Dual Molecular Mechanisms Govern Escape at Immunodominant HLA A2-Restricted HIV Epitope

Serial accumulation of mutations to fixation in the SLYNTVATL (SL9) immunodominant, HIV p17 Gag-derived, HLA A2-restricted cytotoxic T lymphocyte epitope produce the SLFNTIAVL triple mutant “ultimate” escape variant. These mutations in solvent-exposed residues are believed to interfere with TCR recognition, although confirmation has awaited structural verification. Here, we solved a TCR co-complex structure with SL9 and the triple escape mutant to determine the mechanism of immune escape in this eminent system. We show that, in contrast to prevailing hypotheses, the main TCR contact residue is 4N and the dominant mechanism of escape is not via lack of TCR engagement. Instead, mutation of solvent-exposed residues in the peptide destabilise the peptide–HLA and reduce peptide density at the cell surface. These results highlight the extraordinary lengths that HIV employs to evade detection by high-affinity TCRs with a broad peptide-binding footprint and necessitate re-evaluation of this exemplar model of HIV TCR escape.

inTrODUcTiOn CD8 + cytotoxic T lymphocytes (CTLs) form a critical component in immune control of HIV infection. However, the virus has adopted numerous strategies to evade detection by these important immune cells (1,2). Notably, the high rate and poor fidelity of replication produces a viral quasispecies to allow rapid selection of sequence variants that evade CTLs. This immune escape mutation is often associated with loss of viral fitness (3). The effects of immune escape from some HLA class I alleles, particularly HLAs B27 and B57 (3,4), can be significant and delay progression to AIDS. There has been considerable work on immune escape from HIV epitopes presented by the most frequent HLA worldwide, HLA A2 (5). The majority of these studies have focussed on the immunodominant response through the p17 HIV Gag-derived epitope SL9 (amino acids 77-85, sequence SLYNTVATL) (6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23). Responses to SL9 are uncommon in acute infection but are dominant during the long, asymptomatic, chronic infection period where over 70% of HLA A2 + HIV-infected individuals mount a strong response to this epitope (13,18,19). Immune escape from SL9-specific CTL during this period has been studied extensively and is known to follow well-described patterns (6,7). The leucine residues at positions 2 and 9 in SL9 are located within an α-helix (21) and are critical for trimerisation of the P17 Gag protein (24).
Mutations within these residues are believed to produce highly compromised virus (11,24). These residues also act as the principal anchors for securing SL9 into the HLA A2-binding groove (22) and ensure that the virus does not use the archetypical escape strategy of mutating primary MHC anchors to block epitope presentation at the infected cell surface (25)(26)(27). Instead, in the face of SL9-specific CTL pressure, HIV has been observed to undergo preferred mutation at positions 3, 6, and 8 in the epitope (6,7,9). The dominant mutations seen at these positions (Y3F, V6I, and T8V) correlate with a lower viral load suggesting that they result in a loss of viral fitness, a notion that is supported by the rapid reversion to the wild-type SL9 sequence in the absence of T-cell selection pressure (6,7). The available evidence suggests that HIV balances the combined pressures of immune escape vs. viral fitness at the SL9 epitope by successive fixation of mutations at positions 3, 6, and 8 in the subtype A, B, and D SL9 SLYNTVATL consensus sequence over a >10-year period (6,7). Indeed, McMichael and colleagues have described the triple mutant SLFNTIAVL sequence (3F6I8V; mutations in bold underlined text) as the "ultimate" escape mutant in SL9 (6).
Here we use structure and protein biophysics to deconstruct how HIV escapes from the well-studied (9-11) SL9-specific 868 TCR. We add to the existing data by providing the first ever TCR-pMHC structures in this exemplar model of how HIV evades engagement by host TCRs. We find that the 868 TCR interacts with HLA A2-SLYNTVATL (A2-peptide from here on in) with the expected binding topography but exhibits an unusual degree of structural rearrangement upon interaction with A2-SLYNTVATL. The 868 TCR also contacts all five central residues in SL9 (positions [3][4][5][6][7][8] and binds with an unprecedentedly high affinity for a natural TCR. Surprisingly, 4N is the major contact residue-not the 3Y, 6I, or 8T that mutate to precipitate immune escape. This enabled the generation of a high-resolution structure of the 868 TCR in complex with the SLFNTIAVL ultimate escape variant. These results challenge current thinking on the mechanism by which HIV escapes at SL9 and demonstrate the extraordinary tactics that the virus can employ to escape from CTL responses that can involve TCRs with a very broad footprint and a very high natural binding affinity.

MaTerials anD MeThODs construct Design
The 868 TCR α and β chains, HLA A2 heavy chain, and β2m chain were generated by PCR mutagenesis and cloning. All sequences were confirmed by automated DNA sequencing (Lark Technologies). TCR expression constructs were designed with a disulphide linked construct to produce the soluble domains (variable and constant) for both the α (residues 1-207) and β chains (residues 1-247) (28,29). The HLA A2 heavy chain (residues 1-248) (α1, α2, and α3 domains), tagged, or not tagged with a biotinylation sequence, and β2m (residues 1-100) were also cloned and used to make the pMHCI complexes. The TCR α and β chains, the HLA A2 α chain, and β2m sequences were inserted into separate pGMT7 expression plasmids under the control of the T7 promoter (30).

Protein expression, refolding, and Purification
Competent Rosetta DE3 E. coli cells were used to produce the TCR α and β chains, HLA A2 heavy chain, and β2m in the form of inclusion bodies using 0.5 mM IPTG to induce expression and protein were chemically refolded as described previously (31).

surface Plasmon resonance (sPr) experiments
Surface plasmon resonance equilibrium binding analysis was performed using a BIAcore T100™ equipped with a CM5 sensor chip as previously reported (32,33). HLA DR1, generated as in Ref. (34), was used as a negative control on flow cell 1. SPR kinetic analyses were carried out to determine the KD values for the TCR, at 25°C. For all kinetic experiments, approximately 300 RUs of pMHC was coupled to the CM5 sensor chip surface. The TCR was then injected at concentrations ranging from 10 times above and 10 times below the known KD of the interaction at 45  isothermal Titration calorimetry (iTc) Isothermal titration calorimetry experiments were performed using a Microcal VP-ITC (GE Healthcare) as previously described (35), with 30 µM pHLA-I in the calorimeter cell and 210 µM soluble 868 TCR in the syringe. Buffer conditions were 20 mM Hepes (pH 7.4) containing 150 mM NaCl and 20 injections of 2 µl volume each were performed. Results were processed and integrated with the Origin 6.0™ software distributed with the instrument. ITC experiments were performed in duplicate.

adhesion Frequency assay
We used an adhesion frequency assay to measure the 2D affinity of TCR-pMHC interactions at the cell membrane as previously described (36). Briefly, human T-cells transfected with the 868 TCR were mounted onto one micro-pipette and, on the other pipette, human red blood cells coated with pMHC by biotinstreptavidin coupling served as both a surrogate APC and an adhesion sensor for detecting the TCR-pMHC interaction. Site Unanticipated HIV Immune Escape Mechanism Frontiers in Immunology | www.frontiersin.org November 2017 | Volume 8 | Article 1503 densities of TCR and pMHC were measured by flow cytometry as previously described (37). All assays were performed using at least 5 cell pairs, and calculated as an average of 100 cell-cell contacts.  (38). All crystals were soaked in 30% ethylene glycol before cryo-cooling. All crystallisation screens and optimisation experiments were completed using an Art-Robbins Phoenix dispensing robot (Alpha Biotech Ltd, UK). Data were collected at 100 K at the Diamond Light Source, Oxfordshire. All datasets were collected at a wavelength of ~0.98Å using an ADSC Q315 CCD or PILATUS Pixel detectors. Reflection intensities were estimated with the XIA2 package (39) and the data were scaled, reduced and analysed with SCALA and the CCP4 package (40). Structures were solved with molecular replacement using PHASER (41). Sequences were adjusted with COOT (42) and the models refined with REFMAC5. Graphical representations were prepared with PYMOL (43). The reflection data and final model coordinates were deposited with the PDB database (868, PDB: 5NMD; 868-A2-SLYNTVATL, PDB: 5NME; 868-A2-SLYNTIATL, PDB: 5NMF; 868-A2-SLFNTIAVL, PDB: 5NMG; A2-SLYNTIATL, PDB: 5NMH; A2-SLFNTIAVL, PDB: 5NMK).
Thermal stability cD analysis of hla a2 complexes Thermal stability of HLA A2/β2m/peptide complexes was determined by circular dichroism spectroscopy following the change of ellipticities Θ at 218 nm using an Aviv 215 instrument (Aviv Biomedical Inc., Lakewood, NJ, USA). Proteins were dissolved in 137 mM NaCl, 3 mM KCl, 8 mM Na2HPO4, and 1 mM KH2PO4, pH 7.4, at concentrations of ~3 μM as determined spectroscopically using calculated extinction coefficients (44). Melting curves were recorded from 4°C up to a maximum temperature when protein aggregation was observed using a heating rate of ~0.5°C/min. Melting curves were analysed assuming a two-state trimer-to-monomer transition from the native (N) Tissue culture T2 cells were maintained as suspension cells in R10 (RPMI 1640 media, 10% FBS, penicillin, streptomycin, and l-glutamine, all Life Technologies, Paisley, UK) and passaged weekly or when required. Post enrichment, primary CD8 + T-cells were routinely expanded with allogeneic feeder PBMCs and phytohaemagluttinin (PHA) in T-cell media as previously described (47). 868 TCR expression was checked prior to assay using anti-rCD2 antibody and A2-SLYNTVATL tetramer, with the protein kinase inhibitor Dasatinib (48) (Axon Medchem, Reston, VA, USA), and unconjugated anti-PE antibody (PE001, BioLegend, London, UK) used to enhance tetramer staining, as previously described (47,49).

T-cell activation assays
For peptide pulsing assays, T2 cells were incubated with peptide for 1 h in 0.5 ml of AIM-V serum free media (ThermoFisher Scientific) in 5 ml FACS tubes at 37°C and 5% CO2, followed by 4 washes with 4 ml of RPMI 1640 media. In assays using primary CD4 + T-cells as targets, these cells were purified from PBMC using magnetic microbead separation (Miltenyi Biotech). Peptides were synthesised to >95% purity, stored as 20 mM DMSO stocks at −80°C and 1 mM RPMI dilutions made on the day of assay. DMSO controls were treated the same as peptide pulsed. Post pulsing, T2 cells or CD4 + T-cells were incubated in 4.5 ml of AIM-V media at 37°C and 5% CO2 for 3-24 h or used immediately (0 h), such that cells were ready at the same time. 868 TCR transduced and non-transduced T-cells (30,000) were rested overnight in R5 (as for R10 but with 5% fetal bovine serum) then co-incubated in 96 U-well plates with T2 or CD4 + T-cells (60,000)  T2 Peptide-hla-a2 stability assays For conventional peptide-binding assays, T2 cells were pulsed for 22 h with 10 −5 M peptide in 0.2 ml of AIM-V media (ThermoFisher Scientific) in 5 ml FACS tubes at 37°C and 5% CO2, washed then stained with anti-HLA-A2 antibody (BB7.2 BioRad, Hercules, CA, USA) for 10 min at RT, washed with PBS, and fixed with 2% paraformaldehyde for 20 min on ice. Peptides were synthesised to >95% purity, stored as 20 mM DMSO stocks at −80°C, and 1 mM AIM-V dilutions made on the day of assay. DMSO controls were treated the same as peptide pulsed. For peptide on-rates, T2 cells were treated as before but for the durations specified in the results (1-3 h) then stained and fixed.
For peptide off-rates, T2 cells were pulsed overnight, then a proportion of them stained immediately, and fixed as above, with remaining cells being washed in an excess of media (two times using 4 ml) and incubated in 4.5 ml of AIM-V for the specified times (h) at 37°C and 5% CO2, to allow peptide to dissociate, then stained and fixed as before. Flow cytometry was performed on a FACSCanto II (BD Biosciences) and data analysed with FlowJo software (TreeStar).

Peptides Used in This study
This study made use of the HIV SL9 peptide and variants thereof. The sequences and origin of these peptides, and other control peptides utilised, are listed in Table 1.

Tcr binding to a2-slYnTVaTl involves extensive structural reorganisation
Previous studies have demonstrated that over 1.7% of CD8 + T-cells from HLA A2 + HIV-infected patient 868, stained with A2-SLYNTVATL tetramer (10). 75% of these tetramer + cells expressed a TRBV5-6 + TCR and an identical TRBV5-6 β-chain was observed in 10/18 transcripts sequenced in this study (CDR3 sequence CASSLSAVQNNEQF) (10). This same β-chain was later shown to be present in all A2-SLYNTVATL tetramer + cells in a T-cell line grown from patient 868 in 1996 (9). TCR chain antibody staining showed that all of these A2-SLYNTVATL tetramer + T-cells co-expressed a TRAV12-2 chain (11). Our previous study confirmed that CD8 + T-cells expressing this TCR were unable to control HIV virus in vitro (11). Nevertheless, we were able to use phage display to select an artificially enhanced version of the 868 TCR that could recognise the SLFNTIAVL ultimate escape variant; thereby engineering foresight into an immune receptor (11). The detection of common SL9 escape mutants on the surface of HIV-infected HLA A2 + cells by CD8 + T-cells expressing engineered, but not wild-type, 868 TCR reinforced earlier studies suggesting that these mutants are presented by HLA A2 and that escape from the wild-type epitope is mediated by loss of TCR binding (so-called "TCR escape") (6,7,11,18). To understand how positions 3, 6 and 8 in SL9 impinge on TCR binding, we solved the structure of the wild-type 868 TCR in complex with A2-SLYNTVATL to 2.9 Å resolution ( Table 2). The 868 TCR bound to A2-SLYNTVATL with a conventional, centrally-located, diagonal orientation with the TCR α-chain positioned over the α2 helix of MHC class I and the TCR β-chain positioned over the α1 helix ( Figure 1A). The TCR crossing angle of 51.6° lies within the normal range for human TCR-pMHC complexes (50), enabling positioning of the TCR CDR3 loops centrally over the peptide, and the CDR1 and CDR2 loops predominantly over the MHC helices ( Figure 1B). To gain extra insight into the mechanism by which the 868 TCR adopted this binding mode, we also solved the structure of the unbound 868 TCR ( Table 2). Comparison with the 868-A2-SLYNTVATL complex structure showed that extensive structural reorganisation of the CDR3 loops was required to avoid clashes with the solvent-exposed residues in the SL9 peptide. This resulted in an induced fit, or conformational pre-equilibrium binding mode where the TCR exists in two conformations in the unbound state, with the CDR3α loop "moving" 6.9 Å and the CDR3β moving 3.2 Å between the bound and unbound TCR structures ( Figure 1C). Closer examination of the interface between the TCR and the pMHC showed that the CDR3 loops almost exclusively made contacts with the peptide, forming a pincer around the solventexposed 4N and 5T residues in the peptide (Figures 2A,B). The TCR-peptide interaction included 10 hydrogen bonds and 52 vdW contacts, mainly focussed on 4N. The CDR1 and CDR2 loops primarily contacted the α helices of the MHC, including interactions with all three restriction triad residues (MHC residues 65R, 69A, and 155Q) as well as strong interactions with MHC residues 72Q and 154E (Figures 2C,D). Although most of the TCR-pMHC interactions were focussed on these 2 peptide residues and 5 MHC residues, the 868 TCR bound with a broad footprint, making additional bonds with all but 2 peptide residues (1S and 9L) and a further 14 MHC residues. In total, the 868 TCR made 20 hydrogen bonds, and 121 van der Waals (vdWs) contacts with A2-SLYNTVATL (Table 3), which is in the higher range compared to previously published viral TCR-pMHC complexes (35). In summary, the 868 TCR utilised an induced fit, or conformation selection, binding mode to form a pincer-like interaction around the central solvent-exposed portion of the peptide. Additional contacts with distal peptide residues enabled a relatively extensive and broad binding interface between the TCR and pMHC.
Tcr structure with the Ultimate sl9 escape Variant The 868-A2-SLYNTVATL complex structure demonstrated that the dominant contacts with the SL9 peptide were with the 4N residue rather than with the 3Y, 6V, or 8T residues that result in immune escape when substituted to create the SLFNTIAVL ultimate escape variant (6,7). Alanine scan mutation across the SL9 peptide backbone showed an almost complete loss of recognition with the L2A and Y3A changes. Alanine substitution at positions 4, 5, and 6 had the next greatest effect whereas peptides substituted at positions 1, 8, and 9 exhibited recognition equivalent to the wild-type peptide ( Figure 2E). In order to gain insight as to how natural escape mutations impinge on TCR binding, we also solved the structure of the 868 TCR with the SLYNTIATL variant present in the patient at the time the blood sample was used to grow the 868T-cell clone (9), and the SLFNTIAVL ultimate escape variant ( Table 2). Comparison of these structures with the 868 TCR in complex with the A2-SLYNTVATL index sequence showed that the 868 TCR engaged all the antigens in Positions of the 868 TCR CDR loops (1α-red, 2α-green, 3α-blue, 1β-yellow, 2β-cyan, 3β-orange) and the crossing angle of the TCR with the MHC groove is shown. (c) The positions of the 868 TCR CDR3 loops differ considerably between the ligated (red cartoon) and unligated (grey cartoon) structures. a similar conformation and with a similar interaction network (Figures 3A-E). We were initially surprised at being able to solve the structure of the 868 TCR in complex with the SLFNTIAVL peptide as this triple mutant is known to escape from CD8 + T-cells (6,7). Indeed, this mutation is not well recognised by T-cells bearing the 868 TCR (9, 11). The failure of CD8 + T-cells to recognise the SLFNTIAVL mutation has previously been attributed to reduced interactions with the TCR (6,7,18,23). Earlier structural analyses of A2-SLYNTVATL and A2-SLFNTIAVL have concluded that all three substitutions directly affected TCR recognition (6,7). However, we observed only minor differences at the TCR-peptide interface with these two ligands. For example, the 868 TCR made 3 hydrogen bonds and 12 vdWs with peptide residues 3Y, 6V, and 8T in the SLYNTVATL peptide compared to 2 hydrogen bonds and 15 vdWs contacts with 3F, 6I, and 8V in the SLFNTIAVL peptide (Figures 3C,E). The total number of contacts in each complex (868-A2-SLYNTVATL: 141, 868-A2-SLFNTIAVL: 146) was also similar, demonstrating that the 868 TCR could tolerate mutations at residues 3, 6, and 8 in SL9 (Table 3). Thus, because our structural data were inconsistent with the hypothesis that SLFNTIAVL primarily affects TCR binding, and because sum contact numbers do not necessarily represent the accurate binding energy between two molecules, we undertook further structural and biophysical experimentation to determine the mechanism that results in poor T-cell recognition of SLFNTIAVL.

a2-slYnTVaTl Does not alter conformation When bound to 868 Tcr
Previous studies comparing SLYNTVATL and the SLFNTVATL mut ated peptide showed that they adopted different con for mations within the HLA A2 binding groove (7,22). The T-cell recognition and on-rate kinetics with the G10 SL9-specific TCR in this study were very similar with A2-SLYNTVATL and A2-SLFNTVATL, leading the authors to conclude that one conformation,   (Figure 4). These data make it possible to be more definitive about the direction of travel of the previously suggested induced fit model (22) (i.e., the odd one out is A2-SLFNTVATL and this peptide presumably adopts the more conventional structure for A2-SLYNTVATL type ligands prior to, or at the time of, TCR binding). Furthermore, these data do not support the notion that HIV escape is mediated by dynamic differences in ligand engagement during TCR binding to SL9 variants.

The 868 Tcr adopts a similar energetic strategy during binding to sl9 escape Variants
We next explored the evidence for HIV escape being mediated by distinct TCR-binding parameters. It has been previously shown that the G10 SL9-specific TCR binds with a similar affinity to both A2-SLYNTVATL (KD = 2.2 µM) and A2-SLFNTVATL (KD = 5.2 µM) by SPR (22). Another study demonstrated that half maximal pMHC tetramer binding (DR50) to the SL9-specific D3 TCR was near identical for A2-SLYNTVATL, A2-SLFNTVATL, A2-SLYNTVAVL, A2-SLFNTVAVL, and A2-SLFNTIAVL (21). Finally, we have previously shown that the 868 TCR binds to a range of SL9 variants (11) (A2-SLYNTVATL, A2-SLYNTIATL, A2-SLYNTIAVL, A2-SLYNTVAVL, A2-SLFNTIATL, A2-SLFNTVAVL, A2-SLFNTVATL and A2-SLFNTIAVL). Consequently, the available data show that SL9-specific TCRs can engage SL9 variants within an affinity range that would be expected to easily activate T-cells. Here, we reproduced SPRbinding assays with the 868 TCR for the three ligands included in this study. These data demonstrated that the 868 TCR bound to A2-SLYNTVATL with a KD of 82 nM (Figure 5A) confirming that this is, by far, the strongest binding natural TCRpeptide-MHC interaction ever recorded (51,52). Indeed, the affinity was even greater with the A2-SLYNTIATL variant (KD = 38 nM; Figure 5B) that dominated in the patient at the time the blood sample was taken that was used to generate the 868 T-cell line (9,10). 868 TCR interactions with A2-SLYNTVATL and the A2-SLYNTIATL variant were characterised by a relatively slow off-rate (koff = 1.6 × 10 −2 -7.3 × 10 −3 s −1 ) compared to the majority of other natural TCR-pMHC interactions (average koff ~ 2 × 10 −1 s −1 ) (51). Although the strength of the 868-A2-SLFNTIAVL interaction was 21-fold and 45-fold weaker than for A2-SLYNTVATL or A2-SLYNTIATL, respectively, due to a much faster off-rate (koff = 2.1 × 10 −1 s −1 ), the affinity (KD of 1.77 µM; Figure 5C) was still in the range of the very best natural anti-pathogen TCRs (52,53). Thus, this weaker affinity alone was unlikely to explain the ability of the SLFNTIAVL variant to evade T-cell detection. Indeed, we have described a functional HLA A2-restricted TCR that binds its cognate pMHC with over 100-fold weaker affinity than the 868 TCR and A2-SLFNTIAVL (54,55). To get a more complete picture of how the 868 TCR engages its cognate ligands we also undertook thermodynamic analyses by SPR (Figures 5D-F; Table 4) and directly by isothermal calorimetry (ITC; Figures 5G-I). The results of these independent techniques correlated well and showed that 868 TCR engagement was driven by energetically favourable enthalpy and entropy changes to all three ligands. Thus, for the 868 TCR, there was a net increase in the number of electrostatic interactions and a transition from order-disorder during binding. Considering that the 868 TCR CDR3 loops underwent similarly large conformational changes during engagement to all three SL9 variants studied here (shown for 868-A2-SLY interaction in Figure 1C), a movement that would probably be entropically unfavourable, the expulsion of ordered water molecules from the interface, and/or order-disorder transitions in other parts of the molecules presumably contributed to the net favourable entropic change of the interaction. The lower affinity of 868 TCR for A2-SLFNTIAVL compared to A2-SLYNTVATL or A2-SLYNTIATL was associated with a much reduced enthalpic component, but more favourable entropy. We used 2D binding analysis to confirm our observations using SPR (Figures 5J,K).
In accordance with our previous findings with other TCRs (55, 56), we observed an identical relationship between the strength of interaction between 868 and the three SL9-derived ligands by 2D analysis and SPR. 868-A2-SLYNTIATL was characterised by the strongest 2D effective affinity (1.6 × 10 −3 μm 4 ), followed by 868-A2-SLYNTVATL (1.0 × 10 −3 μm 4 ), with 868-A2-SLFNTIAVL generating the weakest effective 2D affinity (5.9 × 10 −4 μm 4 ). To further confirm our observations from 2D and 3D binding experiments, we next examined the binding of pMHC tetramers to the 868 TCR at the T-cell surface. Staining of 868 TCR-transduced CD8 + primary T-cells was similar for tetramers made with A2-SLYNTVATL, A2-SLYNTIATL, and A2-SLFNTIAVL ( Figure 6A). The CD8 coreceptor is known to bind to a site on MHC class I distinct from the TCR docking platform and cooperates with the TCR to aid the engagement of pMHC tetramers (57) and the absence of CD8 binding is known to raise the TCR affinity threshold required for tetramer staining (58,59). 868 TCR + Jurkat T-cells, that do not express CD8, also stained equally well with A2-SLYNTVATL, A2-SLYNTIATL, and A2-SLFNTIAVL tetramers ( Figure 6B). Given our observations that the TCR affinity threshold required for T-cell activation is considerably lower than that required for pMHC tetramer binding (58) and finding that these reagents can even fail to stain fully functional T-cells (47,60), our results suggest that the SLFNTIAVL ultimate escape variant would still be a very  efficient T-cell agonist. Indeed, the 868 TCR-transduced primary CD8 + T-cells exhibited similar reactivity to SLYNTVATL and SLFNTIAVL peptides when using standard T-cell activation assay whereby T-cells, antigen-presenting cells, and peptide are added directly to tissue culture wells and allowed to incubate ( Figure 6C). We conclude that while there is a substantial  reduction of TCR binding to the SLFNTIAVL triple mutant sequence, this reduction cannot explain the observed immune escape per se as assumed by all previous studies in this system (6,7,11,23). The other way by which peptides can escape from host CTL surveillance is by mutating MHC anchors to reduce binding and presentation at the infected cell surface. However, previous studies have shown that the SLFNTIAVL peptide binds to HLA A2 with 50-400% of the affinity the index SL9 sequence depending on the study ( Table 5). Such minor relative differences were not thought to be able to explain how the triple mutant peptide escapes from CTL (6,7,9,18). Our own studies with a modified version of the 868 TCR that recognises the escape variants well showed that this TCR could recognise CD4 + T-cells infected with primary HIV isolates carrying known escape variants (11). Thus, these variants must be presented at the cell surface in the context of HLA A2. Importantly, these variants were not recognised by T-cells expressing the wild-type 868 TCR, thus verifying that these viruses are indeed escape mutants (11). Consequently, the established literature indicates that the common variant peptides in this system bind sufficiently well to HLA A2 to be presented (6,7,9,18) and sufficiently well to the 868 and D3 TCRs to be recognised by T-cells expressing these SL9-specific receptors. These apparently incongruous findings prompted us to next re-examine peptide binding in the SL9 system.

cTl escape by the Ultimate sl9 escape Variant Mediated by a Fast Peptide Off-rate
Detailed analysis of the thermal melting of A2-SLYNTVATL, A2-SLYNTIATL, and A2-SLFNTIAVL by circular dichroism showed the melting temperature (Tm) of the wild-type peptide was higher than that of the triple mutant (57°C compared to 49°C; Figures 7A,B) suggesting that the triple mutant is less stable. These data are in accordance with two previous studies showing that SLFNTIAVL binds to HLA A2 with ~50% reduced affinity compared to SL9 (9, 18) but at odds with two studies indicating that SLFNTIAVL is the better ligand (6,7). Peptide binding to HLA A2 has not previously been thought to result in the immune escape observed with this epitope, which instead has been attributed to a lack of recognition by SL9-specific TCRs (6,7,11,23). Stabilisation of HLA A2 at the T2 cell surface in  Figure 7C) showed that the SLFNTIAVL peptide stabilised HLA A2 with 89% of that seen with the wild-type SL9 peptide after 3 h. This relatively minor difference increased in an overnight assay where stabilisation with SLFNTIAVL was ~65% of that seen with the wild-type peptide. Our previous examinations of HLA A2 peptide stability by SPR have shown that good agonists produce pMHC with a half-life of >6 h in a cell free system (61). Similar experiments, performed by examining the relative binding of 868 TCR to A2-peptide by SPR over a time course at 37°C, showed that both A2-SLYNTVATL and A2-SLYNTIATL had a half-life of >7 h. By contrast, the half-life of A2-SLFNTIAVL on the same chip was <2 h (Figure 8A). We thus reasoned that the stabilisation of peptide-HLA measured in a short-term binding assay like those previously used in this system (9,18) would be determined more by peptide on-rate than the peptide off-rate or dwell time. To further test this possibility, we examined the "on rate" of peptide in T2 cell stabilisation assays. Relative stabilisation was similar for the SLYNTVATL, SLFNTIAVL and the GILGFVFTL influenza matrix-derived peptide that is often used as a positive control for HLA A2 binding (Figure 8B: left panel). We also examined the stability of these A2-peptide complexes at the cell surface by monitoring the amount of HLA A2 on T2 cells incubated with each peptide prior to washing and further culture.  (g-i) Thermodynamic analysis of the 868 TCR-pMHC interaction was also performed using isothermal titration calorimetry 200 instrument to directly measure ΔH. These analyses show consistent values to those generated by SPR. Overall, the 868 TCR uses a generally similar binding mechanism and thermodynamic signature to interact with all three ligands. (J) Effective 2D binding frequency, using at least 5 cell pairs, and calculated as an average of 100 cell-cell contacts. (K) Effective 2D affinity (AcKa) calculated using adhesion frequency assays and reported in the text as geometric mean. Statistics were performed on log-transformed affinities and analysed with two-tailed, unpaired parametric t-tests and assumption of equal SDs. The vast majority of HLA A2 stabilised with GILGFVFTL or SLYNTVATL peptide was still present at the cell surface 3 h after the peptide was removed from the assay by cell washing. In stark contrast, most of the HLA A2 stabilised by the SLFNTIAVL peptide had decayed in half this time (Figure 8B: right panel).
Collectively, these binding data explain how previous studies, using short-term binding assays in the continuous presence of peptide, have shown minor differences in the binding of SL9 and SLFNTIAVL escape variant ( Table 5). They also suggest that the dominant mechanism by which the SLFNTIAVL mutant allows immune escape is due to this peptide producing a much shorterlived peptide-HLA complex. This reduced half-life of antigen at the infected cell surface would be expected to considerably reduce the antigen density on infected cells and thereby facilitate immune escape. To test this notion, we next examined how long peptide pulsed cells could remain as T-cell targets for CD8 + T-cells transduced with the wild-type 868 TCR. These revealing experiments allowed examination of the combined effects of differences in TCR binding and HLA stability in a relevant biological context. Cells pulsed with the SLYNTVATL remained effective targets for 868 TCR-expressing CD8 + T-cells after 24 h of culture ( Figure 8C; Figure S1 in Supplementary Material). By contrast, targets pulsed with the SLFNTVATL, SLYNTIATL, or SLYNTVAVL variant peptides showed reduced levels of recognition after 24 h. The SLFNTIAVL triple mutant sequence produced the largest effect and was recognised well if experiments were performed immediately but recognition dropped to less than 20% of that seen for the SLYNTVATL peptide at t0, after targets were cultured for 24 h prior to the T-cell recognition assay. Similar data were observed when primary CD4 + T-cells, which form the major reservoir for HIV virus in vivo, were used as targets ( Figure 8D). Relative to SLYNTVATL, the SLFNTIAVL pulsed cells made  better targets at t0 (125%) but their ability to activate the 868 TCR cells diminished to 18% when allowed to culture for 24 h prior to assay ( Figure 8D; Figure S2 in Supplementary Material). These results suggest that the major mechanism by which SLFNTIAVL escapes from CTL is through loss of peptide-HLA stability, although reduced TCR interaction with this mutant is also likely to play a secondary role.

DiscUssiOn
Recognition and escape from the HLA A2-restricted HIV epitope SL9 has been examined in well over 50 studies to date, but major questions are still outstanding in this system as there has been no TCR-A2-SL9 co-complex structure. Here, we solved the structure of the free 868 TCR, a TCR first described almost 20 years ago (10), and the structure of this TCR in complex with A2-SLYNTVATL. Comparison of the free and ligated TCR structures suggested an "induced fit" mechanism that involved major structural reorganisation of the antigen binding face of the TCR. The CDR3 loops made the biggest movement (almost 7 Å for CDR3α and >3 Å for CDR3β) to avoid steric clashes with the central portion of the peptide, while the CDR1 and CDR2 loops interacted mainly with MHC via a more "lock-and-key" interaction.
It has been known for over a decade that that the SLYNTVATL and SLFNTVATL bind to HLA A2 with very different conformations, but engage the G10 TCR with near identical on-rates, leading the authors to conclude that one conformation, common to both peptides, was recognised by TCRs (22). Since A2-SLYNTVATL does not change upon TCR binding, we can now be confident that the conformation adopted by this, and many other SL9 variants (21), dominates during TCR recognition. Thus, the evidence suggests that A2-SLFNTVATL undergoes an induced fit upon TCR binding (22) although conclusive proof will await a TCR-A2-SLFNTVATL structure. We further solved the co-complex structure of the 868 TCR with the A2-SLYNTIATL (the variant that predominated in patient 868 in 1996 when the patient blood sample from which this T-cell clone was grown was taken), and the structure of the 868 TCR in complex with the A2-SLFNTIAVL, the ultimate escape variant in this system. Resolution of the latter structure came as a surprise as residues that mutate in positions 3, 6, and 8 to produce this mutant are known to be solvent exposed and had been assumed to act as prominent TCR contacts (6,7,(21)(22)(23). In contrast to this prevailing hypothesis we found that the asparagine at position 4 was the dominant contact residue. Indeed, there were more contacts with this residue than with the position 3 tyrosine, position 6 valine, and position 8 threonine combined. An alanine mutation scan across the peptide backbone showed that while positions 4 and 5 in SLYNTVATL were important for recognition, the most severe non-MHC anchor change was Y3A. In accordance, the position 3Y sub-library also gave the strongest response of the 180 sub-libraries during a 9-mer positional scanning combinatorial library screen of T-cells transduced with the 868 TCR (62). The importance of the tyrosine residue at position 3 is consistent with our previous observation that large bulky amino acids at this position in HLA A2-bound peptides can form a bridge with more C-terminal residues (usually position 5) (63). Replacing residue 3 with a smaller alanine side chain can abolish this important intra-peptide stabilisation and have knock-on effects at peptide residue 4 that abrogate TCR binding. Thus, our findings here point towards a common mechanism of peptide presentation by HLA A2 in which peptide residue 3 can be essential for maintaining antigenic identity.
We next undertook a detailed biophysical and thermodynamic study of 868 TCR binding to A2-SLYNTVATL, A2-SLYNTIATL, and A2-SLFNTIAVL. The 868 TCR exhibited, by far, the strongest affinity of a peptide-HLA ligand for a natural TCR (52) and bound to A2-SLYNTVATL and A2-SLYNTIATL ligands with KDs of 82 and 38 nM, respectively, or over 10× stronger binding than the next highest TCR-pMHC interaction measured by BIAcore (52). The median affinity for human antiviral TCR-pMHC interactions is KD ~5 μM (52). 868 TCR binding was characterised by a relatively slow offrate from the A2-SLYNTVATL and A2-SLYNTIATL ligands. The other SL9-TCR that has been examined by SPR, G10, had an off-rate 3.75 times faster than 868, and a 4.5 times slower on rate, resulting in a weaker affinity of KD of 2.2 µM for A2-SLYNTVATL compared to 868 (22). While binding of the 868 TCR was much reduced for A2-SLFNTIAVL compared to the other ligands, it still had a very respectable affinity (KD = 1.77 μM) that falls at the higher end of the spectrum for TCR agonists (52).
The mode of 868 TCR engagement and its affinity with the A2-SLFNTIAVL ultimate escape variant in this system did not lend full support to the prevailing belief that loss of recognition of SL9 occurs mainly via interfering with TCR binding. The fact that CD8 + T-cells transduced with the 868 TCR could recognise SLFNTIAVL pulsed target cells well and stained with A2-SLFNTIAVL tetramers was also a surprise, but in accordance with an earlier study that showed that A2-SLFNTIAVL bound to the D3 TCR as well as A2-SLYNTVATL in tetrameric form and that the SLFNTIAVL concentration required for half maximal lysis was a highly respectable 4 × 10 −10 M (21). In further accordance with positions 3, 6, and 8 not dominating A2-SL9-specific TCR binding, the position 4 asparagine sub-library was by far the most potent of the 180 sub-mixtures for activating T-cell clone 003 known to express a different SL9-specific TCR (62). These combined data do not support the notion that the SLFNTIAVL escapes from CTL via lack of TCR binding, prompting us to take a closer look at the MHC binding of ligands in this system. Data from three different studies using three different methodologies show that SLFNTIAVL binds to HLA A2 with 50-400% of the affinity of SLYNTVATL. Our previous SPR experiments have shown that good agonists produce pMHC with a half-life of >6 h in a cell-free system (61). Indeed, our data here show that both A2-SLYNTVATL and A2-SLYNTIATL had half-lives of over 7 h on an SPR chip maintained at 37°C compared to <2 h for A2-SLFNTIAVL. We reasoned that the peptide-binding assays that have been performed previously in this system, including by ourselves (9), would be most influenced by peptide on-rate but that the biology would be most linked to peptide off-rate at the target cell surface. HLA A2 + target cells pulsed with SLYNTVATL peptide remained effective targets for SL9-specific CD8 + T-cells after 24 h of culture as has been observed with a different HIV Gag-derived epitope (27). The SLFNTVATL, SLYNTIATL and SLYNTVAVL peptides [which bind to the 868 TCR with dissociation constants of 2.9 µM, <100 nM, and ~300 nM, respectively (52)] were seen well when targets presenting these ligands were used immediately for T-cell assays but exhibited only 67, 60, and 35% recognition of that observed with SLYNTVATL peptide when targets were cultured for 24 h prior to the T-cell assay. The biggest effect was observed with the SLFNTIAVL where >80% of the activity was lost after 24 h of culture. These results were supported by other experiments looking the stability of the A2-SLFNTIAVL complex in comparison to other ligands.
Collectively, our results suggest that the instability of the A2-SLFNTIAVL molecule is likely to provide the major mechanism of escape from SL9-specific CTL. The accumulation of mutations at positions 3, 6, and 8 to fixation over a 10-year period (6,7) demonstrates the extraordinary lengths that HIV will go to in order to avoid detection by these key antiviral cells. Our study further highlights that the results of short-term, steady-state peptide-binding assays can be misleading and suggests that future studies should employ more relevant biological readouts.

acKnOWleDgMenTs
The authors would like to thank Diamond Light Source for beam time (proposals mx4532-1, mx6232-1, mx6232-9 and mx6232-10), and the staff of beamlines DLS I03, DLS I04-1 and DLS I24 for assistance with crystal testing and data collection. We are extremely grateful to Andrew McMichael for providing advice and encouragement and to Anton van der Merwe for discussions about the biophysics and mechanism of SL9 TCR escape.

FUnDing
This work was supported by the UK Biotechnology and Biological Sciences Research Council (Grant BB/H001085/1). AKS is a Wellcome Trust Senior Investigator. FM was funded by a Tenovus PhD studentship. DC is a Wellcome Trust Research Career Development Fellow (WT095767). PR was supported by a RCUK Fellowship, and supported in part by grant GM067079 from the National Institute of General Medical Sciences, National Institutes of Health. JM is an Australian National Health and Medical Research Council (NHMRC) Career Development Fellow (GNT1131732). BDE was funded by NIH (R01 AI096879).