HLA Micropolymorphisms Strongly Affect Peptide–MHC Multimer–Based Monitoring of Antigen-Specific CD8⁺ T Cell Responses Free

In 1996, Altman et al. (1) described how multimers of peptide–MHCs (pMHCs) coupled to fluorochromes can be used to monitor Ag-specific CD8⁺ T cells by flow cytometry. Since this first description, pMHC multimer–based immunomonitoring has become a very widely used technique to understand spontaneous and therapy-induced T cell reactivity in different fields, as illustrated by the >3000 citations to the original work (1). In addition to the use of pMHC multimers for the quantification of Ag-specific T cell responses (2), these tools have also been used for the isolation of Ag-specific T cells for both research purposes and adoptive cell therapy (3) and to describe the composition of the naive T cell compartment (4, 5).

Traditionally, MHC-based immunomonitoring projects have focused on analyses of T cell reactivity toward a small number of epitopes, restricted by a few HLA alleles that are present at high frequency within the Caucasian population. To illustrate this bias, >65% of melanoma-associated Ags described in literature are restricted by the HLA-A*02 allele (6), even though this allele is only one of the many HLA-A alleles that are present. Notably, in non-Caucasian populations, the HLA-A*02:01 subtype (by far the most frequent subtype in Caucasian populations) is present in only a minor fraction of HLA-A*02⁺ individuals (7).

The variation between the subtypes of specific HLA alleles (traditionally referred to as four-digit subtypes) maps to a small number of sequence differences within the α1 or α2 domains of the MHC H chain, which together form the peptide-binding groove. Depending on their location, these sequence differences may influence T cell recognition by modifying the peptide binding properties of the MHC (8) or altering the TCR exposed surface, or they may be without effect.

In this study, we have assessed the extent to which the ability to correctly measure Ag-specific T cell responses against a given epitope by pMHC multimer technology is influenced by such micropolymorphic differences between HLA subtypes. Our results demonstrate that minor variations between HLA subtypes greatly affect the ability to detect Ag-specific CD8⁺ T cell responses in both model systems and clinical samples. Furthermore, in many cases, this lack of T cell detection reflects altered TCR interaction of the pMHC complex, rather than impaired MHC binding of epitopes, and can therefore not be predicted by pMHC binding assays. Collectively, these data indicate that the generation of pMHC-based monitoring technology for large sets of HLA subtypes, as performed in this study for HLA-A*02, will be of importance for the development of personalized pMHC-based immunomonitoring.

Amino acid sequences of the HLA-A*02 alleles were obtained from the ImMunoGeneTics information system/HLA database (http://www.ebi.ac.uk/ipd/imgt/hla/align.html, version 3.8.0. Accessed: June 14, 2012) (9). The full-length protein sequences of HLA-A*02:02, HLA-A*02:05, HLA-A*02:06, HLA-A*02:07, HLA-A*02:11, HLA-A*02:71, and HLA-A*02:77 were aligned to the reference sequence HLA-A*02:01.

Graphic representations of the CMV_NLV-specific TCR (RA14) in complex with the CMV_NLV-HLA-A*02:01 complex (10) (Brookhaven Protein Data Bank code 3GSN) were drawn using PyMOL software.

A streptavidin (SA)-based sandwich ELISA was used to assess the ability of peptides to stabilize the HLA-A*02:01, HLA-A*02:02, HLA-A*02:05, HLA-A*02:07, HLA-A*02:11, HLA-A*02:71, and HLA-A*02:77 subtypes, as described (11), using the PeliScreen HLA class I ELISA kit (Sanquin). In brief, after UV-mediated peptide exchange in the presence of indicated peptides, biotinylated HLA class I molecules were added to wells coated with SA. HLA class I complexes bound to the well were quantified using an HRP-conjugated anti–β₂-microglobulin Ab. Green-colored oxidation product was produced by adding ABTS as a substrate for HRP.

The following peptide sequences were submitted to the NetMHCpan server (http://www.cbs.dtu.dk/services/NetMHCpan/, version 2.4. Accessed: June 14, 2012) (12): CMV_NLV (NLVPMVATV), EBV_GLC (GLCTLVAML), EBV_YVL (YVLDHLIVV), FLU_GIL (GILGFVFTL), GnTV_VLP (VLPDVFIRC), HA-2_YIG (YIGEVLVSV), MAGE-A10_GLY (GLYDGMEHL), MAGE-C2_ALK (ALKDVEERV), MAGE-C2_KVL (KVLEFLAKL), Meloe-1_TLN (TLNDECWPA), and MART-1_ELA (ELAGIGILTV). Predicted IC₅₀ values (in nanomolars) for binding to HLA-A*02:06 were obtained for each peptide. Peptides with IC₅₀ values of <50, 50–500, and >500 nM were regarded as strong, weak, and nonbinders, respectively.

Indicated virus-derived, melanoma-associated, and UV-cleavable peptides were synthesized in-house as described previously (13). Recombinant HLA-A*02 subtype-specific H chains and human β₂-microglobulin L chain were produced in Escherichia coli and isolated from inclusion bodies. MHC class I refolding reactions and purification by gel filtration HPLC were performed as described previously (14). HLA-A*02 subtypes were refolded in the presence of the following UV-sensitive peptides: KILGFVF-J-V, KMDI-J-VPLL, LLDSD-J-ERL, LTA-J-FLIFL, and SVRD-J-LARL. MHC complexes loaded with the indicated peptides of interest were generated by UV-induced ligand exchange. In brief, pMHC complexes with UV-sensitive peptide (100 μg/ml) were subjected to 366 nM UV light (Camag) for 1 h at 4°C in the presence of rescue peptide (200 μM) (13, 15).

For each peptide, pMHC-multimers were generated using two different fluorescent SA conjugates (Invitrogen) (16). T cell clones were stained with dual color–labeled pMHC multimers. Tumor-infiltrating lymphocytes (TILs) were analyzed by combinatorial pMHC multimer stainings, in which pMHC multimers loaded with 25 different peptides were used simultaneously in each staining reaction, with each pMHC combination encoded by a unique two-color combination (17).

PBMC samples were obtained from healthy individuals or from patients with stage IV melanoma in accordance with local guidelines, and following informed consent.

Cell sorting was performed on a FACSAria I (BD Biosciences), MoFlo Legacy (Beckman Coulter), or MoFlo Astrios (Beckman Coulter). Ag-specific T cell clones specific for the viral Ags CMV_NLV (NLVPMVATV), EBV_GLC (GLCTLVAML), EBV_YVL (YVLDHLIVV), and FLU_GIL (GILGFVFTL) were obtained by pMHC multimer–assisted single cell sorting from PBMCs of healthy donors. Ag-specific T cell clones for the melanoma-associated Ags MAGE-A10_GLY (GLYDGMEHL), MAGE-C2_ALK (ALKDVEERV), MAGE-C2_KVL (KVLEFLAKL), GnTV_VLP (VLPDVFIRC) and Meloe-1_TLN (TLNDECWPA) were obtained by pMHC multimer–assisted single cell sorting from TILs and PBMCs of melanoma patients (18). The HA-2_YIG–specific CD8⁺ T cell clone was a gift of M.H. Heemskerk (Leiden University Medical Center, Leiden, The Netherlands). Sorted single CD8⁺ T cells were expanded as described previously (18).

To obtain bulk T cell populations (specific for MART-1_ELA and GnTV_VLP) of an HLA-A*02:06⁺ melanoma patient, TILs were first expanded using a 14-d rapid expansion protocol (19). In brief, irradiated (40 Gy) PBMC feeders of three different healthy donors were combined and taken in culture with TILs (ratio of 20:1 for feeders/T cells) in the presence of anti-CD3 (OKT3; final concentration, 30 ng/ml) and IL-2 (final concentration, 3000 IU/ml). Half of the medium was replaced after 5 d and cultures were split 1:1 on day 7 and when needed afterward. After 12–14 d, T cells were frozen. MART-1_ELA– and GnTV_VLP-specific T cell populations were isolated from expanded TILs by sorting of live single CD8⁺ T cells stained with PE/allophycocyanin-labeled pMHC multimers. Obtained MART-1_ELA and GnTV_VLP cell populations were further expanded using the above-described rapid expansion protocol.

TILs were stored in liquid nitrogen and were thawed 1 d prior to analysis. As a standard, cell numbers were determined by trypan blue staining, with an average recovery of 93%.

Four-digit resolution HLA haplotyping was performed by the Leiden University Medical Center (Leiden, The Netherlands). Typing was performed according to the manufacturer’s protocol using SBT Excellerator HLA-A, HLA-B, and HLA-C typing kits (GenDx).

For pMHC multimer staining of CD8⁺ T cell clones, 1 μl PE-pMHC and 2 μl allophycocyanin–pMHC multimers were used to stain cells in a 15-min incubation at 37°C. Subsequently, cells were stained with anti–CD8-FITC (BD Biosciences) and near-infrared (near-IR) Live/Dead stain (Invitrogen) for 30 min on ice and washed twice with FACS buffer. Background signal was determined by staining with pMHC multimers loaded with a control peptide, HIV_SLY (SLYNTVATL).

For pMHC combinatorial encoding, each pMHC complex was coupled to a unique combination of two fluorochromes (fluorochrome concentrations as described in Ref. 16). The melanoma-associated epitope panel consisting of 145 different peptides has been described previously (16).

Ag-specific T cells were detected using an LSR II flow cytometer (Becton Dickinson) with FACSDiva software (Becton Dickinson). To identify Ag-specific T cells when staining CD8⁺ T cell clones, the following gating strategy was used: 1) selection of live (IR dye–negative) single-cell lymphocytes; 2) selection of anti–CD8- FITC⁺ cells; and 3) selection of CD8⁺ T cells that were double-positive for PE and allophycocyanin pMHC multimers. A total of 50,000 CD8⁺ T cells were recorded. Data shown are representative of three independent experiments.

For combinatorial coding analyses, an 11-color instrument setting was used (16). To identify Ag-specific T cells, the following gating strategy was used: 1) selection of live (IR dye–negative) single-cell lymphocytes; 2) selection of anti–CD8-FITC⁺ and “dump” (anti-CD4, -CD14, -CD16, -CD19; Invitrogen) negative cells; 3) selection of CD8⁺ T cells that were positive in two and only two pMHC multimer channels; and 4) cells positive in only one as well as cells positive in three or more pMHC multimer channels were gated out, as described previously (16).

Predefined cut-off values for the definition of positive responses were ≥0.005% of total CD8⁺ cells and ≥10 events. Five hundred thousand CD8⁺ T cells were recorded per sample. Identified Ag-specific T cell responses were in all cases confirmed in an independent stain using a different fluorochrome combination.

pMHC multimers charged with either GnTV_VLP (VLPDVFIRC), MART-1_ELA (ELAGIGILTV), gp100_KTW (KTWGQYWQV), or tyrosinase_YMD (YMDGTMSQV) peptides, coupled to SA-PE and SA-allophycocyanin, were used to stain expanded TILs of an HLA-A*02:06⁺ melanoma patient at the indicated concentrations. A 1× pMHC multimer staining condition corresponds to an MHC concentration of 6 μg/ml. Anti–CD8-FITC (BD Biosciences) and near-IR Live/Dead stain (Invitrogen) were used to gate on CD8⁺ and living cells.

All FCS files are available for scientific research purposes upon request.

To address to what extent pMHC-based immunomonitoring is influenced by micropolymorphic variation within the HLA class α1 and α2 domains, we selected a set of eight different HLA-A*02 subtypes (HLA-A*02:01, HLA-A*02:02, HLA-A*02:05, HLA-A*02:06, HLA-A*02:07, HLA-A*02:11, HLA-A*02:71, HLA-A*02:77). Of this set of alleles, HLA-A*02:01 is the most frequently occurring subtype within the Western European and North American population and therefore the commonly used HLA-A*02 allele in immunomonitoring studies. When taking HLA-A*02:01 as a reference, the other seven HLA-A*02 subtypes that were chosen differ from HLA-A*02:01 in one to four residues within the α1 and/or α2 domains (Fig. 1A). The location of these polymorphic residues suggests that, in at least in some cases, these alterations are unlikely to be directly “seen” by Ag-specific TCRs (Fig. 1B). As an example, the HLA-A*02:06 sequence differs from HLA-A*02:01 in only one residue, a conservative phenylalanine to tyrosine substitution at position 9 (F9Y) in the H chain. This residue forms a key component of peptide-binding pocket B that accommodates an “anchor residue” at position 2 in HLA-A*02–binding peptides (20–22). Despite the additional OH group present in the HLA-A*02:06 H chain, the B pocket of HLA-A*02:01 and HLA-A*02:06 share the same preference for small aliphatic side chains at this position (22, 23). Therefore, based on their similar peptide-binding preferences, this alteration forms an example of a micropolymorphism that could be expected to have little effect on T cell analysis.

To analyze the ability of this set of subtype variants to detect Ag-specific T cell populations, we first established UV-induced peptide exchange technology for each of them. In prior work, UV-induced peptide exchange was developed for HLA-A*02:01 (24), HLA-A*02:06, HLA-A*02:07, and HLA-A*02:11 (7). Based on the observation that the (predicted) peptide-binding motifs of these HLA-A*02 subtypes are highly similar (23), the previously developed conditional ligands were used to refold the HLA*A-02:02, HLA-A*02:05, HLA-A*02:71, and HLA-A*02:77 alleles. For all new subtypes, this yielded a conditional ligand that led to efficient refolding of HLA class I, and that could be exchanged for peptides of interest by UV light exposure (data not shown).

To subsequently determine the effect of polymorphic differences within the α1 and α2 domains of pMHC multimers on their ability to detect CD8⁺ T cell responses against a series of Ags, HLA-A*02:01–restricted T cell clones specific for 10 different viral and melanoma Ags were generated (CMV_NLV, EBV_GLC, EBV_YVL, FLU_GIL,) or expanded (GnTV_VLP, MAGE-A10_GLY, MAGE-C2_ALK, MAGE-C2_KVL, and Meloe-1_TLN (18), and HA2_YIG). For all these T cell clones, HLA-A*02:01 was the only A*02 subtype present. Thus, no other alleles were expressed that could also form the restriction element of the epitopes concerned. Subsequently, pMHC multimers of all eight HLA-A*02 subtypes were produced for each of the 10 different Ags by UV-mediated peptide exchange (7, 24). The set of CD8⁺ T cell clones was then stained with all pMHC multimer subtypes, loaded either with cognate peptide or with a control peptide.

Representative flow cytometry dot plots of an HLA-A*02:01–restricted CMV_NLV-specific CD8⁺ T cell clone are shown in Fig. 2A. As expected, high-intensity staining (mean fluorescence intensity [MFI], 2.6 × 10³) of this CD8⁺ T cell clone was achieved when subtype-matched (HLA-A*02:01) pMHC multimers were used. A similarly high level of staining was achieved when using HLA-A*02:71 (MFI, 3.0 × 10³) and HLA-A*02:77 (MFI, 1.6 × 10³) pMHC multimers. In contrast, for all four other HLA-A*02 subtypes, signal did not exceed background (MFI range, 0.7–1.1 × 10²). To determine whether a failure to detect these CMV_NLV-specific T cells could have been predicted on the basis of pMHC binding to these HLA-A*02 subtypes, we performed pMHC binding ELISAs for HLA-A*02:01, HLA-A*02:02, HLA-A*02:05, HLA-A*02:07, HLA-A*02:11, HLA-A*02:71, and HLA-A*02:77. Technical limitations precluded pMHC ELISA for the HLA-A*02:06 allele, and for this allele binding affinity predictions were therefore performed using netMHCpan (12) (Supplemental Table I). The CMV_NLV peptide was shown/predicted to bind to seven of eight HLA-A*02 subtypes (Fig. 3, Supplemental Table I), including four of the five alleles that failed to detect the CMV_NLV-specific T cells. Thus, the absence of T cell staining for these alleles cannot simply be predicted on the basis of pMHC binding analysis of the epitope. The same analysis was subsequently repeated for each CD8⁺ T cell clone (Fig. 2B). In all cases, staining with the subtype-matched HLA-A*02:01 multimers (filled histograms) yielded a high-intensity staining. In contrast, staining with any of the seven subtype-mismatched pMHC multimers was highly variable, with some clones being stained by many pMHC multimers (EBV_GLC), whereas others were detected by only a few subtype-mismatched pMHC multimers (MAGE-C2_KVL). Also for these epitopes, pMHC binding analysis could not explain the absence of multimer staining in most cases. For those alleles for which pMHC binding measurements were feasible (i.e., excluding HLA-A*02:06), absence of T cell staining could be explained by impaired pMHC binding in only 3 of 35 cases (Fig. 3; CMV_NLV, FLU_GIL, and MAGE-C2_ALK in the HLA-A*02:07 panel; the MART-1_ELA peptide is ignored in this matter, as the Ag-specific T cell clone is HLA-A*02:06 restricted and peptide binding could not be assessed for this allele). As a side note, the pMHC binding measurements confirm the structural integrity of the vast majority of the pMHC complexes.

In the above analyses, pMHC multimers were generated using HLA-A*02 subtypes for which these 10 epitopes were originally not identified. We next wished to assess whether a set of HLA-A*02:01 multimers loaded with previously described epitopes for this HLA-A*02 subtype could correctly describe T cell reactivity in a patient expressing a related HLA-A*02 subtype. To this purpose, we created a collection of HLA-A*02:01 pMHC multimers loaded with a set of 145 HLA-A*02:01–restricted melanoma-associated epitopes. Subsequently, CD8⁺ T cell responses against this set of epitopes were analyzed in TIL cultures from an HLA-A*02:06⁺ melanoma patient (HLA-A*02:01 and HLA-A*02:06 differ at position 9, F9Y) by combinatorial coding analysis (16). By performing the same analysis for this collection of epitopes complexed with HLA-A*02:06 multimers, it was possible to directly compare T cell reactivity as measured by use of subtype-matched and subtype-mismatched pMHC multimers. These experiments revealed that when subtype-matched (HLA-A*02:06) pMHC multimers were used, four different T cell responses were detected, reactive with the melanoma Ags gp100_KTW, tyrosinase_YMD, MART-1_ELA, and GnTV_VLP (Fig. 4). Importantly, when the same analysis was performed using subtype-mismatched (HLA-A*02:01) pMHC multimers, only the gp100_KTW-specific T cell response was detected at a similar magnitude (0.199 and 0.236% of CD8⁺ T cells using HLA-A*02:06 and HLA-A*02:01, respectively). Specifically, the magnitude of two of the four T cell responses (tyrosinase_YMD and MART-1_ELA) was underestimated by 2- to 3-fold. Furthermore, the T cell response specific for the GnTV_VLP Ag was missed entirely when using HLA-A*02:01 pMHC multimers (Fig. 4), even though this peptide constitutes a bona fide HLA-A*02:01–restricted epitope (25).

To also investigate the reactivity of these HLA-A*02:06–restricted T cell populations with the other HLA-A*02 subtypes, the GnTV_VLP- and MART-1_ELA–specific T cell populations present in this TIL product were isolated using HLA-A*02:06–restricted pMHC multimers. Subsequently, both T cell populations were stained with the panel of HLA-A*02 subtype multimers (representative dot plots shown in Fig. 5A). High-intensity staining (MFI, 2.0 × 10³) of GnTV_VLP-specific T cells was achieved when subtype-matched (HLA-A*02:06) pMHC multimers were used. However, for the other HLA-A*02 subtype multimers, staining patterns were again highly variable and never reached the sensitivity of matched pMHC multimers (MFI range, 0.9–5.8 × 10², Fig. 5A), even though all mismatched HLA-A*02 subtypes were shown to bind the GnTV_VLP peptide (Fig. 3). When the same analysis was performed for the MART-1-HLA-A*02:06–reactive T cell population, a highly variable staining with the seven other HLA-A*02 subtypes was likewise observed (Fig. 5B). pMHC multimer titration experiments (Supplemental Fig. 1) demonstrated that the ability or inability to detect Ag-specific T cell populations cannot be remedied by an increase in the pMHC multimer concentration used for T cell staining.

In this study, we demonstrate that micropolymorphic variation between HLA subtypes can greatly influence the ability to describe human Ag-specific CD8⁺ T cell responses by pMHC multimer staining. We have analyzed this issue for 88 situations in which Ag-specific T cells were stained with subtype-mismatched HLA-A*02 multimers. Of those 88 cases, detection of Ag-specific T cells occurred with a comparable sensitivity as with subtype-matched HLA multimers in only 35% of cases, and in >55% of cases no T cell staining was detected altogether.

The strong effect of minor sequence variation on pMHC-based immunomonitoring that is described in this study extends earlier work that demonstrated specific cases in which sequence variation between HLA subtypes was shown to influence T cell recognition of target cells. Specifically, in work by Yu et al. (26), it was shown that the HLA-B*57:01 and HLA-B*57:03 subtypes, which differ by only two amino acids, can both present the same HIV-1 Gag KF11 epitope, but with T cell recognition only being observed in the context of HLA-B*57:01. Likewise, a differential recognition of the same EBV epitope by Ag-specific T cells has been observed for three of the HLA-B*44 subtypes (27). Structural studies provide strong evidence that the effect of HLA micropolymorphisms on TCR interaction can in many cases not be predicted by peptide binding studies. Specifically, variants of the HLA-B*44 subtype have been shown to present a largely identical epitope repertoire, with T cell recognition primarily being influenced by the effect of the “hidden” polymorphic residue on the conformation of the peptide Ag with the peptide-binding groove (28, 29). Even more striking, crystallographic analysis has demonstrated that micropolymorphisms within the HLA-B*57 allele do not influence the structure of HLA-B*57 complexed with the Gag KF11 epitope in isolation: Only upon TCR ligation, the conformation of the pMHC complex becomes different for HLA-B*57:01 and B*57:03.

Based on this earlier work on target cell recognition, and based on the observation that the peptides used in the present study were shown to bind to most of the HLA-A*02 subtype alleles (Fig. 3), we consider it likely that altered TCR recognition of pMHC subtype variants also forms the major factor preventing accurate T cell detection in the current study. Furthermore, the observation that for polyclonal Ag-specific T cell populations, subtype-mismatched HLA multimers in some cases selectively fail to detect part of the Ag-specific T cell population (Fig. 5B) also provides evidence for altered TCR binding as a contributing mechanism. As such, it will not be feasible to prevent this issue by pMHC binding studies.

The present data clearly demonstrate that human immunomonitoring requires high-resolution matching between the HLA alleles of biological samples and the pMHC reagents used for monitoring. Pioneering work of Dausset, Bodmer, and others resulted in the first descriptions of HLA Ags by serological typing (30). Serological typing of the HLA haplotype is experimentally straightforward and is still commonly used in clinical studies that use pMHC-based monitoring (31, 32). Likewise, the use of low-resolution PCR for HLA haplotype analysis, as still frequently performed (33), does not provide information on the sequence variability that is shown in this study to be of major importance. In future clinical studies, this issue may be readily addressed by incorporation of high-resolution PCR-based haplotyping. Additionally, for the recently developed approaches that aim to obtain HLA haplotype information from whole exome/RNAseq next generation sequencing data, it will also be important to ensure that high resolution is achieved (34–36).

The HLA subtype tools developed in this study for HLA-A*02 should contribute to the development of the personalized immunomonitoring that is made possible by next generation sequencing of pathogens and tumors. As a first example, next generation sequencing of viral quasispecies will make it feasible to follow the dynamic relationship between the T cell repertoire and the viral quasispecies for viruses such as hepatitis C virus and HIV. As a second example, human tumors contain large numbers of mutations that lead to altered (non-self) peptide sequences that are unique to each patient. Because these sequences have not been encountered during T cell development (i.e., are “foreign”), it may be speculated that T cells recognizing these mutated epitopes are of particular importance in (therapy-induced) T cell–mediated tumor regression (37). Recent work by Rosenberg and colleagues (38) and by us (39) has demonstrated how cancer exome sequencing data can be used to reveal T cell responses against patient-specific neoantigens in humans, and how such responses can be influenced by immunotherapy (39). The development of pMHC-based monitoring technology for large sets of HLA subtypes will be of substantial importance for such personalized immunomonitoring, either by the flow cytometry–based approaches used in this study, or by the high-throughput MHC multimer–based mass cytometry recently developed by Newell and colleagues (40).

We thank Can Keşmir (Division of Theoretical Biology, Utrecht University, Utrecht, The Netherlands) for advice on pMHC predictions, Patrick H. Celie (Division of Biochemistry, The Netherlands Cancer Institute, Amsterdam, The Netherlands) for the generation of the three-dimensional molecular models of the subtype MHC molecules, Cécile Alanio and Matthew L. Albert (Centre d’Immunologie Humaine, Institut Pasteur, Paris, France) for critical reading of the manuscript, and members of the Schumacher and Haanen Laboratories for useful discussions.

The authors have no financial conflicts of interest.