ERAP1 binds peptide C-termini of different sequences and/or lengths by a common recognition mechanism

Abstract

Endoplasmic reticulum aminopeptidase 1 (ERAP1) plays a key role in controlling the immunopeptidomes available for presentation by MHC (major histocompatibility complex) molecules, thus influences immunodominance and cell-mediated immunity. It carries out this critical function by a unique molecular ruler mechanism that trims antigenic precursors in a peptide-length and sequence dependent manner. Acting as a molecular ruler, ERAP1 is capable of concurrently binding antigen peptide N- and C-termini by its N-terminal catalytic and C-terminal regulatory domains, respectively. As such ERAP1 can not only monitor substrate’s lengths, but also exhibit a degree of sequence specificity at substrates’ N- and C-termini. On the other hand, it also allows certain sequence and length flexibility in the middle part of peptide substrates that is critical for shaping MHC restricted immunopeptidomes. Here we report structural and biochemical studies to understand the molecular details on how ERAP1 can accommodate side chains of different anchoring residues at the substrate’s C-terminus. We also examine how ERAP1 can accommodate antigen peptide precursors with length flexibility. Based on two newly determined complex structures, we find that ERAP1 binds the C-termini of peptides similarly even with different substrate sequences and/or lengths, by utilizing the same hydrophobic specificity pocket to accommodate peptides with either a Phe or Leu as the C-terminal anchor residue. In addition, SPR (surface plasmon resonance) binding analyses in solution further confirm the biological significance of these peptide-ERAP1 interactions. Similar to the binding mode of MHC-I molecules, ERAP1 accommodates for antigenic peptide length difference by allowing the peptide middle part to kink or bulge at the middle of its substrate binding cleft. This explains how SNP coded variants located at the middle of ERAP1 substrate binding cleft would influence the antigen pool and an individual’s susceptibility to diseases.

Introduction

Endoplasmic reticulum aminopeptidase 1 (ERAP1) plays important roles in various biological functions, including blood pressure regulation, phagocytosis in the innate immunity, and major histocompatibility complex (MHC)-restricted adaptive immunity (Babaie et al., 2020; Fruci et al., 2014). ERAP1 is highly polymorphic and its natural variants, single nucleotide polymorphisms (SNPs), are associated with various human diseases, mostly autoimmune diseases (Alvarez-Navarro and Lopez de Castro, 2014; Fruci et al., 2014; Kenna et al., 2015; Pepelyayeva and Amalfitano, 2019; York et al., 2006). These diseases are linked with human leukocyte antigen (HLA) class I alleles, indicating an association with MHC-I (MHC class I molecules) antigen processing and presentation pathway in disease pathogenesis (Cortes et al., 2015; Lopez de Castro et al., 2016). A well-known case is the association of ERAP1 SNPs with ankylosing spondylitis (AS), a chronic inflammatory rheumatic arthritis. Studies have indicated an epistatic interaction between HLA-B27 and ERAP1 SNPs associated with AS (Vitulano et al., 2017). In addition, Behcet disease (BD) is also identified to be associated with ERAP1 SNPs on HLA-B51 allele (Guasp et al., 2019). Consistent with this epistatic connection, SNP variants of ERAP1 are found to affect the enzyme by diminishing its trimming efficiency or altering its peptide substrate preference in antigen processing (Vitulano et al., 2017). Several SNP variants, such as K528R and Q730E mutations, are located at the ERAP1 regulatory domain (ERAP1_R) that is separated from its catalytic N-terminal domain. This implies that these variants do not directly affect ERAP1’s catalysis, but influence antigen processing indirectly to alter the immunopeptidomes for MHC-I presentation. Studies also suggested that ERAP1 SNP variants can influence peptide trimming by affecting peptide-binding sites in a peptide-dependent manner (Stratikos et al., 2014). Therefore, understanding the peptide binding mechanism of ERAP1 at atomic resolution is critical to understand the underlying molecular mechanism of SNP variants in altering ERAP1’s function and its pathogenic role (Hanson et al., 2019; Reeves et al., 2020).

In cell-mediated immune responses, ERAP1 plays key roles in the N-terminal processing of antigenic peptides to be presented by the MHC-I molecules (Fruci et al., 2014). ERAP1 prefers peptide substrates with C-terminal hydrophobic residues (Evnouchidou et al., 2008), which have been shown to anchor into a hydrophobic pocket on the binding surface of the ERAP1 regulatory domain (ERAP1_R) (Gandhi et al., 2011; Sui et al., 2016). ERAP1 preferentially trims substrates of 9–16 residues in lengths down to 8–10 residues, which is precisely the size range required to fit into the binding cleft of MHC-I molecules (Guo et al., 1992; Silver et al., 1992). This length-dependent trimming of antigenic precursors is believed to be achieved by an allosteric mechanism, called the molecular ruler mechanism, via concurrent bindings of peptide substrates’ N- and C-termini by ERAP1’s catalytic and regulatory domains, respectively (Chang et al., 2005; Gandhi et al., 2011) (Fig. 1). Longer precursor peptides with a vast variety of sequences are thought to be sequentially trimmed by the concerted action of ERAP1 and its homolog ERAP2 (Lorente et al., 2013; Saveanu et al., 2005). These two enzymes complement each other with different sequence specificities to allow processing of antigen precursors with different anchor residues located at N-and C-termini. As an alternative mechanism of antigen trimming, it has been proposed that ERAP1 can trim antigens’ N-termini while they are bound onto the MHC-I molecules (Li et al., 2019). However, this MHC-I mediated process appears to be much slower than the trimming by ERAP1 itself in solution (Mavridis et al., 2020).

Figure 1: Allosteric activation of ERAP1.

(Top) ERAP1 processes peptide precursors by a molecular ruler mechanism to trim one residue at a time from substrates’ N-terminus at the catalytic Zn²⁺ site. To activate this catalytic activity, ERAP1 needs to bind simultaneously the peptide substrate’s N- and C-termini by its N-terminal catalytic and the C-terminal regulatory domains, respectively. This allosteric binding changes ERAP1 into a closed conformation with high activity, to allow an efficient trimming of one residue from the peptide N-terminus. For simplicity, the peptide substrate’s residue positions are numbered forward from the N-terminus as P1, P2, P3, and backward from the C-terminus as peptide carboxyl-terminal PC, the penultimate PC-1, and further towards the N-terminal direction PC-2, PC-3 etc. (instead of P1, P1’, P2’ … PC-1’, PC’).

(Bottom) Sequence comparison of peptides used to study their C-termini recognition by the ERAP1_R regulatory site, aligned backward from their C-termini at the PC position. These include two fused peptides studied in this report by crystallography and SPR binding assays (SPF and LPF), one fused peptide published previously (OPL (Sui et al., 2016)) and a free peptide used in the SPR analyses in solution (seq: LVAFKARKF): an optimized ERAP1 substrate (Evnouchidou et al., 2008). Orange oval represents a protein domain (ERAP1_R) to which peptide sequences shown are attached to the protein C-terminal end to facilitate crystallization (see Methods). Peptide residues included in the final structural models are underlined.

A published structure of ERAP1_R domain in complex with an antigenic peptide C-terminal sequence IINFEKL had provided structural insights to understand how ERAP1 recognizes the antigen C-terminus with a leucine at the PC anchoring position (Sui et al., 2016) (Fig. 1). In support of the molecular ruler mechanism (Chang et al., 2005; Gandhi et al., 2011), this C-terminal binding of an antigenic peptide precursor would allow a concurrent binding of a antigen’s N-terminus, about 9–10 residue long, to be trimmed in the ERAP1’s N-terminal catalytic site that is located about 29 Å from the ERAP1_R’s PC recognition site (Gandhi et al., 2011; Sui et al., 2016). As such, a substrate-length dependent trimming at the ERAP1’s catalytic site can be carried out through an allosteric activation by simultaneously docking the antigen’s C-terminus into the ERAP1’s regulatory site (Fig. 1). Thus the peptide C-terminus recognition by ERAP1_R is critical for shaping the immunopeptidomes to appropriate lengths to fit into the antigen binding cleft of MHC-I molecules (Guo et al., 1992; Silver et al., 1992). However, many molecular details of the ERAP1 molecular ruler mechanism remain to be examined. For example, how could ERAP1 recognize and bind different anchoring residues at peptide PC positions, and how does it accommodate a large pool of peptide precursors with different lengths? Also, how could SNP variants influence ERAP1 substrate sequence preference and result in biased immunopeptidomes of individuals? To address these knowledge gaps, we have carried out structural and biochemical studies of ERAP1 binding using the C-terminal sequence and anchor of a second MHC-I restricted antigenic epitope. We here report two complex structures with the antigenic epitope C-terminal ends bound to the ERAP1_R domain. The peptide in each of the two complexes carries the same C-terminal sequence derived from the same MHC-I restricted epitope, but differs in length by 4 additional alanines. Crystallographic and comparative analyses of these two new complexes with the previously published complex provide a structural basis to understand how ERAP1 can bind to different anchor residues at the antigenic peptide C-terminus. They also show how the sequence and length difference at the middle part of substrate peptides can be accommodate in the ERAP1 substrate cleft. Significance of these intermolecular interactions in crystals have also been validated by SPR (surface plasmon resonance) binding assays in solution. Uncovering of these peptide-ERAP1 interactions has provided insights to understand the molecular mechanism of ERAP1 to process peptides with a vast variety of sequences and a range of peptide lengths.

Materials and methods

Protein expression and purification

To facilitate crystallographic studies of the intermolecular interactions between ERAP1_R domain and substrate’s C-termini, we utilized a single-chain construct strategy described previously for protein purification and crystallization, using an Invitrogen baculovirus-insect cell expression system (Sui et al., 2015; Sui et al., 2016). Briefly, peptide sequences FKARKF and AAAAFKARKF were fused to the C-terminal end of the ERAP1 regulatory domain, referred as SPF and LPF fused peptide constructs, respectively (Fig. 1). More importantly, all the interactions between ERAP1 and peptides discussed in this report are intermolecular in nature, as observed in the crystals and confirmed by the surface plasmon resonance (SPR) analysis in solution (see Results and discussion). Both SPF and LPF constructs carry an N-terminal His₆-tag to facilitate protein purification. Gene cloning, protein expression and purification of these fused constructs were performed following the protocols described before (Sui et al., 2015).

To be applied as analytes for SPR binding assays with ERAP1_R, the His₆-tag of SPF and LPF constructs would need to be removed to avoid their direct binding to the HTG nickel sensor chip (Bio-Rad). For this purpose, a thrombin protease cleavage sequence (LVPRGS) was engineered between the N-terminal His₆-tag and the SPF or LPF single-chain construct. The His_6-tag removal was done by thrombin enzyme cleavage (Waugh, 2011) followed by gel filtration purification. Briefly, after protein purification, His₆-tagged protein was treated with 10U of thrombin enzyme (Calbiochem, San Diego, CA) per mg protein, and incubated at room temperature overnight. After cleavage, the protein was subjected to further purification to remove the cleaved His₆-tag, by incubating with Ni-NTA beads for 20 min with gentle shaking, followed by 5 min of centrifugation at 700g. The protein without His₆-tag was then collected in the supernatant fraction, which was further purified by a Superdex 200 gel filtration column (Amersham Pharmacia, Pittsburgh, PA). Target protein recovery and purity were determined by SDS-PAGE analysis.

Protein crystallization, data collection and processing

Hanging drop vapor diffusion methods were used for crystallization at 4°C, with a reservoir solution of 100mM Tris, pH 8.5, 12% w/v PEG8000. Micro-seedings were applied to further improve the quality of protein crystals. Crystals were flash frozen in liquid nitrogen after soaking in the reservoir solution containing 20% v/v glycerol. Diffraction data were collected on beamline 21-ID-D at Argonne National Laboratory synchrotron (Illinois, USA). The data sets were indexed and integrated using iMosflm and then scaled and merged using CCP4 (CCP4 (Collaborative Computational Project, 1994) to convert intensity to structure factor amplitudes for structure determination.

Structure determination

Complex structures were determined by molecular replacement methods using the Phaser-MR from the PHENIX software package (Adams et al., 2002), with an ERAP1_R domain structure (PDB ID 3RJO) as the search model. Refinements were carried out using the phenix.refine in the PHENIX software package (Adams et al., 2002), followed by simulated annealing refinement (Brunger and Rice, 1997; Hodel et al., 1992). Model buildings and adjustments with Fo-Fc and 2Fo-Fc maps were performed using COOT (Emsley and Cowtan, 2004). Table 1 summarizes the X-ray data and structure refinement statistics. The four crystallographically independent molecules in the P1 unit cell adopt a similar but slightly different conformations and electron density qualities, due to their different packing environments (Fig. S2). All discussions in this report base on a representative molecule with the best density (Chain B and Chain A for the SPF and LPF, respectively).

Table 1.

Crystal information, data collection and refinement statistics

	SPF-ERAP1_R	LPF-ERAP1_R
Resolution (Å)a	60.4–3.2 (3.4–3.2)	58.2–3.0 (3.2–3.0)
Space Groupb	P1	P1
Cell Dimensions:
a, b, c (Å)	58.4, 70.0, 121.2	57.0, 66.7, 121.7
α, β, γ (°)	90.1, 100.9, 90.2	90.1, 101.8, 90.1
No. of molecules per asymmetric unit	4	4
I/sigma-I	3.7 (1.6)	7.6 (2.4)
Completeness (%)	94 (95.8)	90 (87.5)
No. of reflections	65,591 (9,563)	71,232 (10,164)
Unique reflections	29,331 (4,376)	31,557 (4,503)
Rsym (%)c	12.4 (23.2)	6.1 (16)
Structure refinement
Resolution (Å)a	59.5–3.2	55.1–3.0
Rworkd	0.271	0.238
Rfreee	0.316	0.298
R.m.s. deviationsf
Bond-lengths (Å)	0.012	0.006
Bond-angles (°)	1.63	1.46
Ramachandran plot
Most favored regions (%)	94	93
Additional allowed regions (%)	5.5	6.5
Average B-factors (Å2)
Protein	45.87	57.73

^aNumbers in parentheses refer to the outermost (highest) resolution shell.

^bFor comparison, previously reported OPL-ERAP1_R complex was crystallized in the P2₁ space group, with the cell constants of a = 64.2Å, b = 66.8Å, c = 66.5Å, β = 110.2°, and with one molecule per asymmetric unit.

^cR_sym = Σ|I −〈I〉| /Σ(I), where I is the observed intensity and 〈I〉 is the weighted mean of the reflection intensity.

^dR_work = Σ‖F_o| −Fc‖ / Σ|F_o|, where F_o and F_c are the observed and calculated structure factor amplitudes, respectively.

^eR_free was calculated as R_work, but with 5% of the amplitudes chosen randomly and omitted from the start of refinement.

^fR.m.s. deviations are deviations from ideal geometry.

Surface plasmon resonance analysis of binding interactions

Surface plasmon resonance (SPR) was used to detect and quantify the interactions between the peptide C-terminus and the ERAP1_R domain (Nguyen et al., 2015; Torreri et al., 2005), using the Bio-Rad ProteOn XPR 36 protein interaction array system (Bio-rad, Portland, ME). For immobilization strategy, a noncovalent ligand capture method was selected to assure that the ligand protein with His₆- tag binds to the chip with same orientation, thus exposing the regulatory site of ERAP1 for peptide analyte binding. The ProteOn HTG sensor chip (Bio-rad, Portland, ME) was used to capture His₆- tagged ERAP1_R with nickel ions immobilized by the multivalent chelator trisnitrilotriacetic acid (tris-NTA). For all SPR binding assays in this study, an ERAP1_R regulatory domain with a C-terminal His₆-tag was immobilized on the HTG chip as a ligand. Two types of analytes were used to flow across ligand channels for analyzing binding kinetics: fused peptide (SPF and LPF) or free peptide (sequence: LVAFKARKF). Concurrent channels of minus ligand and/or minus analyte are included as reference or negative controls. For reference subtraction, we applied the method of Channel Reference according to the manufacturer protocols (the Bio-rad ProteOn software, Portland, ME), using either interspot or a minus ligand channel.

SPR studies were carried out by the manufacture’s protocol using a running buffer of 20mM HEPES, pH 6.2, 150mM NaCl and 0.005% Tween20. The same buffer was used for dilution of all ligand and analytes. First, the ERAP1_R ligand with a C-terminal His₆-tag was diluted to 5μg/ml and injected immediately after chip activation at the flow rate of 30μl/min for 300 s. After ligand immobilization, the running buffer was passed through both the ligand and analyte channels twice to remove any unbound ligand and to equilibrate the sensor chip. Analytes, either SPF or LPF without His₆-tag (see above for details in preparation) or free peptide, were prepared in different concentrations from 0–20μM and injected at a flow rate of 30μl/min for 200s followed by 300s of dissociation. To minimize non-specific binding of free peptides with the chip’s surface, we repeated SPR binding assays and obtained a similar result by using a modified running buffer with higher concentrations of NaCl and Tween-20: 20mM HEPES, pH 6.2, 400mM NaCl and 0.05% Tween-20.

Protein Data Bank accession codes.

The atomic coordinates and structure factors have been deposited in the Protein Data Bank with ID codes 7MWB and 7MWC for the SPF and LPF complexes, respectively.

Results and discussion

Structural studies of ERAP1_R recognition of peptide carboxylate end with a Phe anchoring residue

We have previously reported structural analyses of the binding mechanism for ERAP1 regulatory domain (ERAP1_R) to recognize antigenic peptide carboxylate end with a Leu anchoring side chain (Sui et al., 2016). These analyses were based on a crystal structure of ERAP1_R in complex with a peptide sequence of an ovalbumin-derived MHC class I-restricted epitope SIINFEKL (Rotzschke et al., 1991), designated as OPL for “ovalbumin peptide with Leu anchor” (Fig. 1). In this report, we extend these studies to understand the molecular mechanism of ERAP1_R to accommodate a large repertoire of peptide substrates with varying sequences and lengths. To this end, we designed two new ERAP1 binding peptides based on an optimized substrate sequence LVAFKARKF from a screening study of ERAP1 substrate preferences (Evnouchidou et al., 2008). One peptide carries the C-terminal sequence of FKARKF, designated as SPF for “short peptide with Phe anchor”, whereas the second peptide has four alanines inserted into the N-terminal side of the SPF sequence, AAAAFKARKF, named as LPF for “long peptide with Phe anchor” (Fig. 1). Crystal structures of ERAP1_R in complex with either SPF or LPF peptide have been determined. All statistics of data collection and model refinement are summarized in Table 1. Figure 2A shows a simulated annealed 2Fo-Fc omit density map at 1.0σ contour level with a continuous main-chain density for the bound SPF peptide near its C-terminus, which has a phenylalanine as the anchor residue at the PC position. The overall complex structure of SPF-ERAP1_R was quite similar to that of the previously reported OPL-ERAP1_R complex (Sui et al., 2016), with a slight difference in relative subdomain orientations. When comparing the complex structure of SPF with that of OPL, it is noticeable that the OPL complex possesses a more “closed” conformation than the SPF complex, by swinging a subdomain (known as domain IV (Kochan et al., 2011; Nguyen et al., 2011)) a bit towards the other subdomain (subdomain III). After aligning the subdomain III of the two complex structures, r.m.s.d. between the two subdomains IV is 0.75Å for all main-chain atoms, in reference to an r.m.s.d. of 0.54Å for subdomains III.

Figure 2: Recognition of SPF peptide C-terminus by the ERAP1 regulatory site.

(A) The gray electron density corresponds to a 2F_o-F_c simulated annealed omit map at 3.2 Å resolution with a contour level of 1.0σ, calculated without any peptide residues. The bound SPF peptide (sequence FKARKF) is shown as a thick stick model and colored by atom types: yellow for carbons, blue for nitrogens, and red for oxygens.

(B) Detailed view of the binding groove for SPF peptide C-terminus, with surrounding side chains of ERAP1 residues shown in grey bonds and labeled. The last two residues of the bound SPF peptide, Lys_PC-1 and Phe_PC are shown as a thick stick model and colored by atom types as in (A). Also shown is Tyr684 of ERAP1 which makes a direct hydrogen-bond contact (black dotted line) with the PC carboxylate end (shown in colored bonds). Salt bridge interactions among Arg807, Lys685 and the PC carboxylate group are shown as blue dotted lines.

Interactions of ERAP1_R binding groove with peptide SPF

The complex structure of the SPF peptide with the ERAP1_R domain provides detailed structural information to study key interactions for accommodating a substrate’s C-terminal end with a Phe anchoring side chain. Throughout this report, we refer to the peptide positions using the scheme denoted in Figure 1, with a simple numbering system P1, P2, P3 … PC etc. (instead of P1, P1’, P2’ … PC-1’, PC’). For example, for the SPF peptide (with a C-terminal sequence of FKARKF), peptide residue positions are denoted backwards from the C-terminus as the Peptide Carboxyl-terminal PC position (Phe), penultimate PC-1 position (Lys), and further towards N-terminal direction PC-2 position (Arg), PC-3 position (Ala) etc. As shown in Figure 2B, the hydrophobic side chain of Phe at the PC position (Phe_PC) points into the hydrophobic pocket defined by ERAP1 hydrophobic residues Leu733, Leu734, Val737 and Phe803. This is the same hydrophobic specificity pocket reported previously in the structure of OPL-ERAP1 complex (Sui et al., 2016), which is positioned about 29 Å away from the N-terminal catalytic zinc site of the full-length ERAP1 structures (Gandhi et al., 2011; Kochan et al., 2011; Nguyen et al., 2011) (Fig. 1). In addition, the PC carboxylate group forms direct hydrogen bonding with Tyr684, and the complex is further stabilized by salt bridges with Lys685 and Arg807 (Fig. 2B). No other hydrogen-bonding or charged interactions were found between other parts of the peptide and the ERAP1_R domain. Thus, the major specific recognition of peptides by ERAP1_R is located at the carboxylate end of the peptide substrates. These specific interactions show the importance of the side chain at the PC position, which serves as an anchor to stabilize the binding of the peptide C-terminus to the ERAP1 regulatory site. At the same time, limited interaction at other parts of peptide also provide a mechanism for ERAP1_R to accommodate antigen peptides with some flexibility in peptide sequences (see below).

Comparing C-terminal bindings of peptides OPL and SPF by ERAP1_R

Structural comparisons reveal common binding properties of ERAP1_R to recognize peptide C-terminal ends even when substrates have different anchor residues at the PC position. For these comparative analyses, we superposed the common interacting residues of ERAP1_R, and then compared specific interactions between peptides and ERAP1 in two complex structures: the new SPF-ERAP1_R complex and the previously reported OPL-ERAP1_R complex (Sui et al., 2016). Side chains of the PC anchoring residues of both peptides (Phe and Leu, respectively) are bound by the same hydrophobic pocket of ERAP1_R surrounded by residues Leu733, Leu734, Val737 and Phe803 (Fig. 3A). In addition, their PC carboxylate group interacts with Tyr684 via a direct hydrogen bond (Fig. 3B). However, some significant differences of binding configuration are noticed as well. For example, for peptide OPL, the main chain of its Lys at the PC-1 position interacts with Lys685 via a direct hydrogen bond (Fig. 3B). Nonetheless, for peptide SPF, the main chain carbonyl oxygen of the PC-1 residue points away from Lys685, resulting in a loss of the hydrogen bond contact (Fig. 3B). In addition, as shown in Figure 3C, the overall backbone conformations of these two peptides are different. Specifically, the main-chain traces of peptides SPF and OPL, between the PC-2 to PC-6 positions, run through different sides of the ERAP1 binding cleft (Fig. 3C). This is most likely due to an electrostatic potential difference between these two peptide sequences. Peptide SPF is positively charged with Lys, Arg and Lys at PC-1, PC-2 and PC-4 positions, respectively (Fig. 1). On the other hand, peptide OPL is likely to be roughly neutral with one positively charged and one negatively charged residues at the PC-1 and PC-2 positions, respectively. Indeed as shown in Figure 3C, main-chain trace of the positively charged SPF peptide (yellow tube) leans toward the right side of the ERAP1 binding cleft which has a negative (red) area of electrostatic surface potential. It is also interesting to note that in these complexes a prominent SNP variant at Gln730 is in close contact to the internal part of the bound peptide near the PC-5 position (Fig. 4A), This contact is located at the wider portion of the ERAP1 binding cleft, roughly 23 Å and 11 Å from the bound peptide N- and C-termini, respectively.

Figure 3: Comparison of SPF and OPL structures bound by the ERAP1 regulatory site.

(A) Schematic drawing of the specificity pocket shared by both bound peptide carboxyl ends Phe_PC and Leu_PC with surrounding side chains of ERAP1 residues shown and labeled. Bound peptides are indicated as thick stick models, while side chains of selected ERAP1 residues are shown as skinny stick models. Atoms are colored by atom type: yellow for carbons of SPF complex, green for carbons of OPL complex, blue for nitrogens, and red for oxygens.

(B) Schematic drawing of interactions of bound peptide PC and PC-1 residues with surrounding ERAP1_R residues. The residues are colored as in (A). Hydrogen-bond and salt-bridge interactions in SPF and OPL complexes are shown in yellow and green dotted lines, respectively.

(C) Comparison of SPF and OPL main-chain traces after a superposition of the ERAP1_R regulatory sites, shown in yellow and green tubes, respectively. ERAP1_R binding site is shown with a surface representation and colored by electrostatic potential, red for negative, blue for positive, and white for neutral. The anchor side chains at the PC position of both peptides are presented as stick models.

Figure 4: Comparison of the SPF peptide and a 15mer peptide bound in the ERAP1 substrate cleft.

(A) Comparing the SPF to a 15mer peptide (pose A of two conformation in (Giastas et al., 2019a), shown in sticks, when bound in the ERAP1 substrate cleft, shown in surface representation. These peptides are superimposed by all main-chain atoms of the ERAP1 regulatory domain (residues 529–939) in both complexes. Catalytic zinc⁺⁺ atom and side chains of Gln730 and Tyr684 of ERAP1, shown in spheres, make close contacts with the peptide P1, PC-5 and PC residues, respectively. Atoms are colored by atom type: yellow for carbons of SPF peptide, cyan for carbons of the 15mer peptide, magenta for carbons of ERAP1 selected residues, blue for nitrogens, red for oxygens, orange for phosphorus, and black for zinc.

(B) Shown is a change of the electrostatic surface potential at the SNP variant Q730E located at the middle section of the ERAP1 substrate cleft: for variant Gln730 (left) and variant Glu730 (right) modeled by a side-chain substitution based on a full-length ERAP1 structure (Nguyen et al., 2011), viewed in an orientation similar to (A). The approximate location of the PC-5 residue of the bound SPF peptide is marked. Blue surface represents a positive electrostatic potential while red represents a negative electrostatic potential. Amino acid side chains at Q730E are shown as stick models and colored by atom type: blue for nitrogens, red for oxygens, light-green and gray for carbons of Gln730 and Glu730, respectively.

Gln730 is highly associated with a predisposition to autoimmune diseases (Harvey et al., 2009), and its mutation Q730E has a major influence on substrate length preference (Stamogiannos et al., 2015). One plausible explanation is that Glu (E) is only a hydrogen acceptor with a negative charge, whereas Gln (Q) would be roughly neutral to serve as a hydrogen acceptor and/or a donor. As such the Gln (Q) to Glu (E) substitution would alter ERAP1’s potential to form hydrogen-bond and salt-bridge contacts with the internal sequences of peptide substrates. Moreover, the negatively charged Glu (E) could change the electrostatic potential at ERAP1_R binding surface. As shown in Figure 4B, the mutation from a Gln to Glu, essentially a change of amino to hydroxyl group, results in a more negatively charged surface. This change could result in a bias in binding and selecting peptide substrates, e.g. a bias towards more positively charged residues with a repulsion against negatively charged residues at peptide internal positions. On the other hand, the nearly neutral Gln (Q) might accommodate a wider range of peptide sequences with less bias against negatively charged peptide residues around the PC-5 position. Recently, a Behçet’s disease-associated ERAP1 allotype that carries a unique combination of SNPs Val349 and Gln725 has been characterized to be a very deficient aminopeptidase (Hutchinson et al., 2021). Similar to Q730E, the R725Q mutation is located at the middle section of substrate binding cleft that may have a synergistic effect on peptide bias with the M349V mutation positioned near the catalytic site.

Binding of LPF by the regulatory site of ERAP1

Although the two complexes with either SPF or LPF peptides grow into different crystal packings and/or environments with different unit cells (Table 1) and with b axis lengths differing by more than 4%, the modes of peptide C-terminus recognition by the ERAP1_R domain are strikingly similar. These structural comparisons provide a further structural basis to understand how ERAP1 could accommodate length flexibility into its substrate binding cleft. In the complex structure of ERAP1_R with LPF peptide, a well-defined C-terminal end further provides evidence that the peptide carboxylate end provides a common anchor for ERAP1 to bind and recognize. Although the electron density maps show a clear conformation for the Phe_PC anchor residue, density for other peptide side chains is weaker (Supplementary Fig. S1). This observation is consistent with the notion that the ERAP1 regulatory site binds peptides mainly by grasping peptide substrates’ C-terminal ends at the PC anchor together with their carboxylate ends. Due to structural flexibility distal from the peptide C-terminus, we can only build the main-chain trace for the internal part of the LPF peptide. Nonetheless, peptide structures at the carboxylate end and the PC anchoring side chain are well defined to analyze its specific contacts with the ERAP1_R binding cleft. As shown in Figure 5, the peptide Phe hydrophobic side chains of both SPF and LPF complexes at the PC position dive into the same specific pocket defined by Leu733, Leu734, Val737 and Phe803. Also similar to previously described structures of OPL complex, the peptide PC carboxylate group of LPF interacts with Tyr684 via a hydrogen bond (Fig. 5). In addition, the PC carboxylate groups also forms salt bridges with the guanidine group of Arg807. Also shown in Figure 5, the C-termini of peptides SPF and LPF are bound by the ERAP1_R with very similar conformation, especially near the Phe anchor residue at the PC position. It is interesting to note that there is a small and concerted shift of the Arg807 guanidine group with peptide carboxylate end in order to maintain a similar salt bridge. On the other hand, weak electron density for other parts of the peptide indicates a minimal binding by ERAP1 at other peptide positions. In fact, this weak binding with the internal part of peptide substrate could provide a mechanism for ERAP1 to accommodate a large pool of antigen peptides. This finding thus reveals another critical feature of ERAP1 for being capable of binding and processing peptide precursors with a vast variety of sequences, and to a less extent with a limited range of peptide lengths (see below).

Figure 5: Comparison of PC and PC-1 residues of peptides LPF and SPF bound by the ERAP1 regulatory site.

Bound peptide residues near their C-termini (at the PC and PC-1 positions) and surrounding ERAP1 residues for SPF and LPF complexes are superposed. Bound peptides are shown as thick stick models, whereas side chains of selected ERAP1 residues are shown as skinny stick models. Atoms are colored by atom type: yellow for carbons of SPF complex, cyan for carbons of LPF complex, blue for nitrogens, red for oxygens, and side chains of ERAP1 hydrophobic residues that form the specificity pocket are shown in light gray. Hydrogen-bond and salt-bridge interactions in these two structures are shown as dotted lines. Note the concerted movements of the Arg807 side chain and the orientation of the carboxylate end to maintain a similar salt-bridge in these two complexes.

Significance of the peptide C-terminus binding by ERAP1 observed in different crystals

X-ray crystallography is a powerful tool to visualize the protein-protein and protein-peptide interactions. However, there are always concerns about whether interactions observed in crystals are biologically/functionally relevant, rather than a crystal-packing artifact (Krissinel, 2011). Now we have determined crystal structures of ERAP1_R in complex with three different peptide sequences and lengths: OPL, SPF and LPF (Fig. 1). They all show the same mode of C-terminus recognition by the ERAP1 regulatory site, even though these complexes were crystallized in very different crystal packings and/or environments, with different space groups and unit cell parameters (Table 1). More specifically, although both crystals of SPF-ERAP1_R and LPF-ERAP1_R were packed in the P1 triclinic space group, the one longer in peptide length (LPF) is packed tighter with shorter unit cell values in the a axis (by 1.4Å) and in the b axis (by 3.3Å) (Table 1). In contrast, the previously reported OPL-ERAP1_R complex was crystallized in the P2₁ space group with different cell constants (a = 64.2Å, b = 66.8Å, c = 66.5Å, β = 110.2°) and packed differently inside crystals (Supplementary Fig. S2). Yet all three complexes reveal near identical peptide-ERAP1 interactions at peptide C-termini: through same hydrogen bonding and salt bridges with ERAP1 residues Tyr684 and Lys685/Arg807, respectively. In addition, side chains of the Phe or Leu anchor residue of peptides dock into the same specificity pocket made up of Leu733, Leu734, Val737, and Phe803 of ERAP1. In fact, this ERAP1 specificity pocket have also been shown to be able to accommodate peptides with a histidine anchor at the peptide PC position (Gandhi et al., 2011). These intermolecular interactions are the only extensive and consistent interactions among different crystals, suggesting that this common mode of peptide-ERAP1 interactions is the main intermolecular recognition in solution that dictated the crystallization packing, rather than the other way around. Indeed, a recently determined structure of a non-cleavable 15mer peptide analog, with both N- and C-termini bound in a P22₁2₁ crystal, shows the same mode of peptide C-terminus recognition by the ERAP1 regulatory site (Giastas et al., 2019a). As shown in Fig. 4A, main-chain trace of the SPF peptide runs along a similar path as a C-terminal conformation of the 15mer peptide. Thus, these consistent ERAP1-PC interactions reported here represents a common PC binding mode among several peptides in various crystal packings/environments. Altogether they point to a critical feature of ERAP1 to act as a molecular ruler for feeding correct length of antigenic peptides, typically 8–9 residues, in order to fit into the MHC-I binding cleft for antigen presentation (Guo et al., 1992; Silver et al., 1992).

SPR analysis of ERAP1_R to bind free peptide substrate or fused peptides

To further confirm that the intermolecular peptide-ERAP1_R interactions observed in multiple crystals reflect functional interactions in solution, we performed surface plasmon resonance (SPR) binding analyses using a few binding substrates as analytes: a free peptide and both the SPF and LPF single-chain fused constructs designed for crystallographic studies discussed above. For these studies, ERAP1_R was immobilized on the chip as ligand. To measure interactions of ERAP1 with a free peptide in solution, we synthesized the optimized ERAP1 substrate LVAFKARKF derived from an ERAP1 substrate preference screening (Evnouchidou et al., 2008), from which the C-terminal sequences of both SPF and LPF peptides are derived (Fig. 1). For comparisons, we also measured binding affinity in solution of the immobilized ERAP1_R ligand with the single-chain fused SPF or LPF peptides generated to facilitate crystallization. These three peptide analytes (SPF, LPF, and free peptide) are different in lengths and sizes but all carry a common sequence FKARKF at their C-terminal ends (Fig. 1). To measure binding affinity with immobilized ERAP1_R domain, different concentrations of analytes were injected, and real-time binding kinetics is shown in sensorgram (Fig. 6). After subtracting non-specific binding from the referenced minus-ligand channel, the net binding data fit best to a “two-state” kinetic model, suggesting a possible conformational change of ERAP1_R upon binding of various peptide analytes. Such a “two-state” kinetic model is consistent with the molecular ruler mechanism of ERAP1 in which an allosteric activation of the catalytic domain via the peptide C-terminus binding by the ERAP1_R domain is required for peptide trimming (Fig. 1). The equilibrium dissociation rate constant (K_D) is calculated based on the association and dissociation rate constants derived for these SPR studies, using the Bio-rad ProteOn Manager software and summarized in Table 2. The Chi² indicates the differences between the experiment data points and the fitted values. As shown in Table 2, all Chi² values were significantly less than 10% of the corresponding Rmax values, indicating a reasonable and fairly confident data fit.

Figure 6: SPR sensorgram for analyzing interactions between ERAP1_R ligand and various peptide analytes.

(A) Interactions of ERAP1_R with SPF analyte at concentrations from 1–15μM.

(B) Interactions of ERAP1_R with LPF analyte at concentrations from 1–10μM.

(C) Interactions of ERAP1_R with free peptide (LVAFKARKF) at concentrations from 5–20μM.

Table 2.

Affinities and kinetic constants calculated from SPR binding assays

Analytes	ka (M−1s−1)	kd (s−1)	KD (μM)	Rmax (RU)	Chi2 (RU2)
SPF	3.5×103	1.12×10−1	32	82.14	6.57
LPF	4.27×103	9.10×10−2	21.3	93.37	1.67
Free Peptide (LVAFKARKF)	7.34×102	1.45×10−2	19.8	10.14	1.43

Affinity constant for the binding of each peptide analyte onto the ERAP1_R ligand was calculated from both the measured on-rate and off-rate kinetic data. The listed kinetic constants are calculated based on the SPR experimental results shown in Figure 6.

The binding affinity constant between the free peptide and the ERAP1_R domain (K_D) is calculated to be 19.8 μM (Table 2). The binding affinity of ERAP1_R with fused peptide analytes also shows low micromolar affinity: 32 μM for SPF and 21.3 μM for LPF. It is plausible that other portions of fused protein also contribute to the binding affinity measured here. However, the molecular packing observed inside the P2₁ and P1 crystals (Supplementary Fig. S2) indicates that binding of the fused peptide to the ERAP1_R substrate cleft is the main driving force for the intermolecular interactions. It was thought that weak interactions with a K_D greater than 100 μM would have a chance of being masked by a crystal packing artifact (Krissinel, 2011). However, SPR analyses show that the binding affinities between peptide C-terminus and ERAP1 regulatory site are all with a K_D value in low micromolar ranges, showing a significantly tighter binding than typical crystallization artifacts. It is worth noting that these peptide analytes carry the same C-terminal sequence, FKARKF, but differ dramatically in lengths and sizes (1 vs 55 kDa). The similarities of K_D value calculated among the free peptide and the two single-chain fused peptides suggest that peptide C-terminal sequences, rather than their length or size, are the major determinants of binding affinity to the regulatory site of ERAP1_R. These SPR binding assays suggest the interactions of ERAP1_R with C-terminal fused peptide in crystals resemble the ones with free peptide in solution. Altogether, this study provides further structural and biochemical evidence for the functional significance of the peptide C-terminus recognition in ERAP1’s molecular ruler catalysis.

Conclusions

It appears that ERAP1 has co-evolved with MHC-I molecules to have a similar preference and flexibility in substrates’ sequences and lengths, with dominant residues at certain positions of antigenic peptides. MHC-I restricted antigens have long been recognized to have canonical sequence motifs with preferred residues at a few anchor positions (Rammensee et al., 1999). Structural studies had revealed that these so-called anchor residues in peptides provide peptide-MHC complex stability by locking into complementary specificity pockets located inside the binding cleft of MHC-I molecules (Guo et al., 1992; Guo et al., 1993). Similar specificity pockets have also been found in the regulatory site of the ERAP1 binding cleft and interact with peptide PC anchoring side chain, in order to carry out its molecular ruler catalysis (Gandhi et al., 2011; Sui et al., 2016). In this study, we determined multiple structures of ERAP1 regulatory domain (ERAP1_R) in complex with different peptide sequences and lengths: SPF and LPF peptides. By comparing to the previously published structure of ERAP1_R in complex with another epitope sequence OPL, current structural and biochemical studies provide further insights into unique properties of ERAP1. In shaping the peptide repertoire available for presentation by MHC molecules, ERAP1 carries out the trimming of antigen precursors by a unique molecular ruler mechanism in a peptide-length and sequence dependent manner. This activity is enabled by ERAP1’s unique capabilities to concurrently bind antigen peptide N- and C-termini by its N-terminal catalytic and C-terminal regulatory domains (ERAP1_R), respectively (Fig. 1). Even though complexes with different peptide sequences and/or lengths form different crystal packings/environments with various unit cell sizes, they all maintain identical intermolecular contacts between the ERAP1 regulatory site around Tyr684 and the peptide C-carboxylate end. In addition, the side chain of the peptide PC anchor residue, either a Phe, Leu or His, tucks into the same hydrophobic specificity pocket around Leu734 in the ERAP1 regulatory domain. These shared interactions are the major binding determinants for ERAP1 to recognize the common or similar features in the immunopeptidomes. It is well known that different MHC-I molecules prefer different PC anchor residues, but predominantly with either a hydrophobic or a positively charged side chain (Guo et al., 1992; Rammensee et al., 1999; Rotzschke et al., 1994). There is thus no surprise that, to generate appropriate peptides for MHC-I presentation ERAP1 has dual specificities on the substrate PC anchor, either a positively charged or a hydrophobic side chain (Evnouchidou et al., 2008). While this study identifies the specificity pocket to bind a hydrophobic anchor, peptides with a charged PC anchor are likely to be accommodated into a different specificity pocket. Recently, a 10mer peptide analog was found to place its charged Lys PC anchor at a different location, by a salt bridge interaction with Asp766 of ERAP1 (Giastas et al., 2019a). However, validity of this site as a specificity pocket for a charged PC anchor remains to be confirmed, since no other extensive interactions between the 10mer PC Lys and ERAP1 residues were noted. Alternatively, we had previously proposed another negatively charged pocket around Glu802 and Glu831 of ERAP1_R regulatory domain as a potential specificity pocket to bind peptides with a Lys or Arg PC anchor (Gandhi et al., 2011). As discussed there, additional binding affinity could come from the conserved Arg841 to make direct contacts with the peptide carboxylate end.

On the other hand, some differences in binding configurations are observed as a mechanism for ERAP1 to accommodate peptides with different sequences and/or lengths. First, in the structure of SPF complex, there is a shift at main chain carbonyl oxygen of PC-1 Lys, whereas the corresponding PC-1 carbonyl oxygen of OPL points at a different orientation to make a hydrogen bonding with Lys685 of ERAP1 (Fig. 3B). Furthermore, main chain traces of peptides OPL and SPF wind through different sides of the ERAP1 binding cleft (Fig. 3C), likely due to their difference in electrostatic potential (Fig. 4B). Furthermore, well defined electron density was observed near the peptide C-terminus and the PC anchor, whereas density for other residues distal to the C-terminal end of peptide is weaker, in particular the side chains of the peptide internal residues. This observation is consistent with the notion that ERAP1 binds mainly on the antigenic peptide N- and C-terminal ends and its PC anchor residue, but allows a dynamic and/or static flexibility in other part of peptide substrates (Fig. 1). The latter feature is unique and critical for ERAP1 to be able to accommodate and trim a large pool of antigenic precursors with various internal sequences and lengths.

Altogether, structural analyses and SPR binding assays using three peptides with different internal sequences and lengths or sizes further confirm the major determinants for ERAP1 regulatory domain to trim and shape immunopeptidomes. It requires a specificity pocket in the ERAP1 to grasp the side chain of PC anchor residue of peptides. ERAP1 also can recognize the common feature of peptide carboxylate ends, by making a hydrogen-bond contact with Tyr684, and forming salt bridges with Lys685 and/or Arg807. Meanwhile ERAP1 also has a somewhat non-discriminatory surface near the middle part of its binding cleft, around the prominent SNP variant at Gln730. These structural features enable ERAP1 to exhibit a high specificity towards peptide C-termini, yet with certain flexibility to accommodate peptides with different internal sequences and lengths in the immunopeptidomes.

Due to its critical functions and associations with chronic and infectious diseases, ERAP1 has been an emerging target for developing potential therapeutics to treat various diseases (Drinkwater et al., 2017; Hanson et al., 2019; Reeves et al., 2020). However, targeting an aminopeptidase to treat human diseases is challenging because there are nine characterized and closely related aminopeptidases in humans (Rawlings et al., 2016). These enzymes belong to the M1 family of peptidases (M1APs) that include the oxytocinase subfamily members: ERAP1, ERAP2 and insulin-regulated aminopeptidase (IRAP) (Hanson et al., 2019; Tsujimoto and Hattori, 2005). Hence, any lead compounds used in therapeutic development must be carefully designed to ensure a selective targeting against a single M1AP over other family members to prevent undesired side effects. Nonetheless, current inhibitor-developing strategies rely heavily on Bestatin derivatives or phosphinic pseudopeptides that aim at the highly conserved catalytic site in the N-terminal domain of ERAP1 (Drinkwater et al., 2017; Georgiadis et al., 2019; Maben et al., 2020; Reeves et al., 2020; Weglarz-Tomczak et al., 2016) (Fig. 1). Although potent, these inhibitors also inhibit many other essential M1AP members, lead to undesired side effects (Drinkwater et al., 2017). In contrast, the ERAP1 C-terminal regulatory domain studied here exhibits a substantial sequence and structural variations among M1AP members (Gandhi et al., 2011; Sui et al., 2016). When aligned to its most closely related enzyme ERAP2, ERAP1 has sequence identities of 65% and 45% for their N-terminal catalytic and C-terminal regulatory domains, respectively. When comparing to another M1AP member aminopeptidase N (Joshi et al., 2017), while their sequence identity of the N-terminal catalytic domain remains to be high (50%), the sequence identity of their C-terminal domains drops to 25%. As such, the ERAP1 C-terminal regulatory site strikes a good balance between a structural uniqueness for specific targeting and a sequence diversity for selectivity over other M1APs. It is important to note that ERAP1 utilizes two distinct binding sites, although in proximity, to recognize the PC-end of immunopeptidomes: a conserved motif (residues around Tyr-684 in Fig. 3B) and a polymorphic specificity pocket (residues around Val-737 in Fig. 3A) to recognize the common PC-carboxylate and the variable PC side-chain anchors, respectively. Recently, small molecules either in crystallization buffer (Giastas et al., 2019b) or from a high-throughput screen (Liddle et al., 2020; Maben et al., 2020) have been found by crystallography or computationally docking to bind to the PC-carboxylate binding motif of the ERAP1 regulatory domain. Two of these compounds have also been shown to modulate ERAP1 activity, likely via an allosteric activation similar to that by short peptides (Gandhi et al., 2011; Nguyen et al., 2011). Together with this conserved PC-carboxylate binding motif, the novel hydrophobic specificity pocket uncovered in this report to differentiate variable PC-anchors may present a new target for selective agents or therapeutics to modulate ERAP1 functions in shaping the immunopeptidomes.

Abbreviations:

ER

endoplasmic reticulum

ERAP1

endoplasmic reticulum aminopeptidase 1

ERAP1_R

ERAP1 regulatory domain

LPF

long peptide with Phe anchor

MHC-I

major histocompatibility complex class I

OPL

ovalbumin peptide with Leu anchor

SPF

short peptide with Phe anchor

SPR

surface plasmon resonance

rmsd

root mean square deviation/displacement