|Home | About | Journals | Submit | Contact Us | Français|
Quantitating the frequency of T cell cross-reactivity to unrelated peptides is essential to understanding T cell responses in infectious and autoimmune diseases. Here we used 15 mouse or human CD8+ T cell clones (11 antiviral, 4 anti-self) in conjunction with a large library of defined synthetic peptides to examine nearly 30,000 TCR-peptide MHC class I interactions for cross-reactions. We identified a single cross-reaction consisting of an anti-self TCR recognizing a poxvirus peptide at relatively low sensitivity. We failed to identify any cross-reactions between the synthetic peptides in the panel and polyclonal CD8+ T cells raised to viral or alloantigens. These findings provide the best estimate to date of the frequency of T cell cross-reactivity to unrelated peptides (~1/30,000), explaining why cross-reactions between unrelated pathogens are infrequently encountered and providing a critical parameter for understanding the scope of self-tolerance.
Specificity is the central feature of immunity. Both innate and adaptive immune reactions are based on discriminating foreign from self molecules. In the case of the major class of CD8+ T cells, Ag discrimination results from the interaction of the TCR with MHC class I molecules bearing peptides typically between 8 and 11 residues in length. Since most activated CD8+ T effector cells are potent killers of cells perceived as foreign, self/non-self discrimination has immediate life and death consequences of obvious danger to the host (1).
For both CD8+ T cells and CD4+ T cells, self/non-self discrimination is based on relatively subtle differences between foreign and self Ags due to the inherent self-specific nature of TCR interaction with MHC class I and II molecules, respectively. The initial threshold for cross-recognition of peptide MHC complexes (pMHC)3 is set during thymic selection, which eliminates T cells that recognize self-pMHC with too high or low affinity. A low level of anti-self reactivity is maintained in mature T cells by peripheral positive and negative selection mechanisms (2).
A critical value for understanding adaptive immunity is the degree of cross-reactions between Ags. This is necessary for understanding how previous exposure to one Ag will influence the response to a novel Ag or a self-Ag that that does not effectively participate in self-tolerance mechanisms. For B cells, the antigenic universe is essentially unlimited, since as Ehrlich first recognized, Abs are fully capable of high-affinity binding to manmade Ags with no equivalent structures in nature. For CD8+ T cells, the antigenic universe is smaller, since recognition is based on discriminating small peptides of finite permutations bound to a set of MHC class I platforms. Given that an average-sized class I peptide ligand of nine residues can have any of the 20 common amino acids at each position, and ~1% of peptides will bind to any given class I allomorph, each TCR selected on a given class I allomorph can potentially interact with 209 × 10−2 (=5.12 × 109) different peptides (not including residues with unusual side chains or post-translational modifications to standard side chains). While this is a daunting number, the practical issue of cross-reactivity between Ags depends on the frequency of cross-reactions between defined Ags, which can (and must) be determined empirically.
A number of reports have addressed the degree of cross-reactivity of TCRs for different peptide Ags. It is important to distinguish, however, between studies that measure the frequencies of cross-reactions between genetically closely related Ags (3) (frel) (e.g., homologous peptides derived from viruses in the same family) vs unrelated Ags (fun) (given the universal origins of life, every Ag is related in the literal sense, and “unrelated” is a relative term). While frel is critical for understanding how infection with a pathogen will affect the subsequent response to a related pathogen (e.g., in the scenario of infections with distinct serotypes of a given virus), it provides limited insight (at best) into fun, which governs the cross-reactivity between unrelated pathogens.
Several elegant studies have used viral vectors or synthetic peptides to estimate fun (4–6). These studies have provided great insight into T cell cross-reactivity, and they have clearly demonstrated that fun is biologically significant. Each is hampered, however, to a various degree by uncertainties in the true functional size of the library used as the denominator in the equation where fun indicates the number of cross-reactions observed divided by the true number of peptides tested. These limitations arise from ambiguities regarding the abilities of peptides tested to either remain soluble in physiological solutions, bind to class I molecules, or compete for binding to class I molecules when tested in pools, or arrive at the cell surface with class I molecules when expressed genetically. Furthermore, studies with random peptides may not accurately reflect the universe of biologically relevant peptides due to evolutionary bias in host or pathogen amino acid sequences away from randomness.
To improve upon these approaches to measure fun, we have exploited large libraries of synthetic peptides generated to study CD8+ T cell recognition of viral and bacterial pathogens relevant to humans or mouse models of human infectious diseases (7, 8). The peptides are either predicted or, in the majority of cases, are shown to bind to a nominal mouse or human class I allomorph. We surveyed the libraries against a panel of mouse or human T cell clones of defined specificities or polyclonal mouse T cells induced by viral infections.
Thymoma cell lines EL-4 and TAP2-deficient RMA/s (H-2b) were maintained in RPMI 1640 containing 7.5% FCS. HLA-A2-restricted CD8+ T cell clones were cloned by limiting dilution of PBMCs from infected donors. Clones were maintained in RPMI with 10% human serum, antibiotics, l-glutamine, and, beginning the third day after restimulation, 200 U/ml IL-2 (R&D Systems) and were restimulated every 14–28 days using irradiated PBMCs at a 1:10 target-to-effector ratio. Clones recognized either the HIV subtype A gag p17 peptide, SLFNTVATL, the influenza M1 epitope, GILGFVFTL, or the CMV lower matrix protein pp65 peptide, NLVPMVATV.
Specific pathogen-free C57BL/6NTac mice (wild type), B6(Ly5.1)-[transgenic (Tg)]OT-I-[knockout (KO)]RAG1 mice (OT-I) specific for the H2-Kb-restricted egg OVA peptide, SIINFEKL, B6-[Tg]F5-[KO]RAG1 mice (F5) specific for the H2-Db influenza A virus NP peptide, ASNENMDAM, and B6-[Tg]MataHari-[KO]RAG1 (MataHari) mice specific for the male Ag, WMHHNMDLI, were from Taconic-National Institutes of Allergy and Infectious Disease colony; B6-[Tg]B16-[KO]RAG1 mice (B16) specific for the human and murine melanoma gp100 peptides, ITDQVPFSV and IMDQVPFSV, were from Taconic. For in vivo influenza infection, 1 ml of infectious PR8 allantoic fluid diluted 1/5 was injected into the peritoneum. For allo-specific responses, >5 × 105 purified BALB/c lymphocytes were injected into the peritoneum of C57BL/6 mice. Memory responses were generated over the course of >6 wk. All procedures involving animals were approved by the Animal Care and Use Committee of the National Institute of Allergy and Infectious Diseases.
Flow cytometry was performed using an LSR II (BD Biosciences). CD69 assays were performed by incubating 5000, 10,000, or 25,000 EL-4 cells overnight with synthetic peptides at various concentrations and 50,000 cells from B6-[Tg]TCR-[KO]RAG1 mice. The following day, cells were stained simultaneously for 30 min on ice with Pacific Blue-conjugated anti-mouse CD8α Ab (eBioscience) and PE-conjugated anti-mouse CD69 Ab (eBioscience), and propidium iodide was added to a final concentration of 1 μM to stain for dead cells.
Five thousand, 10,000 or 25,000 EL-4 cells were incubated for 2 h with exogenous synthetic peptides and then incubated for 48 h with 50,000 cells prepared from secondary lymphoid organs of B6-[Tg]TCR-[KO]RAG1 mice. Next, 0.25 miCi of radioactive [3H]thymidine was added to each well. The following day, the cells were harvested onto filtermats using a FilterMate Harvester (PerkinElmer) and radioactivity was assessed by beta-counting using a 1450 TriLux Microbeta Counter (PerkinElmer). [3H]Thymidine corresponded with cell proliferation.
Peptides predicted to bind MHC class I (H2-Kb, H2-Db, or HLA-A2) were identified from the predicted open reading frames (ORFs) of pathogen sequences using a previously described algorithm (9, 10). Peptides were synthesized by A&A Labs at the John Curtin School of Medical Research Biomolecular Resources Facility or purchased as crude material (Mimetopes/Pepscan Systems). Peptides were solubilized in DMSO and diluted in PBS or RPMI 1640 to their working concentrations. Binding of peptides to MHC was quantified in a substantial portion of peptides based on the inhibition of a standard, radiolabeled peptide as described previously (7, 8).
One hundred thousand of the T cell clone to be typed were incubated with 0.5 μl of anti-hCD8 allophycocyanin-conjugated Ab and 2 μl of 1 of 20 Vβ-specific, FITC- or PE-conjugated Abs (Beckman Coulter). Ab specificities included: Vβ1, Vβ2, Vβ3, Vβ4, Vβ5.1, Vβ5.2, Vβ7.2, Vβ7.7, Vβ8, Vβ11, Vβ12, Vβ13.1, Vβ13.2, Vβ13.6, Vβ14, Vβ16, Vβ17, Vβ20, Vβ21.3, and Vβ22. Cells were then washed in PBS and fixed with 1% formalin for 20 min. Cells were then washed twice more in PBS, resuspended in 250 μl, and acquired using a FACSCalibur flow cytometer (BD Biosciences).
RMA/S cells were incubated overnight at 27°C with 5 μg/ml human β2-microglobulin to increase and stabilize expression of class I MHC. Dilutions of peptide were then incubated with RMA/S cells in a 96-well plate for 1 h at room temperature before being shifted to 37°C for 2 h with the addition of brefeldin A at 10 μg/ml. Cells were then washed to remove excess peptide and stained with the anti-Db Ab B22-249 and counterstained with FITC fluorescent donkey anti-mouse IgG (Jackson ImmunoResearch Laboratories) and analyzed using the LSR II.
For anti-influenza A virus (IAV) responses splenocytes were restimulated in vitro with either PR8-infected or mock-infected EL-4 cells or 100 nM peptide for 5 h at 37°C. After 2 h of incubation, brefelden A was added at a final concentration of 10 μg/ml, and the cells were incubated for a further 3 h to accumulate IFN-γ in the endoplasmic reticulum of activated cells. Cells were stained with ethidium monoazide (to identify live cells) and anti-CD8α Alexa 488 (BD Pharmingen). Cells were fixed with 3.2% paraformaldehyde and stained with anti-IFN-γ Alexa 647 (BD Pharmingen) in 0.2% saponin (Sigma-Aldrich). Sample fluorescence was measured using an LSR II flow cytometer (BD Biosciences) and analyzed with FlowJo software (Tree Star).
ELISPOT plates (Millipore) were coated with anti-human IFN-αB (Mabtech) overnight at 4°C. The plates were washed six times with PBS/0.05% Tween 20 and incubated for 1 h with R10 at room temperature. Fifty microliters of HLA-A2 B cells at a concentration of 2 × 105 cells/ml was added to each well, along with peptide mixtures at a final concentration of 2 μg/ml/peptide. Next, 50 μl of a mixture of either three or four CTL clones with each clone at a concentration of 4 × 103 or 6 × 103 cells/ml was added to a final volume of 100 μl. The plates were incubated overnight at 37°C. After washing the plates, a second biotinylated Ab to human IFN-γ was added and incubated at room temperature for 2 h. After additional washes, the plates were developed with streptavidin-alkaline phosphatase (Mabtech) and colorimetric substrate. Spot-forming units (SFU) were counted using an automated ELISPOT reader (Autoimmun Diagnos-tika). Background counts for negative control wells without peptides were always less than five spots per well. Results were expressed as percentage of SFU compared with an average of the relevant index peptides.
Splenocytes from mice primed with PR8 or allogeneic splenocytes were lymphocytes were extracted from three to four memory mice and pooled in culture. Cultured cells were restimulated with purified lymphocytes from the spleen of either a PR8-infected or a BALB/c mouse and left for 4 days. Following restimulation, the culture was Ficolled to remove dead cells and incubated with 10 ng/ml IL-2 for 3 further days to rest the cells. Finally, cells were Ficolled again and used for peptide surveys.
To produce memory-like phenotypes in culture, transgenic T cells from three mice were pooled and exposed to a fraction of their number (between one-third and one-fourth of their number), which had been prepulsed with agonist peptide. After 5 days of stimulation, the cells were Ficolled and rested in 10 ng/ml IL-2 for 3 days and in 10 U/ml IL-7 for a week. The T cells were Ficolled again and displayed memory phenotypes, including up-regulated CD62L and CD44.
The synthetic peptides correspond to known or predicted ORFs in viruses and bacteria used in the present study, as listed in supplemental Table I.4 Peptides were selected based on predictive binding algorithms for Kb, Db, or HLA-A*0201. Binding affinities were measured empirically for more than a third of Kb and Db peptides and more than two-thirds of the HLA-A*0201 peptides. Peptides with a Kd of ≤500 nM were considered to be high-affinity binders. Extrapolating the fraction of high-affinity binders observed for each pathogen enabled a good estimate of the total number of high-affinity binders for each allomorph (Table I). Of 1501, 1526, and 5993 peptides predicted to bind to Db, Kb, and HLA-A*0201, respectively, 310 (21%), 1065 (56%), and 2475 (41%) are likely to bind with significant affinity (IC50 of <500 nM).
This approach assumes that the predicted binders in the panel are representative of the general population of peptides capable of eliciting a T cell response. To test this assumption, we compared the amino acid composition all of the peptides (450) binding to HLA-A*0201 with an IC50 of <500 nM that are distinct from all bona fide HLA0*0201-restricted epitopes reported in the literature (according to the Immune Epitope Database). Supplemental Table II relates the frequency of each amino acid residue at each position in the peptide in both sets. The overall frequencies are tightly correlated (Pearson r = 0.92, p < 0.0001 for entire set), most prominently at the anchor positions. Thus, it appears that the peptides in our panel are not significantly skewed in composition from the relevant pathogen/self epitope repertoire.
To optimize our ability to detect mouse T cell cross-recognition of peptides, we compared several methods for measuring CD8+ T cell activation, including CD69 up-regulation (via flow cytometry), IFN-γ synthesis (via intracellular cytokine staining (ICS)), and proliferation ([3H]thymidine incorporation) (Fig. 1). Optimization experiments utilized OT-I TCR transgenic T cells, specific for Kb-SIINFEKL. Since all cross-reaction survey experiments described below utilized pools of 16 peptides, we mixed SIINFEKL at equimolar amounts with 15 peptides known to bind H2-Kb or H2-Db with high affinity. Peptides were presented by either 5,000, 10,000 or 25,000 H2-Kb/H2-Db-restricted EL-4 cells. Since CD69-staining displayed the greater sensitivity and independence of APC number (Fig. 1), we used this method to test mouse T cells for their activation by H2-Kb or H2-Db peptide libraries. Human T cell surveys were performed by ELISPOT due to the significantly larger size of the HLA-A*0201 library combined with the success of previous cross-reactivity studies using the ELISPOT method on several of the clones used in this study (11).
To efficiently study the interaction of 15 T cell clones with thousands of peptides, we tested the peptides in pools of 16. Using pooled peptides in this manner, it was essential to consider the potential effects of peptide competition for binding to class I molecules. At high peptide concentrations, the highest affinity non-agonist peptides might reduce the binding of agonist peptides to levels below the threshold of T cell activation. On the other hand, at low peptide concentrations, peptides might not reach a threshold activation level.
To test the effects of peptide competition, we measured activation of a human influenza virus M1 (58-66)-specific T cell clone vs its agonist peptide in the presence or absence of 15 competing peptides. Although the activation curve was shifted by 2 logs by competing peptides, ELISPOT easily detected activation via ELISPOT at submicromolar peptide concentrations (Fig. 2). We similarly examined the effect of competing peptides on OT-I cell activation by Ova257–264 as determined by CD69 expression. The presence of 15 competing peptides shifted the Ova257–264 dose response by only ~1 log.
Despite these encouraging findings, prudence dictated that we perform CD69 surveys using peptide pools at high (each peptide at 100 ng/ml) and low (1 ng/ml) concentrations. To ensure that agonist peptides could be detected under these conditions, positive control agonist peptides were dosed into pools at random during the surveying process, at a concentration equal to the concentration of each of the other 16 peptides. In every case, agonist peptides were detected by each surveying method employed. Thus, we are confident that the methods employed are capable of detecting bona fide cross-reactions between the T cell clones used and potential agonist peptides.
Using CD69 expression to measure CD8+ T cell activation, we tested the Kb- and Db-restricted pathogen-derived peptide library against four well-studied transgenic CD8+ T cells that cover a range of potential cross-reaction scenarios (Table II). One (F5) is specific for a viral peptide, two are specific for mouse-encoded peptides (MataHari, Pmel-1), and one (OT-I) recognizes a chicken peptide. OT-I T cells have a hair-trigger, exhibiting robust homeostatic proliferation and recognizing H2-Kb, H2-Db, and even class II molecules bound to agonist peptides (2). We expected that they would be most likely among our panel to demonstrate cross-reactions. At the other end of the spectrum, MataHari T cells recognize a (male) self-Ag and exhibit only weak homeostatic proliferation, with secondary lymphoid organs shrinking or disappearing entirely, as the majority of T cells are gradually deleted in female mice. The Db-restricted transgenic TCR clones were surveyed against the Db library, while OT-I was surveyed against both libraries, due to its ability to recognize Kb and Db peptide complexes.
Each CD69 up-regulation survey was repeated a minimum of eight times (four times at high and low peptide concentrations) for each transgenic T cell model in at least two separate experiments. Representative results are shown in Fig. 3. In surveying >2000 TCR class I peptide complex interactions (four clones multiplied by 310 predicted Db binders, one clone multiplied by 861 predicted Kb binders), we found a single cross-reaction: MataHari reproducibly recognized Db peptide pool 61. Further testing MataHari T cells against each of the pool peptides immediately identified the cross-reactive peptide, YILCNMALL.
The YILCNMALL peptide derives from a predicted VV ORF (B ORF D, Moss 240, B068, protein 54). This peptide shares essentially no sequence homology with HY peptide, WMHHNMDLI, beyond the common Db anchor residue at position 5. The critical parameter in assessing the physiological significance of T cell cross-reactions is the copy number of pMHC needed for CD8+ T cell activation. This value can be calculated from the peptide affinity for Db and the number of peptide-receptive molecules present on the APC surface. Peptide affinity is closely approximated by the concentration needed for half-maximal stabilization of peptide-receptive cell surface class I molecules. Using TAP-defective RMA/S cells we found that YIL binds to Db with a KD of 40 nM, while the agonist peptide WMH binds with a KD of 3 nM (Fig. 4A).
We next compared the agonist activities of two peptides based on MataHari CD69 up-regulation. This revealed that WMHHMD LI peptide achieves 50% maximal activation of MataHari T cells at a concentration 240-fold less than that required for the cross-reactive YILCNMALL peptide (0.025 vs 6 nM). Based on the affinities of the peptides and the presence of ~2.5 × 104 peptide-receptive Db molecules on the surface of EL-4 cells (12), we can calculate that half-maximal MataHari activation requires ~200 copies of the defined agonist vs 4000 copies of the cross-reactive VV peptide. Notably, the latter number is at the upper range of copy numbers of most defined immunogenic viral peptides, which typically range from <100 to several thousand copies per cell (13).
Thus, the lone cross-reaction observed in our surveys of four mouse transgenic TCR clones is mediated by a peptide of unrelated sequence whose activity requires numbers of class I peptide complex that are achieved by a small fraction of naturally processed viral peptides.
HLA-A*0201-specific peptides derived from HBV, HCV, SARSV, LCMV, VV, and IAV were surveyed against 10 well-characterized, immunodominant, virus-specific human CD8+ T cell clones, including a Vβ17-bearing clone representative of the anti-M1 (58-66) response to influenza, five clones representative of the anti-p17 Gag response in chronic HIV-1 infection, and four clones representative of the anti-pp65 response to CMV (Table III).
Although only three peptide specificities were represented among the human clones, each of the pp65 specific clones expresses a unique Vβ-chain, and even the Gag-specific clones expressing Vβ7.2 were derived from different individuals, making it nearly certain that the clones express unique TCRs.
As with the mouse surveys, we used pools of 16 peptides to detect cross-reactions. Each of the 10 clones was surveyed against 5993 predicted HLA-A*0201 peptides (including 2475 predicted binders) using IFN-γ ELISPOT to measure activation. Despite >24,000 potential TCR-pMHC interactions, we detected only infrequent, weak, and sporadic cross-reactivity with a few peptide pools (Fig. 5). These pools were titrated over a range of concentrations to test whether they represented partial cross-reactions or simply noise in the assay. Careful retesting failed to reveal any bona fide cross-reactions.
Using the same panel, we examined a human TCR transgenic TCD8+ clone from a mouse that expresses the 1G4 TCR specific for HLA-A2 complexed with a peptide derived from the human NY-ESO-1 protein (E. Shenderov et al., manuscript in preparation). To enable thymic selection, this transgenic model was generated using the B6 HHD mouse, which expresses a single-chain β2-microglobulin HLA-A2 (with the Kb α3-domain to enable interaction with mouse CD8) on a Dbo, β2-microglobulin−/− background (14). B6 mice are not known to express a gene product with a homologous peptide predicted to bind to HLA-A2 (the only close match provided by a protein BLAST search has Gly at the critical COOH-terminal anchor position). Naive 1G4 TCR TCD8+ failed to respond to any of the pools, as determined by CD69 up-regulation using autologous splenocytes as APCs (supplemental Fig. 1). In contrast, in all cases tested, spiking random pools with the superagonist peptide SLLMWITQV (9V), or the weak agonist peptide SLLMWITQC (9C), corresponding to the NY-ESO-1 sequence, elicited a clear positive response.
Typical antiviral CD8+ T responses in any given individual encompass measurable T cell responses to ~10 different viral peptides (15). If each antipeptide response includes ~250 clonotypes, then in a typical antiviral response (16) ~2500 clones are activated and could give rise to memory cells that potentially cross-react with the 10 immunogenic peptides expressed by future pathogens, resulting in ~25,000 potential TCR-pMHC cross-reactions between the two pathogens. Under these assumptions, the number of potential TCR-pMHC cross-reactions we have surveyed (~25,000) represents a substantial sampling of likely cross-reactions.
There are, however, a number of factors that could undermine the relevance of our approach. First, there may be highly biased distribution for cross-reactions among TCR clonotypes, with a small fraction of clones responsible for the bulk of cross-reactions (17). Second, significantly >10 determinants could expand populations of T cells of sufficient size to bias responses to unrelated future pathogens.
To address these concerns, we surveyed in vitro-expanded polyclonal mouse alloreactive and IAV-specific CD8+ T cell responses against the Kb and Db peptide libraries. IFN-γ ELISA of IAV peptide-stimulated cultures demonstrated Ag specificities proportional to those found in ex vivo immunodominance hierarchies, suggesting that the cultured cells accurately mimic the specificities of repertoires generated in vivo.
Surveys of both anti-IAV or B6 anti-BALB/c populations failed to reveal clear cross-reactions (supplemental Fig. 2). The few weak reactions we detected against pools were not confirmed in subsequent experiments. Importantly, there is a limitation to this experimental strategy. If cross-reactions are largely based on “private” specificities responses (3), this would severely compromise our ability to validate true cross-reactions between experiments, which of necessity, utilized different mice. Nevertheless, the broad consensus of these results with the previous experiment argues strongly for a very low frequency of cross-reactions.
It is widely thought (but poorly documented) that memory CD8+ T cells are more easily triggered than are naive CD8+ T cells. In this case, we may have missed bona fide cross-reactions in our peptide surveys by not using memory CD8+ T cells. To address this concern, we generated OT-I cells with a memory-like phenotype (up-regulated CD62L and CD44) by in vitro culture with Ag in media supplemented with IL-2 and IL-7. These “memory-like” cells were not, in fact, triggered by lower amounts of agonist peptide (data not shown). Surveying these cells against the Kb and Db libraries failed to reveal cross-reactive peptides. We also surveyed naive OT-I cells in the presence of 5 ng/ml LPS using bone marrow-derived dendritic cells rather than EL-4 cells as APCs (Fig. 6). Although LPS combined with dendritic cells increased the sensitivity of OT-Is to agonist peptide and partial agonist EIINFEKL by ~10-fold (supplemental Fig. 3), it increased the background noise of the surveys substantially and failed to reveal further bona fide cross-reactions.
These data support the relevance of our surveys using TCR T cells and human CD8+ T clones to determine the relevant value of fun in vivo in individuals with a typical sequential exposure to multiple pathogens.
The degree of T cell cross-reactive recognition of peptides is a critical parameter for understanding recognition of self and foreign peptides. Of particular interest are the effects of previous exposure to genetically unrelated organisms on the “naive” response to subsequent challenge organisms. It is well established that activated memory T cells specific for a virus can suppress the responses of naive T cells specific for the same virus (18). Thus, even relatively weak T cell responses to nondominant determinants during an initial infection can greatly modify the immunodominance hierarchy in T cell responses to subsequent infection by enhancing the response to subdominant cross-reactive determinants at the cost of responses to other determinants in the hierarchy.
Importantly, for a cross-reaction between determinants to be biologically relevant, it must be based on recognition of physiological levels of pMHC, which typically range from 10 to a few thousand copies per APC. Relevant levels are achieved in most in vitro applications using synthetic peptides corresponding to naturally processed peptides at concentrations of no more than 100 nM. In contrast, many examples of cross-reactions in the literature are based on recognition of APCs exposed to super-micromolar concentrations of peptides, for example, the cross-reaction reported by Anderson et al. (19) now known to be physiologically irrelevant (20).
T cell recognition of peptides closely related in sequence is no doubt the most common form of cross-reaction between pathogens. Typically, such sequence homology has a genetic basis in a close evolutionary relationship between the source organisms of the cross-reacting determinants. A practical and highly relevant example would be recurrent infection with IAVs, where CD8+ T cell responses to highly conserved peptides encoded by nonglycoprotein genes can be boosted on a yearly basis. While such cross-reactions are clearly relevant to human immunity to viruses, those like IAV have multiple serotypes that enable reinfection and they are irrelevant to the question at hand, namely the frequency of cross-reactions between viruses from different families.
TCR cross-reactions can be divided into two categories: “random” cross-reactions, where peptide ligands bear little or no clear sequence homology, and cross-reactions based on sequence homology. In the former case, ignorance of the governing chemical principles dictates that the frequency of cross-reactions must be determined empirically. In the latter case a “Fermi estimate” (http://en.wikipedia.org/wiki/Fermi_estimate) for the frequency of such cross-reactions between two distinct viruses can be made using the following approximate values (chosen to err toward over-estimation): (1) APCs express class I molecules associated with <100 different viral peptides in physiologically relevant quantities that induce TCD8+ responses (21). (2) Of peptides bound to a given class I molecule, TCR recognition is predominantly based on interaction with four residues typically oriented toward the TCR (22). (3) At each of these positions, recognition is lost by substitution with 16 of the 20 possible residues (11). Based on assumptions 2 and 3, the frequency of cross-reactions among any two presented peptides by a given class I molecule is calculated to be 1 in 164 = 1/65,500. Given assumption 1, the chance of a sequence-based cross-reaction between any two viruses is <[1 − (1 − 1/65,500)100] = 0.0015, or < 1/655.
In the present study we use libraries of peptides unrelated in sequence to the defined agonist peptides of 15 TCD8+ clones to show that frequency of “random” cross-reactions is, perhaps surprisingly, of a similar magnitude (~1/30,000) to the predicted frequency of cross-reactive recognition of peptides of similar sequences. Using the combined frequencies of cross-reactions to similar and random sequences (1/30,000 + 1/65,500 1/20,000), the estimated frequency of cross-reactions between any two unrelated viruses (with <100 determinants each) is ≤[1 − (1 − 1/20,000)100], or ≤1/200. How does this number compare with the number of reported cross-reactions in the literature?
While there are a number of published examples of cross-reactions between defined peptides from unrelated viruses, only a few have been shown to represent bona fide cross-recognition at physiological pMHC levels (23, 24). Surveys of human PBLs against extremely large peptide libraries from VV or CMV, both large DNA viruses with hundreds of gene products and dozens of defined immunogenic peptides, revealed only a few examples of peptide recognition in absence of serological evidence of prior infection with the homologous virus (7, 25). This is consistent with the conclusion that relevant human T cell cross-reaction between heterologous viruses is infrequent.
We can use the level of cross-reaction observed in these studies to generate another Fermi estimate of fun. Combining the two studies, cross-reactions were detected with ~10 of the class I-binding ligands surveyed in 31 naive individuals. If the assays used could detect ~1000 different CD8+ T cell specificities (0.1% of total CD8+ T cells, the limit of detection of the hCMV flow cytometry study, which is ~ 10-fold less sensitive than ELISPOT used for the VV study), and ~1000 peptides surveyed for each patient generated a sufficient number of class I complexes to exhibit agonist activity (higher than the likely number of ligands in the hCMV study and lower than the VV study), then fun 10/1000 peptides × 1000 specificities × 31 individuals or 1/300,000, which is within reasonable distance of our more empirical data.
Does a fun of 1/30,000 support a high degree of allorecognition by T cells, estimated to be on the order of 10−1 between rodents expressing different class I allomorphs (26)? Allorecognition is thought to be based largely on peptide-specific cross-reaction on an allomorphic MHC molecule (27). If every gene can provide peptides for allorecognition, we can make another Fermi estimate for the fraction of CD8+ T cells that recognize self peptide MHC class I allomorphs. Given ~30,000 genes of ~330 residues average length, 107 different peptides of any given length are potentially synthesized, of which ~20% (2 × 106) can be generated in immunogenic quantities by cellular proteases, and 3% (60,000) will bind with sufficient affinity to at least one of the three different class I molecules that can serve as restriction elements in inbred animals (13). Since gene expression at immunogenic levels is no doubt limited in APCs to much less than all 30,000 genes, the number of self peptides expressed in sufficient quantities to be antigenic is likely to be on the order of 6,000, which is consistent with our fun estimate of 1/30,000 and the published 10% alloreactivity rate.
In summary, in this study we provide an empiric determination of fun based on the use of a large panel of defined peptides whose binding to class I has either been established directly or estimated based on reasonably safe assumptions. Although further refinement of fun is needed, particularly with regard to using a wider variety of TCD8+ clones, our estimate is consistent with the known cross-reactivity of CD8+ T cells for viral and alloantigens, and it provides a value that should be of use to systems biologists interested in modeling T cell responses.
Glennys Reynoso provided outstanding technical assistance.
1This work was supported by the Division of Intramural Research, National Institute of Allergy and Infectious Diseases (to J.R.B. and J.W.Y.), by the Medical Research Council Human Immunology Unit (to T.D. and A.M.), and by Cancer Research UK (Grant C399/A2291) (to V.C.).
3Abbreviations used in this paper: pMHC, peptide MHC complex; frel, frequency of T cell cross-reactivity to closely related peptides; fun, frequency of T cell cross-reactivity to unrelated peptides; IAV, influenza A virus; ICS, intracellular cytokine staining; KO, knockout; ORF, open reading frame; SFU, spot-forming unit; Tg, transgenic; VV, vaccinia virus.
4The online version of this article contains supplemental material.
Disclosures: The authors have no financial conflicts of interest.