Discriminative T-cell receptor recognition of highly homologous HLA-DQ2-bound gluten epitopes

Celiac disease (CeD) provides an opportunity to study the specificity underlying human T-cell responses to an array of similar epitopes presented by the same human leukocyte antigen-II (HLA-II) molecule. Here, we investigated T-cell responses to the two immunodominant and highly homologous HLA-DQ2.5-restricted gluten epitopes, DQ2.5-glia-α 1a (PFPQPELPY) and DQ2.5- glia-ω 1 (PFPQPEQPF). Using HLA-DQ2.5:DQ2.5-glia-α 1a and HLA-DQ2.5:DQ2.5-glia-ω 1 tetramers and single-cell αβ T-cell receptor (TCR) sequencing, we observed that despite similarity in biased variable-gene usage in the TCR repertoire responding to these nearly identical peptide:HLA-II complexes, most of the T cells are specific for either of the two epitopes. To understand the molecular basis of this exquisite fine specificity, we undertook Ala substitution assays revealing that the p7 residue (Leu/Gln) is critical for specific epitope recognition by both DQ2.5-glia-α 1a and DQ2.5-glia-ω 1 -reactive T-cell clones. We determined high-resolution binary crystal structures of HLA-DQ2.5 bound to DQ2.5- glia-α 1a (2.0 Å) and DQ2.5-glia-ω 1 (2.6 Å). These structures disclosed that differences around the p7 residue subtly alter the neighboring sub-structure and electrostatic properties of the HLA-DQ2.5-peptide complex, providing the fine specificity underlying the responses against these two highly homologous gluten epitopes. This study underscores the ability of TCRs to recognize subtle differences in the peptide:HLA-II landscape in a human

Here, we have studied the T-cell response towards the epitopes DQ2.5-glia-α1a (PFPQPELPY) and DQ2.5-glia-ω1 (PFPQPEQPF).These immunodominant gluten epitopes in CeD are highly homologous, with only two amino acid differences in the 9-mer core region.Despite the epitope similarity, we found that the majority of patient-derived T cells, both isolated from blood or the celiac gut lesion, are specific for either of the two epitopes with limited cross-reactivity.This suggests that there are important differences in how T cells recognize these peptide:HLA-II complexes, as the immune system of the patients are exposed simultaneously to both epitopes when they consume gluten-containing food.To explore the basis for specific recognition of the homologous epitopes, we performed T-cell proliferation assays with amino-acid substituted epitopes, undertook single-cell TCRαβ-gene sequencing and solved the crystal structures of HLA-DQ2.5 complexed with either DQ2.5-glia-α1a or DQ2.5-glia-ω1.Despite biased usage of common V genes, the majority of the T cells were specific to only one epitope.The p7 residue in both epitopes was critical for the specific TCR recognition by the discriminatory T-cell clones (TCCs).The crystal structures of the two peptide:HLA-DQ2.5 complexes, whilst similar, exhibited local structural perturbations around the p7 residue.Hence, this study demonstrates the ability of TCR to distinguish subtle differences in peptide:HLA-II topology.Consequently, in human T-cell mediated diseases like CeD, discrete alterations in the peptide:HLA-II landscape can profoundly shape the disease-driven immune response.

Majority of T cells in blood and gut of CeD patients specific for DQ2.5-glia-α1a or DQ2.5glia-ω1 show exquisite fine specificity
We identified distinct populations of CD4 + T cells that bound the HLA-DQ2.5:DQ2.5-glia-α1aor HLA-DQ2.5:DQ2.5-glia-ω1tetramers in blood and gut of 12 CeD patients.(Fig. 1A).The relative ratio of CD4 + T cells positive to both DQ2.5:DQ2.5-glia-α1a-andDQ2.5:DQ2.5-glia-ω1-tetramerswith CD4 + T cells that are positive to either of the two tetramers in samples obtained from 11 CeD patients was 0.06 (mean).This suggests that the majority (~ 95 %) of these CD4 + T cells recognize either of the two tetramers (Fig. 1B, Table S1) indicating that despite only two amino acids difference in the 9mer core region of the two gliadin peptides, most of the in vivo primed T cells in blood and gut of CeD patients discriminate between the two epitopes.
The HLA-DQ:gluten tetramers (including HLA-DQ2.5:DQ2.5-glia-α1aand HLA-DQ2.5:DQ2.5glia-ω1)give staining of effector memory CD4 T cells of CeD patients, but not of healthy subjects (14,15).We validated the specific staining of HLA-DQ2.5:DQ2.5-glia-α1avs HLA-DQ2.5:DQ2.5-glia-ω1tetramers with TCCs that were generated from tetramer-sorted cells by antigen-free expansion and cloning by limited dilution.On re-testing of antigen specificity, all (nine) TCCs generated from PBMCs of the patient CD1383, stained with the tetramer used for sorting and showed a proliferative response towards the respective peptide (Fig. 1C and Fig. 1D).One of the five TCCs generated from HLA-DQ2.5:DQ2.5-glia-α1atetramer binding CD4 + T cells and one of four of the TCCs generated from HLA-DQ2.5:DQ2.5-glia-ω1tetramer binding CD4 + T cells stained with both the tetramers and gave a response to both peptides in T-cell proliferation assays.The cross-reactive T-cell clones displayed a higher fluorescence intensity to the tetramer originally used for their isolation.None of the nine TCCs stained with the other HLA-DQ2.5:glutentetramers (HLA-DQ2.5:DQ2.5-glia-α2,HLA-DQ2.5:DQ2.5-glia-ω2and CLIP2), indicating the antigen specificity of the tetramer staining (Fig. 1C).In addition, none of these TCCs showed response to any other gluten epitopes when tested in a T-cell proliferation assay against a panel of known gluten peptides at a high concentration of 10 µM (Fig. 1D).Taken together, the results of the HLA-DQ:gluten tetramer staining of T cells in blood and gut as well as the proliferation assay and HLA-DQ:gluten tetramer staining of established T-cell clones (the low number here prevents an exact estimate of cross-reactivity), strongly indicate that T cells of CeD patients generally discriminate between the two epitopes.

The p7 residue of the DQ2.5-glia-α1a and DQ2.5-glia-ω1 epitopes is critical for T-cell recognition
To explore the differences in recognition of these two highly homologous epitopes we measured the effect of single Ala substitutions at each position in the two epitopes (Table 1) in a T-cell proliferation assay.The sequences of the peptides used in these T-cell proliferation assays were identical to the epitope sequences that were part of the HLA-DQ:gluten tetramers.In this analysis TCCs that stained with the HLA-DQ2.5:DQ2.5-glia-α1aor HLA-DQ2.5:DQ2.5glia-ω1tetramers and that had reactivity with only one of the two epitopes, were tested (Table 2).Reactivity pattern of the DQ2.5-glia-α1aspecific-andDQ2.5-glia-ω1-specific TCCs to the single Ala-substituted DQ2.5-glia-α1a and DQ2.5-glia-ω1 peptides were normalized to the response to the wild-type peptide.Both DQ2.5-glia-α1a-(Fig.2A) and DQ2.5-glia-ω1-specific (Fig. 2B) TCCs lost reactivity on Ala substitution at p2 and p7.However, substitution at p9, a position that differs between the two epitopes, had no effect on the reactivity of the TCCs.This is in line with the observations from a previous study where p7 was critical, p9 was dispensable and the p2 had variable effects on Tcell recognition of the DQ2.5-glia-α1a epitopes by the DQ2.5-glia-α1a-specificTCCs (10).Unique to all DQ2.5-glia-α1a-specific TCCs was the loss of reactivity on Ala substitution of residues p6 and p8.Two of the three TCCs lost reactivity on substitution at p3 and p4 while the third TCC was sensitive to substitution at p5. DQ2.5-glia-ω1-specific TCCs exhibited varying responses to Ala substitutions at positions other than p2 and p7.Substitution at p3 and p8 resulted in loss of reactivity in three of four TCCs.In brief, the p2 and p7 amino acids were important for T-cell recognition of both the DQ2.5-glia-α1a and DQ2.5-glia-ω1 epitopes.
We also performed similar analysis of unique clonotypes obtained by paired TCRα and TCRβ sequencing of single CD4 + T cells isolated from in vitro cultured T-cell line (TCL) generated from gut biopsies of CeD patients using the HLA-DQ2.5:DQ2.5-glia-α1a(n = 50) or HLA-DQ2.5:DQ2.5-glia-ω1tetramers (n = 20) (Table 4 and Fig. S1, Fig. S2B).The TRAV and TRBV usage for both epitopes in data derived from T cells that were directly isolated from blood or gut biopsies was generally reproduced in the data obtained from TCLs.
We then aligned the CDR3α and CDR3β sequences of the TCRs specific for DQ2.5-glia-α1a and DQ2.5-glia-ω1 (Fig. 3C) to analyze the CDR3 amino acid usage and positioning.The T cells were isolated using HLA-DQ2.5:DQ2.5-glia-α1aor HLA-DQ2.5:DQ2.5-glia-ω1tetramers, either directly from blood or gut biopsies or from in vitro cultured TCLs from gut biopsy.We did not observe any significant selection pattern in the amino acid usage and positioning in the CDR3 sequences.
We then examined the most frequently expressed V-genes in TCRs specific for the four immunodominant gluten epitopes (DQ2.5-glia-α1a,DQ2.5-glia-ω1, DQ2.5-glia-α2 and DQ2.5glia-ω2).We analyzed unique clonotypes obtained by paired TCRαβ sequencing of single T cells isolated either from TCLs or directly from blood or gut biopsies of CeD patients using HLA-DQ2.5 tetramers (Fig. S2B).TRAV4 was among three most frequently used V-gene segments in the TCR repertoires specific for all the four epitopes.As all these gluten epitopes are restricted by HLA-DQ2.5, this feature of TRAV4 bias could be dependent on HLA interactions.

Crystal structures of the DQ2.5-glia-α1a and DQ2.5-glia-ω1 complexes
To gain an understanding of how two highly homologous epitopes give rise to two separate populations of T cells, we examined the corresponding peptide:HLA-II landscapes.Given the limited extent of cross-reactivity between the two gliadin determinants, it was unclear whether they would sit differently or within the same register in the HLA-DQ2.5 molecule.The observations made from the Ala substitution experiment performed on the TCCs is also based on the assumption that the substituted peptides bind in the identical binding registers.To establish this, we determined, to high resolution, the crystal structures of HLA-DQ2:DQ2.5-glia-α1a(2.0 Å) and HLA-DQ2:DQ2.5-glia-ω1(2.6 Å) (Fig. 4A, 4B and Table 5) for data collection and refinement statistics).These complexes were solved with same linker peptide for both epitopes and accordingly any structural variation between these two binary complexes could be attributed to differences in the two epitopes bound within the antigen-binding cleft.While the crystal symmetry for the two crystals were different (P222 for DQ2.5-glia-α1 versus C121 for DQ2glia-ω1), the crystal packing of the two structures was overall similar and, notably, with no crystal contacts involving the peptide positions p7 and p9.The HLA-DQ2.5:DQ2.5glia-α1astructure aligned well with previously determined binary HLA-DQ2.5:DQ2.5-glia-α1astructure that was solved at lower resolution (16).
The higher resolution structure (reported here) will be compared to the HLA-DQ2.5:DQ2.5glia-ω1structure.The electron densities corresponding to the two peptides were clear (Fig. S4).

Discussion
The CD4 + T-cell response to gluten epitopes presented by the disease-predisposing HLA-DQ molecules is an essential part of the pathogenesis of CeD.Here we have studied two homologous and immunodominant gluten epitopes, DQ2.5glia-α1a (PFPQPELPY) and DQ2.5-glia-ω1 (PFPQPEQPF).Despite only two amino acids difference in the 9-mer core region of the epitopes, the majority (~ 95 %) of T cells in blood and gut of CeD patients discriminate between the two epitopes with only a proportion of the T cells being cross-reactive.We found that the p7 residue was uniformly critical in both epitopes for discriminatory TCR recognition by all TCCs.Accordingly, we addressed the molecular basis underpinning the fine specificity of this response.
The p7 residues served as anchor residues in both HLA-DQ2.5:DQ2.5-glia-α1aand HLA-DQ2.5:DQ2.5-glia-ω1and their sidechains were only partially exposed in the binary crystal structures.Notwithstanding, it is striking that the p7 residue is critical for specific T-cell recognition of both DQ2.5-glia-α1a and DQ2.5glia-ω1 as well as DQ2.5-glia-α2 and DQ2.5glia-ω2.In ternary crystal structures of three unique complexes of TCR:HLA-DQ2.5:DQ2.5glia-α2, the large sidechain of p7-Tyr forms critical interactions with the TCR, explaining its vital role for T-cell recognition of this epitope.However, the ternary structure of the S2 TCR (TRAV4*01-TRBV20-1*01) specific for the DQ2.5-glia-α1a epitope in complex with HLA-DQ2.5 (10) revealed that this TCR does not make a direct contact with the less accessible p7-Leu residue.In spite of this, the TCC from which the S2 TCR was derived, along with ten other DQ2.5-glia-α1a-specificTCCs, uniformly lost reactivity towards DQ2.5-glia-α1a on p7 substitution in T-cell proliferation assays (10).The S2 TCR does discriminate between DQ2.5glia-α1a and DQ2.5-glia-ω1, suggesting that even though the p7-Leu is not making direct contact to the TCR it is, in some way, influencing antigen recognition.To address this conundrum, we determined the crystal structures of HLA-DQ2.5 complexed with the DQ2.5-glia-α1a and DQ2.5-glia-ω1 epitopes, which revealed that the two epitopes were accommodated by HLA-DQ2.5 using the same register.While the two peptide-HLA-DQ2.5 binary complexes were similar, differences around the p7 residue altered the neighboring sub-structure of the HLA-DQ2.5 molecule and associated electrostatic properties.Despite these relatively subtle differences in peptide:HLA-II topologies observed for the two immunodominant gluten epitopes, the S2 TCR must be able to detect them in a discriminatory manner.In the S2 TCR ternary complex, these p7 neighboring residues (specifically, HLA-DQ2.5 Asp66β, Glu69β and Arg70β) form an extensive interface with CDR1α, CDR3β and αframework residues of the S2 TCR, suggesting that the nature of this local environment is involved in epitope discrimination.Similar to the previous study, we found that for DQ2.5-glia-α1a-specific TCCs the p7-Leu was uniformly critical in specific TCR recognition.The DQ2.5glia-α1a-specific TCCs analyzed in the two studies use many different TRAV/TRBV pairings suggesting that this critical role of p7 is not contingent of a particular TCR gene usage.Interestingly, the p7-Gln of the DQ2.5-glia-ω1 epitope was also critical in specific TCR recognition for the DQ2.5-glia-ω1-specificTCCs that also express fairly diverse TCRs.These observations suggest that the different p7 amino acids in DQ2.5-glia-α1a and DQ2.5-glia-ω1 epitopes induce subtle differences that are sensed by the TCRs giving discriminatory recognition of the two peptide:HLA-II complexes.Similar features of impact on TCR recognition by a MHC buried anchor residues, were observed for I-Ek and a hemoglobin epitope where substitution at the p6 residue affected T-cell recognition (17).Hence, these results indicate that anchor residues that are buried in the MHC II and make no direct contact with the TCR, can indirectly influence the specific TCR recognition.
Examining the wealth of TCR gene sequence data now accumulated for the four immunodominant HLA-DQ2.5-restrictedgluten epitopes (DQ2.5-glia-α1a and DQ2.5-glia-ω1, DQ2.5-glia-α2 and DQ2.5-glia-ω2) in CeD ( 12), we observe that TRAV4 is among the most frequently expressed TRAVs in T cells specific for all of the four epitopes.This TRAV4 bias does not appear to be associated with a common TRBV usage or conservation of CDR3 sequences.The crystal structure of DQ2.5-glia-α1a-specific TCR using TRAV4 revealed that the TRAV4 bias against DQ2.5-glia-α1a is an effect of interactions between germline-encoded TCR residues, most prominently Tyr38α, with residues of the β-chain of HLA-DQ2.5 (Arg70β and Arg77β) (10).Similarly, HLA-DQ2.5:DQ2.5-glia-α2specific TCRs encoded by TRAV26-1, a phylogenetically close V gene with high sequence relatedness with TRAV4, formed analogous sets of interactions in ternary crystal structures.Therefore, we suggest that the common TRAV4 bias in TCRs specific for immunodominant gluten epitopes restricted by HLA-DQ2.5 could be an outcome of conserved interaction between germline residues encoded by TRAV4 and HLA-DQ2.5.
CeD provides an opportunity to study the natural immune response in humans towards a natural antigenic system that comprises a vast array of similar peptide sequences.Here we show that despite minor differences in peptide:HLA-II topologies and high similarity in TCR V-gene usage, the majority of the T cells discriminate between the two homologous and immunodominant gluten epitopes.For CeD, this implicates that highly homologous peptides have the potential to engage separate T-cell populations thereby resulting in broader and more robust T-cell responses to gluten.Of general implication, the study underscores the exquisite sensitivity of TCRs to detect subtle differences in peptide:HLA-II complexes.

Patient material
We obtained ~60 ml of citrated full blood and six gut biopsies taken as part of a gastroduodenoscopy, using an Olympus H-180 endoscope and a regular biopsy forceps (Olympus), from both untreated and treated CeD patients.PBMCs were isolated by Ficoll-based density gradient centrifugation from the blood samples and cryopreserved for later use.Gut biopsies were processed to obtain single-cell suspension prior to cryopreservation.In brief, freshly collected biopsies were washed twice with 2 mM EDTA in 2% FCS at 37 ºC for 10 min (to remove epithelial layer) followed by collagenase digestion (1 mg/ml) at 37 ºC for 30-60 min, homogenization by syringe and filtration.We used previously established and cryopreserved T-cell lines (TCLs) that had been generated from single gut biopsy, of which some are published (18,19).Two new TCLs (CD1174 and CD1178) were generated by incubating gut biopsy with native chymotrypsin-digested gluten and deamidated chymotrypsin-digested gluten (both at 20 µg/mL) for 3-5 days followed by addition of interleukin (IL)-2/IL-15.

Generation and TCR sequencing of TCCs
The TCCs were generated from tetramer-sorted CD4 + T cells and sequenced as described in previous publication (12).In brief, TCCs were generated using cloning by limited dilution and expanded without antigens followed by mRNA isolation, switching mechanism at the 5'terminus of the RNA transcript based cDNA synthesis, PCR amplification and finally Sanger sequencing.TCCs from patient CD442 and CD1340 used for Ala scans in the current study were generated in the course of another project (7).

T-cell proliferation assay
T-cell proliferation assays were carried out as described previously (21).In brief, 75,000 APCs (HLA-DQ2.5homozygous Epstein-Barr virus (EBV)-transformed cells) were irradiated 75 Gy and incubated with 10 µM peptides at 37 o C for 24 h before the adding 50,000 T cells.48 hours later, cultures were pulsed with 1 µCi 3 Hthymidine followed by harvesting and scintillation counting after 16-20 hours.The TCCs were identified as peptide-specific if the stimulation index (ratio of cpm after antigen stimulation and cpm after medium stimulation) was higher than 3.

Single-cell TCR gene sequencing
We used a previously published protocol, based on nested PCR amplification using multiplex TRAV and TRBV primers, for single-cell TCR sequencing (22).The protocol was slightly modified by performing cDNA synthesis and the first PCR reaction in two separate steps.In short, single cells were sorted directly in 96-well plates containing 5 µl capture buffer (20 mM Tris-HCl (pH 8.0), 1% NP-40 and 1 U/µl RNase Inhibitor (optional).The cDNA mix (5 µl of 1X SSII RT buffer, 1 mM dNTP, 2.5 mM DDT, 1 µM oligo d(T), 1 µM reverse TRAC (5'-AGTCAGATTTGTTGCTCCAGGCC-3') and TRBC (5'-TTCACCCACCAGCTCAGCTCC-3') primers, 1.5 U/µl RNase Inhibitor, 2.5 U/µl Superscript II in final 10 µl reaction volume) was added and cDNA synthesis was carried out at 42 ºC for 50 min followed by an inactivation step at 72 ºC for 10 min.The original protocol was followed to obtain a purified PCR product, which was then sequenced with 250-bp pairedend sequencing using Illumina MiSeq platform at the Norwegian Sequencing Centre (Oslo University Hospital).All the raw data generated in this study has beenuploaded to the European Genome-phenome Archive (EGAS00001003245).

Processing of TCR gene sequences
Processing of the raw sequencing data obtained from Illumina NGS was carried out in a multistep pipeline.The pipeline composed of: quality filtering of low-quality reads (Q < 30), annotation of the header with barcodes (row, plate and column) and gene (TRA/TRB) information, pairing and assembly of the annotated forward and reverse reads.These assembled reads were then collapsed to give up to three highest ranking reads (based on dupcount) to remove duplicates.In the next step V, D, J genes and CDR3 junction sequences were identified using IMGT-HighV-Quest online tool (23).A processing workflow implemented as an in-house Java program together with a custom MySQL database was used to further extract the sequences.In brief, only productive sequences with dupcount > 100 were collapsed based on identical V gene, J gene and CDR3 (nucleotide level).These sequences were then refiltered, so that only T cells with at least one TCRα and TCRβ chain and dual TCRα or TCRβ chains (but not both dual TCRα and TCRβ, maximum 3 chains) were considered for downstream analysis.We then categorized T cells as a clonotype if they have identical V and J gene (subgroup level), together with identical CDR3 region allowing for one nucleotide mismatch.This data was then used for further analysis of clonal expansion and V-gene usage.The paired TCRα and TCRβ sequencing of single T cells isolated from TCLs of CeD patients CD364, CD373, CD412 and CD436 used in V-gene usage analysis were generated in the course of another project for tracking clonotypes (7).

Statistical analysis
The Morisita-Horn index for analyzing similarity in TRAV and TRBV usage across TCRs specific for the four immunodominant epitopes was generated in R (3.3.2) with the use of the Vegan (2.4-2) package.We calculated pairwise similarities using sim.tablefunction in Vegan package, setting 2 as the order of diversity measure (q) to generate the Morisita-Horn indices.