Novel HLA-B27-restricted Epitopes from Chlamydia trachomatis Generated upon Endogenous Processing of Bacterial Proteins Suggest a Role of Molecular Mimicry in Reactive Arthritis*

Background: Reactive arthritis is an HLA-B27-associated disease triggered by Chlamydia trachomatis. Results: Three chlamydial peptides endogenously presented by HLA-B27 were identified. All were homologous to human-derived sequences, and one showed conformational similarity to a self-derived HLA-B27 ligand. Conclusion: Molecular mimicry between chlamydial and self-derived HLA-B27 ligands is not uncommon. Significance: Molecular mimicry may contribute to the pathology of reactive arthritis. Reactive arthritis (ReA) is an HLA-B27-associated spondyloarthropathy that is triggered by diverse bacteria, including Chlamydia trachomatis, a frequent intracellular parasite. HLA-B27-restricted T-cell responses are elicited against this bacterium in ReA patients, but their pathogenetic significance, autoimmune potential, and relevant epitopes are unknown. High resolution and sensitivity mass spectrometry was used to identify HLA-B27 ligands endogenously processed and presented by HLA-B27 from three chlamydial proteins for which T-cell epitopes were predicted. Fusion protein constructs of ClpC, Na+-translocating NADH-quinone reductase subunit A, and DNA primase were expressed in HLA-B27+ cells, and their HLA-B27-bound peptidomes were searched for endogenous bacterial ligands. A non-predicted peptide, distinct from the predicted T-cell epitope, was identified from ClpC. A peptide recognized by T-cells in vitro, NQRA(330–338), was detected from the reductase subunit. This is the second HLA-B27-restricted T-cell epitope from C. trachomatis with relevance in ReA demonstrated to be processed and presented in live cells. A novel peptide from the DNA primase, DNAP(211–223), was also found. This was a larger variant of a known epitope and was highly homologous to a self-derived natural ligand of HLA-B27. All three bacterial peptides showed high homology with human sequences containing the binding motif of HLA-B27. Molecular dynamics simulations further showed a striking conformational similarity between DNAP(211–223) and its homologous and much more flexible human-derived HLA-B27 ligand. The results suggest that molecular mimicry between HLA-B27-restricted bacterial and self-derived epitopes is frequent and may play a role in ReA.

MHC class I (MHC-I) molecules present endogenous peptides derived from self-proteins or intracellular pathogens at the cell surface for recognition by cytotoxic T lymphocytes (CTL). 6 HLA-B27, an allotype that is present worldwide, shows one of the strongest associations between MHC-I and a human disease (1)(2)(3). This association concerns a group of inflammatory rheumatic diseases termed spondyloarthropathies, which include ankylosing spondylitis (AS), where this allele occurs in about 90% of patients, and reactive arthritis (ReA), where the prevalence of HLA-B27 is less well defined but probably around 30 -50% (4). This latter disorder is triggered by various Gramnegative bacteria (5). Although it is frequently a self-limited disease, ReA evolves sometimes toward AS, particularly among HLA-B27 ϩ individuals. In contrast to AS, where HLA-B27 is probably a true pathogenetic factor, epidemiologic and other studies suggest that in ReA, it may influence the severity of clinical manifestations rather than being a truly causative allele (4,6,7). guarantee that this peptide is the actual immunogenic epitope in vivo.
The direct biochemical identification of endogenous chlamydial T-cell epitopes from infected cells has been accomplished only in the mouse system (35,36). It is hardly feasible in humans, due to the very low amounts of bacterial epitopes on infected cells, the difficulties associated with working with large amounts of Chlamydia-infected human cells, and, especially, the down-regulation of MHC-I expression and induction of apoptosis by C. trachomatis (19,37). Thus, we developed an alternative strategy involving the stable expression of chlamydial fusion proteins on HLA-B27 ϩ human cells. Endogenously processed chlamydial peptides, including a predicted T-cell epitope, were identified by comparing the HLA-B27-bound peptidomes from transfected and untransfected cells. These studies (38,39) were based on comparative MALDI-TOF MS and concerned three chlamydial proteins containing sequences highly homologous to known human-derived HLA-B27 ligands or from which synthetic peptides were recognized by CTL from ReA patients: DNA primase (DNAP) (CT794), Na ϩ -translocating NADH-quinone reductase subunit A (NQRA) (CT634), and pyrroloquinoline-quinone synthase-like protein (PqqC) (CT610).
In two different studies, based on a predictive search for HLA-B27-restricted chlamydial ligands in ReA patients (32,33), a sequence from ClpC protein, spanning residues 7-15, was recognized as a synthetic peptide by CD8 ϩ T-cells from multiple individuals, suggesting that this epitope could be immunodominant. Here we used MS techniques of high sensitivity and accuracy to investigate the endogenous processing and presentation of this and other HLA-B27-restricted peptides from ClpC and other chlamydial proteins. Molecular dynamics simulations were also carried out to analyze the relationship between chlamydial and homologous human-derived B27 ligands at the conformational level.
Flow Cytometry-The C1R transfectants were analyzed by measuring their EGFP-associated fluorescence. Briefly, 1 ϫ 10 6 cells were washed twice with 200 l of PBS and centrifuged at 1500 rpm for 5 min. The detection was carried out in a flow cytometer FACSCalibur (BD Biosciences). All data were acquired using CellQuest TM Pro version 4.0.2 software (BD Biosciences) and analyzed using FlowJo version 7.5 (Tree Star, Inc.).
Isolation of HLA-B27-bound Peptides-B*27:05-bound peptides were isolated from about 1 ϫ 10 10 cells or, for some analyses, 1 ϫ 10 9 C1R-B * 27:05 cells, as described previously (42). Briefly, cells were lysed in the presence of a mixture of protease inhibitors (Complete, Roche Applied Science). The soluble fraction was subjected to affinity chromatography using the W6/32 mAb (IgG2a; specific for a monomorphic HLA class I determinant) (43). HLA-B27-bound peptides were eluted with 0.1% aqueous TFA at room temperature, filtered through Centricon 3 devices (Amicon, Beverly, MA), concentrated, and either used as a peptide pool or subjected to reverse phase HPLC fractionation at a flow rate of 100 l/min, as described previously (44). Fractions of 50 l were collected and stored at Ϫ20°C until use.
Synthetic Peptides-These were obtained using standard N-(9 fluorenyl)methoxycarbonyl chemistry and purified by HPLC. The correct molecular weight of purified peptides was verified by MALDI-TOF MS.
MALDI-TOF MS-HPLC fractions were analyzed using a MALDI-TOF/TOF mass spectrometer (4800 Proteomics Analyzer, Applied Biosystems, Foster City, CA) as described previously (38) and processed using the Data Explorer software version 4.9 (Applied Biosystems).
Electrospray-LTQ-Orbitrap MS/MS-Peptide mixtures were desalted and concentrated with Micro-Tip reverse-phase columns and analyzed by LC-MS/MS using an Orbitrap XL mass spectrometer (Thermo Fisher Scientific) fitted with a capillary HPLC (Eksigent, Dublin, CA) as described previously (45), with minor modifications. Briefly, the peptides were eluted at flow rates of 0.25 l/min, with linear gradients of 7-40% acetonitrile in 0.1% formic acid, for 90 min, followed by 17 min at 95% acetonitrile in 0.1% formic acid. In some cases, the same gradient was used during 214 min, with a final isocratic elution for 29 min. The spectra were collected in the Orbitrap mass analyzer using full ion scan mode over the mass-to-charge (m/z) range 400 -2000, which was set to 60,000 resolutions. The most intense seven masses from each full mass spectrum, with single, double, and triple charge states, were selected for fragmentation by collision-induced disintegration in the linear ion trap.
Electrospray-LTQ-Velos MS/MS-Particular peptides were searched in 10 l of individual HPLC fractions by MS/MS in a dual mode, using selected multiple ion monitoring and dynamic exclusion mode in an LTQ-Velos instrument. Briefly, each particular fraction was dried down and resuspended in 9 l of 0.1% formic acid and analyzed in an Agilent 1100 system coupled to a linear ion trap LTQ-Velos mass spectrometer (Thermo Fisher Scientific). The peptides were separated by reverse phase chromatography using a 0.18 ϫ 150-mm Bio-Basic C18 RP column (Thermo Fisher Scientific) and eluted using an 80-min gradient from 5 to 40% solvent B (solvent A, 0.1% formic acid in water; solvent B, 0.1% formic acid, 80% acetonitrile in water) at 1.8 l/min. Peptides were detected in selected multiple ion monitoring mode at single, double, and triple charged states. In parallel to the selected multiple ion monitoring mode, a full ion scan over the m/z range 400 -2000 (1-s scans) was also performed, followed by data-dependent MS/MS scans, using an isolation width of 2 m/z units and normalized collision energy of 35%, and dynamic exclusion was applied for 30 s. Alternatively, 10-l aliquots of various consecutive HPLC fractions were pooled together and analyzed in the same way. The synthetic peptides were detected using only the selected multiple ion monitoring mode as above, except that a 35-min elution gradient was used.
Database Searches-The Mascot server 2.2 (Matrix Science Inc., Boston, MA) (46) was used as the main search engine. The search parameters were 0.5 Da mass tolerance for both precursor and fragment ions for MS/MS spectra from LTQ-Velos and 0.01 and 0.5 Da for precursor and fragment ions, respectively, for data from LTQ-Orbitrap. Met oxidation and Asn and Gln deamidation were selected as variable modifications. A small sequence database consisting of the chlamydial ClpC (Swiss-Prot accession B0B7K2), DNAP (B0B920), and NQRA (O84639) sequences as well as HLA-B27 (P03989), HLA-B35 (P30685), HLA-C04 (P30504), and EGFP (GenBank TM accession AAB02576.1) was used for the specific search of chlamydial peptides. In addition, all raw files were run against the human subset of the Uniprot database (release 57.6, 07/2009, with 20,331 entries), using the same parameters described above. Those sequences showing the highest scores in these preliminary searches were analyzed manually and validated by comparison with the experimental MS/MS spectrum of the corresponding synthetic peptide.
The search for homology between chlamydial peptides and human proteins was carried out using the UniProtKB/Swiss-Prot database (release 07/2012, with 20,231 entries) and the BLASTP 2.2.26ϩ software.
Proteasome Cleavage Predictions-Proteasome/immunoproteasome cleavage was predicted with previously described algorithms (47) available on the Proteasome Cleavage Prediction Server.
Homology Modeling-Three-dimensional models for the complexes between B*27:05/␤ 2 m and DNAP(211-221), DNAP(211-223), or B27(309 -320) were built by homology modeling. A total of 23 x-ray structures of HLA-B27⅐peptide complexes were aligned using the MAFFT software (48). Because all of the x-ray complexes contained bound 9-mers, the alignments of these peptides with the longer ones in our study was done by introducing gaps at internal peptide positions. The four N-terminal and two C-terminal positions on each peptide were constrained, whereas certain flexibility was allowed for their central parts. B*27:05 in complex with the pVIPR(400 -408) peptide in its canonical conformation (Protein Data Bank code 1OGT) (49) was finally selected as template, due to its high resolution (1.47 Å), and the alignment was subjected to homology modeling using the MODELLER program.
Setup of the Systems and Molecular Dynamics (MD) Simulations-For each HLA-B27⅐peptide complex, the setup entailed the following steps: (a) adding missing heavy and hydrogen atoms (50) to assign atom types and charges according to AMBER ff10 force field (51) and to determine the protonation state of ionizable residues at pH 7; (b) employing the tleap module from the AmberTools package (52) to immerse each system within a 10-Å box of TIP3P (53) explicit water molecules and to add Na ϩ counterions; (c) energy-minimizing the positions of water molecules and ions using the conjugated gradient method for 3000 steps while the atomic coordinates in the complexes were kept constrained, followed by equilibration at 298 K for 10 ps, maintaining the constraints; (d) transforming the constraints into progressively lower restraints and energy-minimizing the whole complexes, including the water molecules and the ions, as above.
MD simulations were carried out starting from the energyminimized structures. All calculations were performed with the NAMD version 2.8 program (54) using constant temperature (298 K) and pressure (1 atm). Short and long range forces were calculated every one and two time steps, respectively (each time step ϭ 2.0 fs), constraining the covalent bonds involving hydrogen atoms to their equilibrium values. Long range electrostatic interactions were accounted for using the particle mesh Ewald approach (55). The systems were heated up to 298 K and then equilibrated at this temperature for 200 ps. The equilibration was performed under harmonic restraint conditions on all of the heavy atoms. These restraints were gradually reduced until they were almost removed. Finally, these equilibrated structures were further simulated for an additional 50 ps with a minimal restraint. These were the starting points for a 30-ns MD production period during which the system coordinates were collected every 2 ps for further analysis.
Analysis of MD Trajectories-The stability of a given complex was evaluated by calculating the root mean square deviation (RMSD) of the C␣ atoms along the trajectories, using as reference their starting structures. Additionally, the root mean square fluctuation (RMSF) of each residue, relative to the corresponding average value, was calculated once each snapshot had been fitted to its initial structure. Further analysis was carried out by clustering the sampled conformational space during the trajectory production period (last 10 ns), using the ptraj module from the AmberTools package, the snapshots sampled as described above, and the average linkage algorithm based on the peptide backbone atoms. Adaptive Poisson-Boltzmann Solver (56,57) was used to perform the Poisson-Boltzmann electrostatic calculations for the most representative structures in each cluster. Dielectric constants were set to 4 and 80 for protein and solvent, respectively. Other parameters were set as default. The free energy of binding between each peptide and the B*27:05 molecule was calculated by the MM-ISMSA approach (58). We also calculated the pairwise decomposition of the free energy of binding following the scheme developed in MM-ISMSA to determine the main residues responsible for the interaction of the peptides with B*27:05. Mean and S.D. for the free energy of binding was calculated for the MD trajectories fit to a normal frequency distribution using R (59). Contacts between residues were analyzed following the MM-ISMSA methodology.

Expression of Chlamydial ClpC Fusion Proteins-ClpC is an
ATP-dependent protein-unfolding subunit of the bacterial ClpCP protease complex (60,61). In C. trachomatis, it has 854 amino acid residues and binds ATP through two nucleotidebinding domains, AAAϩ (Fig. 1A). EGFP-ClpC fusion proteins were expressed in C1R-B*27:05 cells in order to detect endogenously processed HLA-B27 ligands from this protein, including a predicted T-cell epitope, ClpC(7-15). Our initial attempts to express the whole ClpC protein using full-length cDNA failed to generate stable C1R transfectants. To avoid functional interference of the ClpC protein in human cells, two fusion protein constructs, ClpC(1-570) and ClpC(1-512), with partial or total deletions of the C-terminal AAAϩ domain, were made in which residues 1-570 or 1-512, respectively, were fused at the C-terminal end of EGFP (Fig. 1A). Stable transfectants in C1R-B*27:05 cells were obtained for both constructs, whose expression levels and correct size were determined by flow cytometry (Fig. 1B) and Western blot (Fig. 1C), respectively. The ClpC(1-512) transfectant in C1R-B*27:05 was used for further experiments, due to its higher expression compared with ClpC(1-570).

One ClpC-derived Ligand Distinct from the Predicted T-cell Epitope Is Endogenously Presented by HLA-B*27:05 on C1R
Cells-A first approach to search for endogenously processed ClpC-derived HLA-B27 ligands was the comparative analysis of HLA-B27-bound peptides from untransfected C1R-B*27:05 cells and the ClpC(1-512) transfectant, based on identity of chromatographic retention time (RT) and molecular weight, through systematic comparison of the MALDI-TOF MS spectra from correlated HPLC fractions. Although this strategy was successful in previous studies with other fusion proteins (38,39), it failed to identify any ClpC-derived peptides. Thus, two further approaches were undertaken (Fig. 1D). The first one involved high throughput sequencing, using LTQ-Orbitrap MS/MS, performed on the unfractionated B27-bound peptide pool from ClpC(1-512)-transfected C1R-B*27:05 cells.
The second one involved a targeted search for specific candidates in the fractionated B27-bound peptide pool performed on HPLC fractions at the RT Ϯ 3 min of each of the corresponding synthetic peptides. The relevant HPLC fractions, either individually or pooled together, were subjected to MS/MS fragmentation of all ions corresponding to the m/z ratios of the candidate peptide, using a LTQ-Velos mass spectrometer.
The MS/MS spectra from the unfractionated B27 peptidome from the ClpC(1-512) transfectant obtained in the LTQ-Orbitrap were searched against a small database including ClpC and a few other chlamydial proteins. Two putatively significant matches with sequences containing the canonic B27 binding motif R2 from ClpC were obtained. Manual inspection of the corresponding MS/MS spectra showed a good match with the theoretical fragmentation of only one of these sequences, SRLDPVIGR, spanning ClpC residues 203-211 ( Fig. 2A). A search against the human proteome database did not show a match of this MS/MS spectrum with any human peptide. SRLDPVIGR did not match any human sequence upon BLAST analysis, confirming the bacterial origin of this peptide.
We next determined whether this peptide was just overlooked in our previous MALDI-TOF comparison or hidden by a co-eluting human HLA-B27 ligand. For this purpose, the RT of the synthetic peptide in the same chromatographic condi- Candidates sequences Comparison with the synthetic peptide (7) Manual and assisted interpretation (6) Validation Specific search at multiple charge states LTQ-Velos (Individual fractions or minipool of fractions at RT ± 3 min) HLA-B27-bound peptides: isolation HPLC and MALDI-TOF MS analysis (3) RT determination for each target peptide  (1). Alternatively, a specific search was performed by determining the RT of a target synthetic peptide (2) and analyzing the corresponding individual fractions, or a minipool of neighbor fractions around the RT of the synthetic peptide, from an HPLC-fractionated B27-bound peptide pool (3) and looking for the specific ion peaks at various charge states in an LTQ-Velos mass spectrometer (4). MS/MS spectra were submitted to automatic interpretation using the Mascot software (5). Each candidate sequence was revised manually and assisted by the MS-product tool (6). Final confirmation was done by comparing the MS/MS spectrum of the assigned peptide with that of the synthetic peptide (7).    Because the Orbitrap-based sequencing described above failed to detect the predicted T-cell epitope ClpC(7-15), NRAKQVIKL, an alternative approach was used for the specific search of this and the related peptide ClpC(7-17), NRAKQ-VIKLAK, which also has the B*27:05 binding motif, in the HPLC-fractionated HLA-B27-bound peptide pool from the ClpC(1-512) transfectant. Both peptides were synthesized and used for a targeted search (Fig. 1D), monitoring the m/z ratios corresponding to [M ϩ 2H] 2ϩ and [M ϩ 3H] 3ϩ ions of both peptides. These analyses failed to show any reliable fragmentation compatible with ClpC (7)(8)(9)(10)(11)(12)(13)(14)(15) or ClpC (7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17).
Novel Chlamydial Peptides from Other Proteins Processed and Presented by HLA-B27 in Live Cells-Several chlamydial peptides endogenously processed and presented by HLA-B27 were identified in previous studies from our laboratory (38,39) by comparative MALDI-TOF MS of HPLC-fractionated B27bound peptide pools from C1R-B*27:05 transfectants expressing chlamydial NQRA, PqqC, or DNAP fusion protein constructs (Table 1). Due to the limitations of this approach, revealed by our results on ClpC, a search for novel peptides from NQRA and DNAP was undertaken, using more sensitive MS techniques.
NQRA-The NQRA(330 -338) peptide, MRDHTITLL, was recognized in vitro, as a synthetic peptide, by CD8 ϩ T-cells from a ReA patient (32), but it was not found in C1R-B*27:05 cells expressing the EGFP-NQRA(1-465) fusion protein in a MALDI-TOF-based study (39). Thus, the most intense ions in the full MS spectrum of the pooled fractions corresponding to the RT Ϯ 3 min of the synthetic peptide in the fractionated HLA-B27-bound peptide pool from the EGFP-NQRA(1-465) transfectant were subjected to MS/MS fragmentation. The MS/MS spectrum of one of the main ion peaks in the full MS scan, with m/z 558.33, was compatible with the [M ϩ 2H] 2ϩ species of the oxidized form of MRDHTITLL. Its correct assignment was confirmed by comparison with the MS/MS spectrum of the corresponding synthetic peptide in its oxidized form (Fig. 3). This result demonstrates the endogenous processing and presentation by HLA-B27 of the predicted chlamydial epitope NQRA(330 -338) in NQRA transfectant cells. This is the second HLA-B27-restricted T-cell epitope with demonstrated relevance in Chlamydiainfected ReA patients that has been shown to be generated in live cells.
DNAP-The unfractionated HLA-B27-bound peptide pool from C1R-B*27:05 transfected with the EGFP-DNAP(90 -450) fusion protein (38) was subjected to MS/MS analysis in an LTQ-Orbitrap mass spectrometer and searched against a small database including the chlamydial DNAP fusion protein sequence. A parental ion of m/z 508.62, compatible with DNAP(211-223) (RRFKEGGRGGKYI) was identified (Fig. 4A). This peptide was two residues longer than one previously found from this protein, DNAP(211-221) ( Table 1). Both sequences show high homology with a natural ligand of HLA-B27, arising from the endogenous processing of the HLA-B27 heavy chain, B27(309 -320) (RRKSSGGKGGSY) (62). To confirm the tentative assignment from the Orbitrap analysis, a targeted search for this peptide (  (Fig. 4B), whose identity was confirmed by comparison with the MS/MS spectra of the synthetic peptide.
High Homology between the ClpC and NQRA-derived HLA-B27 Ligands and Human Sequences-To explore the possible molecular mimicry between the B27-restricted peptides from C. trachomatis found in this study and putative self-derived HLA-B27 ligands, we looked for human sequences showing high homology to ClpC(203-211) and NQRA(330 -338). The search was performed against the human proteome, looking for sequences containing Ͼ50% amino acid identity with the bacterial peptides and the main binding motif of HLA-B27 ligands, R2. Only human sequences with residues present among known HLA-B27 ligands (63,64) with a frequency of Ͼ1% at the anchor P1, P3, and P⍀ positions were considered. Multiple human sequences homologous to the ClpC-and NQRA-derived peptides were found ( Table  2). Most of the sequences showed predictive scores compatible with proteasome/immunoproteasome cleavage at their C-terminal residue (Ͼ0.5).
MD Simulation of Chlamydial DNAP and Homologous Human-derived HLA-B27 Ligands-To explore the similarity of DNAP(211-221) and DNAP(211-223) with B27(309 -320) at the three-dimensional level, comparative MD simulation of their interaction in complex with B*27:05 was carried out. The initial, energy-minimized, three-dimensional structures of the complexes involving the three peptides, all built by homology modeling, and pVIPR(400 -408) in its canonical conformation were subjected to MD simulations for 30 ns. After this time, the stability of the trajectories was analyzed. Both the mean C␣ RMSD and the mean RMSF for the B*27:05 heavy chain and ␤ 2 m were similar among the three complexes (Fig. 5, A and B). In contrast, the mean RMSD and RMSF values for the peptides were more variable, spreading from 0.58 to 2.25 Å and from  6, A and B). Large RMSF values (above 3.0 Å) were observed for certain residues (Fig. 6B), such as Arg-8 in DNAP(211-221) and Gly-6, Gly-7, and Lys-8 in B27(309 -320). The very low RMSD fluctuation of DNAP(211-223) after the first 5-10 ns of MD simulation and the smaller RMSF values, relative to DNAP(211-221) and B27(309 -320), suggest a less flexible structure of the former peptide. Clustering Analysis Reveals Distinct Peptide Flexibility and Conformations-A total of 5000 structures sampled during the last 10 ns of the MD simulation were subdivided in up to five clusters on the basis of similarity (RSMD) in the peptide backbone. Two predominant clusters were found for DNAP(211-221), one for DNAP(211-223), three for B27(309 -320), and one for the x-ray template ( Table 3). The distinct flexibility of the three peptides revealed by this analysis was further apparent upon considering the intracluster RMSD variability. This was calculated as the distance to the centroid, which is the average distance of all members of a cluster to its geometrical center. This parameter reflects the dispersion of data inside a given cluster. Smaller (0.43), intermediate (0.54), and larger values (0.7) were found for the major clusters of DNAP (211-223),         configuration preferences of the peptides in their bound states (Fig. 6D). For B27(309 -320), rep1 and rep2 showed similar conformations and small differences in their molecular surface, but rep4 was significantly different. For DNAP(211-221), the representative conformers of its two main clusters were very similar and were different from those of B27(309 -320). In contrast, the only major cluster in DNAP(211-223) showed a striking similarity to B27(309 -320), looking like an intermediate form of rep2 and rep4 of this peptide. DNAP(211-223) also showed a surface charge distribution with similarities to both rep2 and rep4 of B27(309 -320) (Fig. 6E).
Binding Energy-MM-ISMSA was used to estimate the total free energy of binding of the peptides in the binding groove of B*27:05 and the contribution of each peptide residue to the total free energy of binding. The N-and C-terminal residues each contributed ϳ20 kcal/mol to the total binding of each peptide. Residue 2 showed the highest contribution, ϳ25 kcal/ mol, whereas the central regions of the peptides showed greater variation and a smaller contribution (Fig. 5C). These results are in full agreement with the known canonical interactions governing binding of MHC-I ligands.

DISCUSSION
Two issues were addressed in this study: first, the endogenous processing and presentation of predicted T-cell epitopes, recognized as synthetic peptides by CTL from Chlamydia-infected ReA patients, and second, the structural similarity between chlamydial and human-derived HLA-B27 ligands. Our approach was the direct identification of endogenously processed chlamydial peptides using high sensitivity and accuracy MS. Although, ideally, this search should be performed on Chlamydia-infected cells, this approach is virtually unfeasible in humans, due to induction of MHC-I down-regulation and apoptosis (38). Some chlamydial proteins are injected into the cytosol through the type III secretion system (65)(66)(67)(68). However, many others reach cytosolic cross-presentation pathways (69,70) after uptake of bacterial debris from infected cells undergoing apoptosis and are subjected to proteasomal degradation, similarly to endogenous proteins. Thus, the endogenous processing of chlamydial fusion proteins is likely to mimic that in infected cells to a large degree, as confirmed by the direct identification of chlamydial T-cell epitopes using fusion proteins in this and a previous study (39). However, proteasome-independent pathways might also generate chlamydial MHC-I ligands after transfer of bacterial components following the fusion of inclusion-derived vesicles with the endoplasmic reticulum (71) and perhaps also through non-cytosolic cross-presentation pathways. Thus, some chlamydial antigens may not be revealed with our approach.
Although studies based on MALDI-TOF MS allowed us to identify several HLA-B27 ligands from C. trachomatis, the limitations of this approach justified a more in depth search using electrospray-based MS techniques to look for novel chlamydial epitopes. Despite the technical improvements, the direct iden- tification of immunologically relevant bacterial peptides by biochemical methods is less sensitive than CTL because these can recognize minute antigen amounts, down to a few copies, at the cell surface (72). Although the relatively high expression of bacterial fusion proteins results in the generation of many more copies of chlamydial peptides than on infected cells, partially compensating for the lower sensitivity of biochemical analyses, the different thresholds relative to CTL recognition must always be kept in mind.
Our study focused on three chlamydial proteins. For two of them, ClpC and NQRA, HLA-B27-restricted T-cell epitopes had been predicted (32,33). For the third one, DNAP, an endogenous peptide, DNAP(211-221), with high homology to a natural human-derived B27 ligand, was previously reported (38). Both the transcriptional profile (73) and the proteomic characterization of the Chlamydia life cycle (74) indicate that ClpC is expressed in the infectious elementary body and, at a higher level, in the replicative but non-infectious reticulate body and is up-regulated by IFN-␥ (75). The presence of ClpC in both developmental stages and its up-regulation in an inflammatory context is compatible with the possibility that HLA-B27-restricted T-cells, directed against epitopes from this protein, may be relevant in controlling both the bacterial infection and the development of ReA. Detection of NQRA in the itations are clear in the previous failure to predict some chlamydial B27 ligands that are endogenously processed and presented in live cells, including ClpC(203-211) identified in this study. Moreover, because monoclonal T-cells can recognize many distinct peptides (34), T-cell recognition of a synthetic peptide in vitro does not necessarily identify the natural epitope. Conversely, the identification of chlamydial peptides processed and presented by HLA-B27 in live cells does not indicate their immunological relevance in the absence of their positive identification by T-cells. Despite their limitations, prediction algorithms are useful for detecting epitopes generated in vivo because they help in focusing MS-based search strategies toward specific peptides in complex pools, as demonstrated by our previous identification of an endogenous HLA-B27-restricted chlamydial T-cell epitope (39). Another predicted epitope, from NQRA, was found in the present study. Thus, NQRA(330 -338) is the second known chlamydial T-cell antigen processed and presented in live cells by HLA-B27 and recognized by specific CTL from ReA patients. This demonstrates the similarity of epitope processing between fusion proteins and infected cells.
Our failure to detect the predicted T-cell epitope ClpC(7-15), despite an intensive search with highly sensitive techniques, must be interpreted with caution. We cannot rule out that this peptide might be present in our cell lines in very low amounts that challenge detection by MS but are still sufficient for T-cell recognition. With this possibility in mind, our results suggest that this peptide may be produced with low efficiency, if at all, in vivo.
C. trachomatis is a large organism and is potentially the source of many HLA-B27-restricted ligands. The use of fusion proteins necessarily limits our analysis to a few epitopes. However, the endogenous generation of HLA-B27 ligands from each bacterial protein tested suggests that HLA-B27-restricted T-cell responses in ReA patients may be directed against multiple chlamydial antigens. That all of the reported peptides showed significant homology with human sequences suggests that autoimmune cross-reaction of Chlamydia-specific T-cells with self-derived HLA-B27 epitopes through molecular mimicry might not be uncommon.
The chlamydial DNAP shows a particularly interesting example of molecular mimicry between bacterial and self-derived HLA-B27 ligands. HLA-B27 presents an 11-mer from this protein, DNAP(211-221), with high homology to the humanderived HLA-B27 ligand B27(309 -320), which is one residue longer than the chlamydial peptide (38,62). The finding now of the C-terminally extended variant DNAP(211-223), whose proteasomal generation was predicted in a previous study (62), increased the probability of molecular mimicry between peptides from DNAP and the human-derived ligand. MD simulations suggest that DNAP(211-221) and DNAP(211-223) adopt distinct conformations. Both peptides showed limited flexibility and a peptide-specific predominant conformation. In contrast, B27(309 -320) was significantly more flexible. This is in agreement with x-ray data showing a single defined conformation of DNAP(211-221) and a diffuse electron density corresponding to the central region of B27(309 -320) in complex with B*27:05. 7 The limited flexibility of the two chlamydial peptides, especially DNAP(211-223), observed in our MD simulations was apparently determined by intrapeptide hydrogen bonds established within their central regions, which are more frequent among long peptides, and by peptide-specific interactions of their central regions with HLA-B27 residues.
The higher flexibility of the human-derived peptide is likely to provide a wider spectrum of antigenically distinct conformations. The striking similarity of the conformation and surface charge distribution of DNAP(211-223) with some of the main conformational clusters of B27(309 -320) could favor T-cell cross-reaction between both peptides. A peptide bound in a flexible and variable conformation in its middle part may be amenable to recognition by more T-cell clones, with preference for single conformations, than a peptide bound with lower flexibility. For instance, T-cell-mediated self-reactivity has been related to peptide antigens bound to HLA-B27 in dual conformation (76,77). The antigenic similarity between the DNAPderived peptides and the homologous self-derived B27 ligand must be confirmed in functional assays with peptide-specific T-cells. Although we recognize the importance of functional studies in this context, we were unable to perform them because it was extremely difficult to gain access to HLA-B27 ϩ patients with Chlamydia-induced ReA, a disease becoming increasingly rare or not unambiguously diagnosed (4) in Western countries. Attempts to stimulate peptide-specific, HLA-B27-restricted, CTL in vitro from a few individuals were unsuccessful. Due to the difficulties inherent to raising peptidespecific CTL in vitro, even from infected individuals, these studies must be performed with a sufficient number of patients, which was unfeasible because they were not available. In the absence of formal confirmation with T-cells, both the sequence homology and the predicted conformational features of DNAP(211-223) and B27(309 -320) suggest a mechanism for increasing T-cell cross-reaction between endogenous chlamydial and self-derived HLA-B27 ligands through 7 B. Loll, B. Uchanska-Ziegler, and A. Ziegler, unpublished observations. presentation of related peptides of distinct length and conformation, homologous to self-peptides with high flexibility in their bound state.
In conclusion, the high accuracy and sensitivity of current MS technologies brought about a major improvement in the detection of naturally processed HLA-B27 ligands from C. trachomatis, allowing us to detect three novel peptides from distinct proteins, including the second known HLA-B27-restricted epitope recognized by T-cells from ReA patients. Both the homology of all of the reported peptides with human sequences carrying the binding motif of HLA-B27 and the finding of a peptide from DNAP with significant sequence and conformational similarity to a human-derived HLA-B27 ligand suggest that molecular mimicry between bacterial and self-derived HLA-B27 ligands may play a role in ReA. This mechanism could provide an autoimmune component that would exacerbate the proinflammatory role of HLA-B27, influencing disease severity and evolution toward chronicity.