Identification of T cell antigens in the 21^st century, as difficult as ever

Abstract

Identifying antigens recognized by T cells is still challenging, particularly for innate like T cells that do not recognize peptides but small metabolites or lipids in the context of MHC-like molecules or see non-MHC restricted antigens. The fundamental reason for this situation is the low affinity of T cell receptors for their ligands coupled with a level of degeneracy that makes them bind to similar surfaces on antigen presenting cells. Herein we will describe non-exhaustively some of the methods that were used to identify peptide antigens and briefly mention the high throughput methods more recently proposed for that purpose. We will then present how the molecules recognized by innate like T cells (NKT, MAIT and γδ T cells) were discovered. We will show that serendipity was instrumental in many cases.

Introduction

In T cell biology, there are two diametrically opposed situations when it comes to determine specificity. The easy one is when the antigen is known and used to pull out T cells. An animal is immunized, the lymph nodes collected, and T cell reactivity examined by in vitro restimulation and then narrowed down to a unique MHC molecule and a particular peptide. Clones of T cells or hybridoma T cell can be produced in that process and T cell receptors sequenced. This type of examination led to the development and use of peptide-MHC tetramers. The second situation is the one that we will develop in this review and has frustrated generations of immunologists, have a T cell and no antigen. Antigen determination has been and is very challenging when the cell of interest is a canonical CD4 or CD8 T cell that recognizes a peptide in the context of an MHC molecule (peptide MHC complex or pMHC). However, this effort is herculean when it comes to non-conventional T cells of which some are not even MHC restricted, e.g. γδ T cells. The reasons for the challenges of antigen determination are twofold and paradoxical, T cells are exquisitely sensitive to activation while analytical biochemical methods are not. Over the past 30 years, immunologists have seen three waves of approaches to tackle antigen specificity. The first one was purely biochemical, find a cell that expresses the antigen, produce vast quantities of it, separate the extracts by chromatography, isolate peaks with bioactivity, repeat. The exercise was daunting, led to some resounding successes, but overall did not open the door of antigen isolation to all. The second approach came at the peak of the molecular biology days of immunology: expression cloning. Many flavors of this technique were developed but all are cumbersome, resource and hands-on consuming. The third approach was a revisitation of the first one and relied on mass spectrometry (MS) and two misconceptions, the theorized high sensitivity of MS, and the assumption that MS could replace all traditional biochemical methods. Here, it is important to properly position MS in the context of antigen determination. MS can be highly sensitive through variations such as multiple reaction monitoring, but it is otherwise not much more sensitive than silver staining or Western blot. The second issue is that the isolation of a mass by MS out of a bioactive fraction, does not lead to the straight chemical characterization of the molecule with that given mass. Steps of fragmentation, re-analysis, and synthesis of the compound for MS analysis and comparison with the original sample is necessary; all these steps require larger amount of material, a bottleneck to which many have succumbed.

In the true spirit of science and with an undying ability of bouncing from failure to failure without loss of enthusiasm, the field is now investing heavily in artificial intelligence-based techniques to attribute antigen specificity to T cells by modelling their T cell receptor structure and binding surfaces. In this case, the determination of the TCR sequence is the only requirement, but the approach cannot work for the full discovery of unknown non-peptidic and/or non-MHC-restricted antigens.

We will review some of the most important steps and progresses that have been made in the field of determining the antigen specificity of T cells.

The most classical scenario: definition of the specificity of pMHC restricted CD4 and CD8 T cells

1- Isolation of T cells:

To ask the question of specificity, one first needs a T cell. Our ability to isolate individual clonal T cells is not that ancient. T cell cloning from primary source was published in 1979¹ and succeeded because of its alloreactive nature that allowed restimulation with irradiated cells, and the use of T cell growth factor soup, the ancestor of interleukin 2. Asking for antigen specificity was not the concern at that time, and we do not believe that it has been answered yet for that first T cell also known as CTLL-2. An easier way of isolating T cells of single specificity came shortly thereafter with the immortalization of T cells by fusion with an immortal partner and the production of the well-known T cell hybridomas^2,3. The advantage of T cell hybridomas is that they do not need antigen-specific T cell stimulation for growth as primary T cell clones do, even if this requirement can be overcome by using T cell receptor complex ligands such anti-CD3 and anti-CD28 antibody⁴. T cell hybridomas like any other heterokaryon are genetically unstable and have been marred by the loss of critical components such as CD3 subunits when kept in culture for extended periods⁵. The contemporary approach to access clonal T cells is through single cell T cell receptor (TCR) αβ paired sequencing⁶ followed by re-expression in “avatar” T cells. The avatar can be a primary cell such a bone marrow progenitor if in vivo studies are necessary⁷, or more commonly an αβ TCR negative cell line, BW51.47 for mouse studies⁸ or Jurkat for human ones⁹. Re-expression is usually achieved by retroviral or lentiviral infection using either a bicistronic vector or a self-cleavable P2A sequence inserted between the α and β chain cDNAs^7,10.

2- Biochemical approaches to determine T cell specificity, the hard road.

So, in the mid-80s, immunologists could isolate single T cells but the nature of the ligand seen by T cells and the nature of the TCR itself were still in full discussion and it took a few years and the first structure of an MHC molecule^11,12 to settle all issues and focus the search of T cell specificity onto peptides. These early days were heroic and here we will relate two stories, one for CD8, one for CD4 T cells that both demonstrate the use of biochemistry to identify peptides and specificity. The first one was a tour de force and delivered the nature of the ligand that once bound to H-2L^d, triggers activation of the famous CD8 2C T cell^13–15. We are in the 1990s and peptide sequence means Edman degradation, and Edman degradation equals large quantities of material and hopefully no N-terminal blockage¹⁶. The front of the paper describes the characterization of the biological activity of cell extracts from tissues (spleen and thymus) and six cell lines separated by reverse phase HPLC and assayed in cell killing assays with the T cell clone. Knowing that 60 fractions were tested for each HPLC run, the amount of work for this initial characterization is simply stunning. The results showed multiple peaks of activity, each reasonably broad and not sufficient in quantity to proceed further. To narrow down the search, the next step was to purify the MHC molecule, elute the bound peptides, and repeat the HPLC/bioassay search. Comparing H-2L^d with H-2D^d bound peptides delivered a much cleaner profile but not enough material to proceed to the dreaded sequencing step. To achieve and complete the treat, 28.7 g of spleen (from 283 BALB/c mice) were used. After two rounds of C18 and three rounds of C8 chromatography the sequence of the isolated peptide was determined by Edman degradation for the first 6 residues, LSPFPF, while the next two were tentatively determined by mass spectrometry, DL/I, and confirmed for its ability to stimulate by chemical synthesis¹⁷. It has been 30 years since this magnificent achievement by H. Eisen and his team. The second example that we want to discuss is about a famous CD4 T cell, BDC2.5, that was isolated in 1990 by K. Haskins K, and M. McDuffie from the spleen of NOD mice and was diabetogenic upon transfer¹⁸ or transgenesis of its TCR¹⁹. This particular T cell was maintained as a primary T cell clone and required stimulation with freshly isolated islets from NOD mice to be maintained. As you may know, islet isolation is tedious, cumbersome, expensive, and labor intensive and keeping the clone alive was a challenge, knowing its sequence a priority. On the good side, it was known that the antigen was present in the pancreatic islets, on the bad side, the antigen was in the islet. A very experienced pair of hands will give you 300-500 islets per day or 3 to 5 x 10⁵ cells (~10³ cells per islet), two to three logs off what a spleen would provide and was illustrated above for the 2C TCR antigen. Therefore, the biochemistry became a nearly impossible task and could only reach the conclusion that the antigen was in the membranes of the granules of β cells²⁰. In the 1990s, this “quantity” bottleneck of classical biochemistry came up over and over again forcing the community to developing alternative approaches. The most successful one was the production of random peptide libraries, genetically encoded²¹ or chemically made^22,23, that were used to stimulate T cells and then refined to single sequences by successive rounds of screening. This world of “mimotopes” emerged from studies of the immunopeptidome^24–26 and pMHC structural studies^27,28 that defined two principles, anchors residues and the associated definition of down-facing residues (anchors) and up-facing residues (exposed to TCR). Relatively strict rules can be determined for most MHC class I molecules with primary and secondary anchors²⁹ and the construction of libraries which will have 2-3 anchors and only 5-6 potential up-facing residues, limiting the size of the library, therefore simplifying the screen accordingly. For MHC class II molecules, the situation is more complicated especially for the I-A/HLA-DQ family of molecules for which anchors are dispensable and mainly use side chains of the α helices for interaction with the peptide³⁰. Behind the construction of these libraries and the determination of mimotopes to interrogate specificity was the unyielding hope that somehow mimotopes would look very much like the natural peptide. This situation has been encountered but remains the exception^31,32. One of the reasons for this disappointing outcome is that the simple fact to choose nominally anchor residues changes T cell recognition to a substantial extent which has been documented in cancer and autoimmunity^33,34. As importantly, selecting up-facing residues one at a time with a screen that only looks for improved activation, ends up selecting peptides that are far from the natural sequence in which suboptimal residues are often encountered; in some respect, selecting mimotopes is very similar to the production of Altered Peptide Ligands³⁵. As a consequence, in mimotopes libraries, it is never a single peptide that is found but a large family of related members^22,23. This long digression in the world of mimotopes is very relevant to what happened to the determination of the specificity of BDC2.5 T cells. Indeed, while the BDC2.5 mimotopes allowed an easy way to stimulate the BDC2.5 T cell in vitro and in vivo, and the making of tetramers³⁶, they also initiated a search for the natural peptide. In all BDC2.5 mimotopes, the 4-residue N-terminus is very tolerant to substitutions while the 5 C-terminal amino acid are almost always identical, WARMD. Today, a BLAST search for that short sequence will provide hundreds of potential candidates, some of which will be stimulatory. In any case, in 2010 as the search for the BDC2.5 antigen was still ongoing, biochemistry pointed to the importance of chromogranin A as a potential source of antigen and a WSRMD sequence (residues 358-362) within its C-terminus was compatible with the mimotope motif³⁷. However, the stimulatory properties of native chromogranin A sequences were weak and variable from lab to lab, and the search continued for another 6 years until the discovery of potential “hybrid peptides” produced by the fusion of fragments of C-peptide and fragments of chromogranin A³⁸. We will not enter the still controversial debate about this BDC2.5 peptide and the related one about hybrid peptides^39–41 but it is important to learn from that 25-year search for one peptide: biochemistry is very difficult and limited by accessible quantities of material, following the track of mimotopes is paved with dangers as we optimize anchor residues in mimotopes, and the identification of naturally processed peptides by mass spectrometry is an approach in which we put too much faith without knowing some of the pitfalls^40,41 and the current limitations of the existing databases.

3- One step beyond mimotopes

As the biochemical approaches remained so challenging, alternative methods of addressing antigen specificity were developed. J. Kappler following on his tethering of peptides to the C-terminus of the β chain of MHC class II molecules⁴² was the first one to propose libraries of pMHC complexes first for class I⁴³, then for class II⁴⁴, that enabled the screen with soluble versions of a given TCR. The approach was successful but the need to produce soluble versions of every TCR to be tested remained cumbersome and did not become a popular approach. To alleviate some of these issues, the same concept was developed in yeast with “platform pMHC’ molecules complexed with libraries of tethered peptides⁴⁵. In both instances, three issues are persistent: 1- the necessity of using linkers to tether the peptide to the MHC. This exercise requires a substantial mutagenesis of the C-terminal part of the of MHC class I molecule to allow the peptide to land into the groove, and it might be consequential for some T cells. Similarly, the linker leaves out of consideration for specificity all C-terminal flanking residues for MHC class II molecules, and those are sometimes critical for T cell recognition^46,47. 2- the quality of display of the pMHC is largely dependent on the inherent stability of the molecule being used. Some MHC class II are instable and difficult to display⁴⁸. 3- the peptide libraries to be displayed are limited in size and nature. Post-translational modifications or spliced/hybrid peptides will not be displayed. While the approach has certain advantages over the classic mimotope one, it remains with similar limitations that have slowed its embrace by the immunology community.

4- Expression cloning

Before the days of Crispr/Cas9 knockout T cell antigen specificity was often examined by surveying cell lines that would stimulate or not a particular T cell of interest. The complementation of a non-stimulatory cell line with the appropriate cDNA offers the possibility of seeing expression, processing in the appropriate subcellular compartments and presentation on the restricting MHC molecule. This type of approach has been quite successful for the identification of microbial antigens^49,50 and in the early days of cancer antigen discovery^51–53 leading to the discovery of the MAGE series of antigens in melanoma, and NYESO-1 in multiple cancers. However, the approach remained difficult and poorly amenable for the identification of unknown self-antigens that require genome wide screening until recently. Indeed, a new technique called T-Scan has shown great potential by adding to the retroviral expression of antigens in presenting cells, a reporter detection system based on the cleavage and activation of a GFP bipartite molecule upon delivery and cleavage by granzyme B⁵⁴. Labeled cells are subsequently isolated by flow cytometry and the antigen identified by next generation sequencing. This technique, although limited to CD8 T cells should have great applications in cancer biology next to other approaches amenable when a small number of neoantigens needs to be tested against a small number of T cells⁵⁵.

Today, any of these cloning techniques meant at identifying tumor antigens or autoantigens can be complemented by CRISPR/Cas9 genome wide screens that flip the basic principle of expression cloning and instead examine the loss of a single gene whose product is antigenic.

5- Screening with pMHC tetramers

In a much simpler scenario, sometimes the antigen is known first, and specific T cells need to be identified second. Here the simplest approach is to develop pMHC tetramers with overlapping peptides from the protein, isolate specific T cells from the polyclonal population, sequence individual TCR pairs from single cells, and confirm specificity after re-expression of the isolated TCR in a surrogate T cell^56,57. For CD4 T cells this type of approach is still very limited given the difficulty to produce pMHC class II tetramers. For CD8 T cells, pMHC class I tetramers are more suited to the task as they can be produced from a generic stock using photocleavable⁵⁸ or temperature mediated⁵⁹ peptide exchange and loading. These two approaches work reasonably well and allow schemes of multiplexing with standard fluorophores or metal-labeled secondary reagents⁵⁶ for 20 to 100 different specificities^56,57. A first enrichment step using magnetic beads recognizing a fluorochrome present in the tetramer can be used before cell sorting if large numbers of CD8 T cells are available as provided by leukapheresis⁶⁰. However, regardless of the elegance of the approaches, pMHC tetramers have not offered the universal key to screen T cells for specificity. The cited papers have been published a decade ago and few laboratories have had the knowledge and resources to follow up on them. Recombinant protein expression, peptide synthesis, parallel production of hundreds of reagents, quality control, polymorphism of MHC molecules, are some of the main hurdles that anyone will face. The possibility to increase the tetramer combinatorial to much higher levels with DNA barcodes will not alleviate these fundamental inherent issues of tetramers^61,62. Furthermore, when the frequency of specific T cells is very low (10⁻⁵ for instance for some circulating T cells recognizing tumor-antigen in the blood⁶⁰) it should always be kept in mind that even highly specific tetramers will be fraught with high background (e.g. at frequencies of 10⁻⁵ specific T cells and a reagent that stain with an efficiency of 99.99 %, a false positive event every 10 000 cells will result in a sorted population that is 10% specific and 90% non-specific T cells). In that scenario, specificity of the tetramer-positive sorted cells should always be checked using an orthogonal method such as TCR cloning and re-expression followed by functional assays.

6- Screening with MHC-TCR chimeric reporter molecules

Recently, a very astute approach was developed for peptide screening for CD4 T cells, based on a design inspired by CAR T cells⁶³: Here, the extracellular part of the TCR is substituted by an MHC class II molecule displaying a tethered peptide and expressed in a TCR negative T cell with an NFAT reporting construct. Upon recognition of the pMHC by the T cell of interest, the NFAT reporter is activated and the antigen presenting cell sorted by flow cytometry. A similar technique called SABR was developed at about the same time for CD8 T cells, based on pMHC equipped with an intracellular CD28-CD3ζ signaling module and a NFAT-GFP reporting system⁶⁴. This spring, a sophisticated method called RAPTR (Receptor–antigen pairing by targeted retroviruses) was reported and designed around an entry-deficient VSV-pseudotyped retrovirus⁶⁵. In this approach, the deficient virus is rescued by co-expression of a membrane-displayed receptor such as pMHC. Fluorescence upon cell entry and expression of the viral cargo allows cell sorting and isolation while pMHC can be barcoded and produced as libraries to allow a large screen.

These techniques have limitations similar to the ones mentioned for pMHC tetramers and pMHC display: random peptide libraries are extremely challenging to display (low affinity of peptides, size of peptides, lack of optimization of the linker for each peptide), poor expression of some MHC molecules, and presence of a linker on the C-terminus of the peptide. However, given the current difficulties to determine the antigen specificity of CD4 T cells, these techniques offer a substantial step forward.

7- A step into the future: computational prediction of antigen specificity

As we move more and more into artificial intelligence-based technologies in life sciences, the field of computational biology has joined forces with immunologists to try to crack one and for all the problem of antigen specificity determination for T cells. There is currently more review than experimental papers on the subject, but the principles deserve to be discussed. The structural determinants of antigen specificity that can be analyzed computationally come from three sources, 3D structures of pMHC, TCR, and pMHC/TCR complexes, TCR sequences of pMHC tetramer-sorted cells, paired or unpaired, and mutagenesis of TCRs, MHCs, and peptides within a given specificity. There is currently no methodology capable of computing all these parameters together. The simplest approaches like TCellMatch⁶⁶ will analyze only CDR3 sequences. Others will add the layer of TCR αβ pair sequencing⁶⁷ and gain in accuracy as will mixed models such as RACER in which both sequence and structural information are combined in a pairwise energy model⁶⁸. Another complex model called ERGO collapses TCR and MHC specificity on the peptide sequence itself⁶⁹. Other methods are purely statistical and: use available large datasets of TCRs specific for a given peptide-MHC combination, to assign a tentative antigen specificity based on homology^70,71. While all these approaches claim some degree of success, they all face identical issues. The main one, common to every modelling approach is the size and quality of the various databases we can mine. If TCR sequences are in large number, fewer provide pairing, and fewer have a good validation of antigen specificity. Even more limiting is the structural information that we can exploit for MHC, peptide binding to MHC, TCR, and TCR-pMHC complexes which all together are in the low thousands. An additional issue is to assign specificity to CD4 or CD8 TCRs and one would expect a large overlap between the two repertoires⁶⁷. Finally, like in the case of mimotopes, computational searches will deliver lists of potential related candidates⁷² not a single answer. Here, we must be modest and patient and not be blinded by the word artificial intelligence. AlphaFold for the prediction of structures was successful because it gathered all structural information to this day, ~200,000 of them, and that this information is of high quality⁷³. For TCR specificity the complexity is much higher with three intricate levels that combine MHC structure, peptide structure and binding, TCR structure and conformational changes upon binding. Finally, the prediction programs for peptide binding to MHC molecules are still very underperforming, especially for class II isotypes^74–78.

8- The unique situation of non-conventional T cells; more difficult than difficult.

We will finish this review like we started it with anecdotes and stories that illustrates the hurdles that one faces when attempting to identify the nature of a T cell antigen. Because of their nature and functions, innate T lymphocytes add to the complexity of finding their antigens for at least three reasons, some are not restricted to MHC or MHC-like molecules (γδ T cells), some are restricted but do not bind peptide ligands (NKT, MAIT cells), and some bind ligands that have been chemically modified (MAIT cells)(Table 1). Thus, most of the approaches that we have discussed are not amenable to finding ligands for innate T lymphocytes.

Table 1:

Some of the most important references that have led to the identification of exogenous and endogenous ligands for innate T lymphocytes. This list is not exhaustive, and we apologize to the authors of some of the many references that could have been listed.

Cell type	Some direct links to important references
NKT T cells
	Discovery of αGalalctosylceramide: Natori et al., 1994, DOI: 10.1016/S0040-4020(01)86991-X Morita et al, 1995, DOI: 10.1021/jm00012a018 Kawano et al., 1997, DOI: 10.1126/science.278.5343.1626 Exogenous versus endogenous ligands : Mattner et al., 2005, DOI: 10.1038/nature03408 Early search for a ligand: Stanic et al., 2003, DOI: 10.1073/pnas.0430327100 Facciotti et al., 2012, DOI: 10.1038/ni.2245 Endogenous ligands, anomers and contamination: Zhou et al., 2004, DOI: 10.1126/science.1103440 Brennan et al., 2011, DOI: 10.1038/ni.2143 Brennan et al., 2014, DOI: 10.1073/pnas.1415357111 Kain et al., 2014, DOI: 10.1016/j.immuni.2014.08.017
γδ cells
	MHC molecules as ligands: Bonneville et al., 1989, DOI: 10.1073/pnas.86.15.5928 Schild et al., 1994, DOI: 10.1016/0092-8674(94)90170-8 Rice et al., 2021, DOI: 10.1073/pnas.2110288118 Phospho-antigens, butyrophilins and stress: Fournié and Bonneville, 1996, DOI: 10.1016/0923-2494(96)89648-9 Morita et al., 1995, DOI: 10.1016/1074-7613(95)90178-7 Rigau et al., 2020, DOI: 10.1126/science.aay5516 O’Brien et al, 1991, DOI: 10.1111/j.1600-065x.1991.tb00827.x Microbial antigens: Zeng et al., 2012, DOI: 10.1016/j.immuni.2012.06.011
MAIT cells
	The ligand is of microbial origin: Le Bourhis et al., 2010, DOI: 10.1038/ni.1890 Gold et al., 2010, DOI: 10.1371/journal.pbio.1000407 The ligand is a derivative of the Rib pathway: Kjer-Nielsen et al., 2012, DOI: 10.1038/nature11605 Corbett et al., 2014, DOI: 10.1038/nature13160 Soudais et al., 2015, DOI: 10.4049/jimmunol.1403224 Alternative ligands Keller et al., 2017, DOI: 10.1038/ni.3679 Mak et al., 2017, DOI: 10.1038/ncomms14599 Salio et al., 2020, DOI: 10.1073/pnas.2003136117

NKT cells

Semi-invariant NKT cells were identified 30 years ago and remained difficult to study for a decade because the MHC restriction element was unknown and non-classical, and the ligand even more elusive⁷⁹. The restriction element, CD1d, a non-MHC encoded MHC-like molecule was discovered first⁸⁰ and studied like a classic MHC molecule, leading to the identification of peptides binding to it⁸¹. The real story came to light in two successive papers, one identifying a natural product from a marine sponge capable of binding to CD1d and stimulatory of NKT cells⁸⁰, and one detailing the structure of CD1d and its uniquely hydrophobic structure capable to accommodate acyl chains ⁸². The ligand was α-galactosylceramide, a glycosphingolipid, apparently unique to the prokaryotic world because of the α anomeric linkage of its sugar. From this day, the search for an endogenous ligand was on and focused exclusively on β anomeric glycolipids, the only ones known to be synthesized by mammalian cells. The physical nature of this ligand, a lipid, limited greatly all biochemical approaches that had been used for classic MHC molecules and required the use of detergents in the first step of isolation. The engineering of surface protease-cleavable CD1 molecules was elegant but unfruitful⁸³. Instead, most of the field went for a genetic approach and a screen of cell lines or animals deficient in particular pathways of glycolipid synthesis. Over the next 10 years, almost all of the three dozen enzymes of these pathways were examined; the quest delivered some information. In animals, many deficiencies resulted in gross dyslipidemia and the complete disappearance of all NKT cells because of a lysosomal storage disease⁸⁴; others allowed the identification of ligands such as isoglobotrihexosylceramide which are stimulatory of NKT cells⁸⁵ but not conserved through species⁸⁶. Other studies pointed towards very common ligands such as β glucosylceramide⁸⁷ but the synthetic product was not stimulatory unless contaminated by α anomers^88,89. In the end, using an antibody specific of the CD1d-α-galactosylceramide complex, our group demonstrated that α anomers of glycolipids were produced in small quantities in mammalian cells and were the likely endogenous ligands for NKT cells^89,90.

MAIT cells

The identification of the MAIT cell ligand followed a very similar trajectory. Discovered in the same era as NKT cells, MAIT cells are restricted on the MHC-like molecule MR1 ⁹¹. For the next 12 years after this assignment of restriction, researchers in the field tried to elute material from MR1 molecules using classical biochemical approaches. This search faced two difficulties, MR1 is expressed by cells at a very low level, and the production of its recombinant form is also very low yield. The first hint at the antigen specificity of MAIT cells came from the discovery of their reactivity against a broad range of microbes and a heat-sensitive soluble bacterial fraction ⁹² . The nature of this molecule was determined by the groups of McCluskey and Rossjohn⁹³ who exploited the refolding reaction of recombinant MR1 in the presence of filtered supernatant from S. Typhimurium. Indeed, the presence of the soluble fraction of many bacterial cultures but also from plain culture medium is sufficient to enhance MR1 refolding. 6-formyl-pterin (derived by photodegradation from the folates present in the media) was the first compound to be identified as bound to MR1. Then, based on this pterin structure, lumazines were identified and shown to be MAIT agonists⁹⁴. Further work identified a stronger MAIT agonist as a bipartite molecule made of a precursor of the vitamin B2 of bacterial origin, and a small endogenous dialdehyde, glyoxal or methylglyoxal⁹³. It is still unclear whether glyoxal or methylglyoxal derive from the bacteria or the eukaryotic cells. No genetic or computational method could have achieved this ligand identification. The search for alternative antigens for MAIT cells has been going on for the past few years with little success. Some non-stimulatory ligands have been isolated⁹⁵ as well as a small number of weak agonists⁹⁶, none of which appear to support important biological functions.

The important lesson from MAIT cells and MR1 is that for small molecule ligand identification, one is trapped in biochemistry and metabolomics approaches.

γδ T cells ligands

Very few γδ T cells are restricted on MHC molecules, but some are such as the ones seeing sulfatide in the context of CD1^97,98. However, in most instances γδ T cells will bind to MHC and MHC-like molecules in an unorthodox manner and no similarity with αβ TCRs. The series is rather long but each ligand is unique to a small subset of γδ cells and each sees the MHC molecule differently: T10/T22 reacts to the G8 TCR⁹⁹, I-E^k is seen by multiple γδ T cells¹⁰⁰, HLA-24¹⁰¹, HLA-B27¹⁰², HLA- A2¹⁰³ and HLA-DR¹⁰⁴ are all seen by subsets of humans γδ T cells. None of these γδ T cell ligands have been identified using similar approach, the latest, the HLA-DR antigen was identified for Vγ3Vδ1 T cells by CRISPR/Cas9 screen ¹⁰⁴. Next to these MHC ligands, γδ T cells can also see small molecules like MAIT cells do. The best documented family of those is for the large Vγ9JP/Vδ2 family of human T cells and consists of phospho-antigens such as (E)-4-hydroxy-3-methyl-but-2-enyl pyrophosphate that were first isolated from Mycobacterium using biochemical methods¹⁰⁵. However, this scenario is too simple for an innate T lymphocyte as complex as a γδ T cell. While the phospho-antigen moiety is necessary for stimulation with phospho-antigens, cell-cell contacts must be established and appear to be dominated by the butyrophilin family of molecules amongst which the protein butyrophilin 3A1 (which was discovered by knock down and knockout in cell lines^106,107). To make the story better, other butyrophilins such as BTN2A1 have been identified as necessary for stimulation with phospho-antigen^108,109. The isolation of BTN2A1 was very long and complex and achieved using very different means, namely tetramerized soluble TCR staining of target cells and a genome-wide CRISPR screening for the first group¹⁰⁸, radiation hybrids of Chinese Hamster Ovarian cells containing human chromosome 6 and screening for abrogation of phosphor-antigen reactivity for the other one¹⁰⁹. A similar situation, e.g. a very complex antigen, has been revealed for the main murine skin Vγ5Vδ1 T cell (DETC) with the isolation of skint-1 as a necessary adhesion molecules for the selection of DETC but an unlikely antigen that binds TCR^110,111. A large collection of other γδ T cell ligands have been described for discrete populations¹¹² and in none of these examples a unique strategy can be defined, illustrating well the difficile art of the isolation of T cell antigens.

Conclusion

Facing the explosion of TCR sequences that are deposited in public databases, the determination of their specificity will become more and more of an issue. While high throughput computational methods are the only viable solution, they will take time to develop and validate. It is likely that for the next decade, each of us will have to rely on the difficult and painstaking methods that we have briefly surveyed; this situation is especially true for individuals focused on innate T lymphocytes.

Figure 1:

Schematic description of the four main physical approaches to identify antigens for T cells. A- Biochemical methods of isolation of the antigen. A succession of column chromatography runs followed by functional testing of each fraction leads to the isolation of a “pure” fraction that is used for sequencing. B- Screening with cells expressing or not the antigen. Removal or addition of particular genes will lead to the loss or appearance of T cell activation, respectively. C- Screening with re-engineered retroviruses that tag unique T cells. VSV-pseudotyped retroviruses are engineered to express an entry-deficient VSV-G protein that is rescued by the co-expression of surface pMHC. Infected cells are isolated and sequenced. D- Screening with overlapping peptides and libraries of pMHC tetramers.