Sugiyarto, G. & Elliott, T. Induction of protective antitumor immunity through attenuation of ERAAP function. J. Immunol. 190, 5839-5846 (2013). https://doi.org/10.4049/jimmunol.1300220 14 Nagarajan, N. A. et al. ERAAP Shapes the Peptidome Associated with Classical and Nonclassical MHC Class I Molecules. J. Immunol. 197, 1035-1043 (2016). https://doi.org/10.4049/jimmunol.1500654 15 Tsao, H. W. et al. Targeting the aminopeptidase ERAP enhances antitumor immunity by disrupting the NKG2A-HLA-E inhibitory checkpoint. Immunity 57, 2863-2878.e2812 (2024). https://doi.org/10.1016/j.immuni.2024.10.013 16 Lim, Y. W. et al. Germline genetic polymorphisms influence tumor gene expression and immune cell infiltration. Proc Natl Acad Sci U S A 115, E11701-e11710 (2018). https://doi.org/10.1073/pnas.1804506115 17 Reeves, E. et al. HPV Epitope Processing Differences Correlate with ERAP1 Allotype and Extent of CD8(+) T-cell Tumor Infiltration in OPSCC. Cancer Immunol Res 7, 1202-1213 (2019). https://doi.org/10.1158/2326-6066.Cir-18-0498 18 Robinson, J., Barker, D. J. & Marsh, S. G. E. 25 years of the IPD-IMGT/HLA Database. HLA 103, e15549 (2024). https://doi.org/10.1111/tan.15549 19 Kuiper, J. J. et al. EULAR study group on 'MHC-I-opathy': identifying disease-overarching mechanisms across disciplines and borders. Ann. Rheum. Dis. 82, 887-896 (2023). https://doi.org/10.1136/ard-2022-222852 20 Cortes, A. et al. Major histocompatibility complex associations of ankylosing spondylitis are complex and involve further epistasis with ERAP1. Nat Commun 6, 7146 (2015). https://doi.org/10.1038/ncomms8146 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 22 21 Evnouchidou, I. et al. Cutting Edge: Coding single nucleotide polymorphisms of endoplasmic reticulum aminopeptidase 1 can affect antigenic peptide generation in vitro by influencing basic enzymatic properties of the enzyme. J. Immunol. 186, 1909-1913 (2011). https://doi.org/10.4049/jimmunol.1003337 22 Hutchinson, J. P. et al. Common allotypes of ER aminopeptidase 1 have substrate-dependent and highly variable enzymatic properties. J. Biol. Chem. 296, 100443 (2021). https://doi.org/10.1016/j.jbc.2021.100443 23 Stamatakis, G., Samiotaki, M., Temponeras, I., Panayotou, G. & Stratikos, E. Allotypic variation in antigen processing controls antigenic peptide generation from SARS -CoV-2 S1 spike glycoprotein. J. Biol. Chem. 297, 101329 (2021). https://doi.org/10.1016/j.jbc.2021.101329 24 Georgiadis, D., Mpakali, A., Koumantou, D. & Stratikos, E. Inhibitors of ER Aminopeptidase 1 and 2: From Design to Clinical Application. Curr. Med. Chem. 26, 2715-2729 (2019). https://doi.org/10.2174/0929867325666180214111849 25 Fougiaxis, V. et al. ERAP Inhibitors in Autoimmunity and Immuno-Oncology: Medicinal Chemistry Insights. J. Med. Chem. 67, 11597-11621 (2024). https://doi.org/10.1021/acs.jmedchem.4c00840 26 Kokkala, P. et al. Optimization and Structure-Activity Relationships of Phosphinic Pseudotripeptide Inhibitors of Aminopeptidases That Generate Antigenic Peptides. J. Med. Chem. 59, 9107-9123 (2016). https://doi.org/10.1021/acs.jmedchem.6b01031 27 Papakyriakou, A. et al. 3,4-diaminobenzoic acid derivatives as inhibitors of the oxytocinase subfamily of M1 aminopeptidases with immune-regulating properties. J. Med. Chem. 58, 1524-1543 (2015). https://doi.org/10.1021/jm501867s 28 Vourloumis, D. et al. Discovery of Selective Nanomolar Inhibitors for Insulin-Regulated Aminopeptidase Based on α-Hydroxy-β-amino Acid Derivatives of Bestatin. J. Med. Chem. 65, 10098-10117 (2022). https://doi.org/10.1021/acs.jmedchem.2c00904 29 Giastas, P., Neu, M., Rowland, P. & Stratikos, E. High-Resolution Crystal Structure of Endoplasmic Reticulum Aminopeptidase 1 with Bound Phosphinic Transition- State Analogue Inhibitor. ACS Med Chem Lett 10, 708-713 (2019). https://doi.org/10.1021/acsmedchemlett.9b00002 30 Maben, Z. et al. Discovery of Selective Inhibitors of Endoplasmic Reticulum Aminopeptidase 1. J. Med. Chem. 63, 103-121 (2020). https://doi.org/10.1021/acs.jmedchem.9b00293 31 Liddle, J. et al. Targeting the Regulatory Site of ER Aminopeptidase 1 Leads to the Discovery of a Natural Product Modulator of Antigen Presentation. J. Med. Chem. 63, 3348-3358 (2020). https://doi.org/10.1021/acs.jmedchem.9b02123 32 Deddouche-Grass, S. et al. Discovery and Optimization of a Series of Benzofuran Selective ERAP1 Inhibitors: Biochemical and In Silico Studies. ACS Med Chem Lett 12, 1137-1142 (2021). https://doi.org/10.1021/acsmedchemlett.1c00235 33 Giastas, P. et al. Mechanism for antigenic peptide selection by endoplasmic reticulum aminopeptidase 1. Proc Natl Acad Sci U S A 116, 26709-26716 (2019). https://doi.org/10.1073/pnas.1912070116 34 Koumantou, D. et al. Editing the immunopeptidome of melanoma cells using a potent inhibitor of endoplasmic reticulum aminopeptidase 1 (ERAP1). Cancer Immunol Immunother 68, 1245- 1261 (2019). https://doi.org/10.1007/s00262-019-02358-0 35 Temponeras, I. et al. Distinct modulation of cellular immunopeptidome by the allosteric regulatory site of ER aminopeptidase 1. Eur. J. Immunol. 53, e2350449 (2023). https://doi.org/10.1002/eji.202350449 36 Nikopaschou, M. et al. ERAP1 Activity Modulates the Immunopeptidome but Also Affects the Proteome, Metabolism, and Stress Responses in Cancer Cells. Mol. Cell. Proteomics 24, 100964 (2025). https://doi.org/10.1016/j.mcpro.2025.100964 37 Lillie, T. et al. EMITT-1: Proof-of-mechanism immunopeptidome (ImPD) effects at target PK exposure, in a phase 1 study of GRWD5769 (a first-in-class inhibitor of Endoplasmic Reticulum Aminopeptidase 1 [ERAP1]) in patients with solid malignancies. Journal of Clinical Oncology 42, 2589-2589 (2024). https://doi.org/10.1200/JCO.2024.42.16_suppl.2589 38 Hryczanek, R. P. et al. Optimization of Potent and Selective Cyclohexyl Acid ERAP1 Inhibitors Using Structure- and Property-Based Drug Design. ACS Medicinal Chemistry Letters (2024). https://doi.org/10.1021/acsmedchemlett.4c00401 39 Baker, J. A., Altman, M. D. & Martin, I. J. Interpretation of in Vitro Metabolic Stability Studies for Racemic Mixtures. ACS Medicinal Chemistry Letters 9, 843-847 (2018). https://doi.org/10.1021/acsmedchemlett.8b00259 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 23 40 Pittet, M. J. et al. Ex vivo characterization of allo-MHC-restricted T cells specific for a single MHC-peptide complex. J. Immunol. 176, 2330-2336 (2006). https://doi.org/10.4049/jimmunol.176.4.2330 41 Pepelyayeva, Y. et al. ERAP1 deficient mice have reduced Type 1 regulatory T cells and develop skeletal and intestinal features of Ankylosing Spondylitis. Sci Rep 8, 12464 (2018). https://doi.org/10.1038/s41598-018-30159-5 42 Keller, M. et al. The proteasome immunosubunits, PA28 and ER-aminopeptidase 1 protect melanoma cells from efficient MART-126-35 -specific T-cell recognition. Eur. J. Immunol. 45, 3257-3268 (2015). https://doi.org/10.1002/eji.201445243 43 Vonrhein, C. et al. Data processing and analysis with the autoPROC toolbox. Acta Crystallogr D Biol Crystallogr 67, 293-302 (2011). https://doi.org/10.1107/s0907444911007773 44 Winter, G. et al. DIALS: implementation and evaluation of a new integration package. Acta Crystallogr D Struct Biol 74, 85-97 (2018). https://doi.org/10.1107/s2059798317017235 45 Otwinowski, Z. & Minor, W. Processing of X-ray diffraction data collected in oscillation mode.

Enzymol. 276, 307-326 (1997). https://doi.org/10.1016/s0076-6879(97)76066-x 46 Kabsch, W. XDS. Acta Crystallogr D Biol Crystallogr 66, 125-132 (2010). https://doi.org/10.1107/s0907444909047337 47 Evans, P. R. & Murshudov, G. N. How good are my data and what is the resolution? Acta Crystallogr D Biol Crystallogr 69, 1204-1214 (2013). https://doi.org/10.1107/s0907444913000061 48 Winn, M. D. et al. Overview of the CCP4 suite and current developments. Acta Crystallogr D Biol Crystallogr 67, 235-242 (2011). https://doi.org/10.1107/s0907444910045749 49 Bricogne, G. et al. BUSTER version 2.11.7. Global Phasing Ltd., Cambridge, UK. (2017). 50 Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr D Biol Crystallogr 66, 486-501 (2010). https://doi.org/10.1107/s0907444910007493 51 Smart, O. S. et al. Grade version 1.2.20. Global Phasing Ltd., Cambridge, UK. (2011). 52 Camurri, G. & Zaramella, A. High-Throughput Liquid Chromatography/Mass Spectrometry

for the Determination of the Chromatographic Hydrophobicity Index. Anal. Chem. 73, 3716-3722 (2001). https://doi.org/10.1021/ac001388j 53 Robinson, M. W. et al. Use of Calculated Physicochemical Properties to Enhance Quantitative Response When Using Charged Aerosol Detection. Anal. Chem. 89, 1772-1777 (2017). https://doi.org/10.1021/acs.analchem.6b04060 54 Sou, T. & Bergström, C. A. S. Automated assays for thermodynamic (equilibrium) solubility determination. Drug Discov. Today Technol. 27, 11-19 (2018). https://doi.org/https://doi.org/10.1016/j.ddtec.2018.04.004 55 Bassani-Sternberg, M. Mass Spectrometry Based Immunopeptidomics for the Discovery of Cancer Neoantigens. Methods Mol Biol 1719, 209-221 (2018). https://doi.org/10.1007/978-1- 4939-7537-2_14 56 Zinn, N. et al. Improved Proteomics-Based Drug Mechanism-of-Action Studies Using 16-Plex Isobaric Mass Tags. J Proteome Res 20, 1792-1801 (2021). https://doi.org/10.1021/acs.jproteome.0c00900 57 Kong, A. T., Leprevost, F. V., Avtonomov, D. M., Mellacheruvu, D. & Nesvizhskii, A. I. MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry -based proteomics. Nat. Methods 14, 513-520 (2017). https://doi.org/10.1038/nmeth.4256 58 Yang, K. L. et al. MSBooster: improving peptide identification rates using deep learning-based features. Nat Commun 14, 4539 (2023). https://doi.org/10.1038/s41467-023-40129-9 59 Yu, F., Haynes, S. E. & Nesvizhskii, A. I. IonQuant Enables Accurate and Sensitive Label - Free Quantification With FDR-Controlled Match-Between-Runs. Mol. Cell. Proteomics 20, 100077 (2021). https://doi.org/10.1016/j.mcpro.2021.100077 60 Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47 (2015). https://doi.org/10.1093/nar/gkv007 61 López de Castro, J. A. How ERAP1 and ERAP2 Shape the Peptidomes of Disease - Associated MHC-I Proteins. Front Immunol 9, 2463 (2018). https://doi.org/10.3389/fimmu.2018.02463 62 Aldhamen, Y. A. et al. Autoimmune disease-associated variants of extracellular endoplasmic reticulum aminopeptidase 1 induce altered innate immune responses by human immune cells. J Innate Immun 7, 275-289 (2015). https://doi.org/10.1159/000368899 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint 24 63 Chen, L. et al. Silencing or inhibition of endoplasmic reticulum aminopeptidase 1 (ERAP1) suppresses free heavy chain expression and Th17 responses in ankylosing spondylitis. Ann. Rheum. Dis. 75, 916-923 (2016). https://doi.org/10.1136/annrheumdis-2014-206996 64 D'Amico, S. et al. ERAP1 Controls the Interaction of the Inhibitory Receptor KIR3DL1 With HLA-B51:01 by Affecting Natural Killer Cell Function. Front Immunol 12, 778103 (2021). https://doi.org/10.3389/fimmu.2021.778103 65 Mpakali, A. & Stratikos, E. The Role of Antigen Processing and Presentation in Cancer and the Efficacy of Immune Checkpoint Inhibitor Immunotherapy. Cancers (Basel) 13 (2021). https://doi.org/10.3390/cancers13010134 66 Schmidt, K. et al. ERAP1-Dependent Antigen Cross-Presentation Determines Efficacy of Adoptive T-cell Therapy in Mice. Cancer Res. 78, 3243-3254 (2018). https://doi.org/10.1158/0008-5472.Can-17-1946 67 Reits, E. A. et al. Radiation modulates the peptide repertoire, enhances MHC class I expression, and induces successful antitumor immunotherapy. J. Exp. Med. 203, 1259-1271 (2006). https://doi.org/10.1084/jem.20052494 68 Martin, A. L. et al. Anti-4-1BB immunotherapy enhances systemic immune effects of radiotherapy to induce B and T cell-dependent anti-tumor immune activation and improve tumor control at unirradiated sites. Cancer Immunol Immunother 72, 1445-1460 (2023). https://doi.org/10.1007/s00262-022-03325-y 69 Maben, Z., Arya, R., Georgiadis, D., Stratikos, E. & Stern, L. J. Conformational dynamics linked to domain closure and substrate binding explain the ERAP1 allosteric regulation mechanism. Nat Commun 12, 5302 (2021). https://doi.org/10.1038/s41467-021-25564-w 70 Wei, Y. et al. Efficacy and safety of combination immunotherapy for malignant solid tumors: A systematic review and meta-analysis. Crit Rev Oncol Hematol 138, 178-189 (2019). https://doi.org/10.1016/j.critrevonc.2019.04.008 71 Babaie, F. et al. The roles of ERAP1 and ERAP2 in autoimmunity and cancer immunity: New insights and perspective. Mol. Immunol. 121, 7-19 (2020). https://doi.org/10.1016/j.molimm.2020.02.020 72 Britanova, O. V. et al. Targeted depletion of TRBV9(+) T cells as immunotherapy in a patient with ankylosing spondylitis. Nat. Med. 29, 2731-2736 (2023). https://doi.org/10.1038/s41591- 023-02613-z 73 Arakawa, A. et al. ERAP1 Controls the Autoimmune Response against Melanocytes in Psoriasis by Generating the Melanocyte Autoantigen and Regulating Its Amount for HLA - C*06:02 Presentation. J. Immunol. 207, 2235-2244 (2021). https://doi.org/10.4049/jimmunol.2100686 74 Yang, X. et al. Autoimmunity-associated T cell receptors recognize HLA-B*27-bound peptides. Nature 612, 771-777 (2022). https://doi.org/10.1038/s41586-022-05501-7 75 Tran, T. M. et al. Paradoxical Effects of Endoplasmic Reticulum Aminopeptidase 1 Deficiency on HLA-B27 and Its Role as an Epistatic Modifier in Experimental Spondyloarthritis. Arthritis Rheumatol 75, 220-231 (2023). https://doi.org/10.1002/art.42327 76 Cauli, A. et al. The arthritis-associated HLA-B*27:05 allele forms more cell surface B27 dimer and free heavy chain ligands for KIR3DL2 than HLA-B*27:09. Rheumatology (Oxford) 52, 1952-1962 (2013). https://doi.org/10.1093/rheumatology/ket219 77 Reeves, E., Islam, Y. & James, E. ERAP1: a potential therapeutic target for a myriad of diseases. Expert Opin Ther Targets 24, 535-544 (2020). https://doi.org/10.1080/14728222.2020.1751821 78 Babaie, F. et al. Inhibition of ERAP1 represses HLA-B27 free heavy chains expression on polarized macrophages and interrupts NK cells activation and function from ankylosing spondylitis. Clin. Immunol. 248, 109268 (2023). https://doi.org/10.1016/j.clim.2023.109268 79 Dragovic, S. M. et al. Proteasomes, TAP, and endoplasmic reticulum-associated aminopeptidase associated with antigen processing control CD4+ Th cell responses by regulating indirect presentation of MHC class II-restricted cytoplasmic antigens. J. Immunol. 186, 6683-6692 (2011). https://doi.org/10.4049/jimmunol.1100525 80 Goto, Y., Ogawa, K., Nakamura, T. J., Hattori, A. & Tsujimoto, M. TLR-mediated secretion of endoplasmic reticulum aminopeptidase 1 from macrophages. J. Immunol. 192, 4443-4452 (2014). https://doi.org/10.4049/jimmunol.1300935 81 Pepelyayeva, Y. & Amalfitano, A. The role of ERAP1 in autoinflammation and autoimmunity. Hum. Immunol. 80, 302-309 (2019). https://doi.org/10.1016/j.humimm.2019.02.013 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted November 17, 2025. ; https://doi.org/10.1101/2025.11.17.686761doi: bioRxiv preprint