Human thymopoiesis produces polyspecific CD8+ α/β T cells responding to multiple viral antigens

Article activity feed

1. 

   Version published to 10.7554/elife.81274 on eLife

   Mar 30, 2023
2. 

   eLife

   Dec 2, 2022

   Author Response

   Reviewer #1 (Public Review):

   In this paper, Quiniou and colleagues show via orthogonal methods that human thymopoiesis releases a large population of CD8+ T cells harboring a/b paired TCRs that (i) have high generation probabilities, (ii) have a preferential usage of some V and J genes, (iii) are shared between individuals and (iv) can each recognize and be activated by multiple unrelated viral peptides, notably from EBV, CMV and influenza.

   Major strengths of the paper

   Quiniou et al. generated single-cell sequencing datasets of the earliest stages of TCR beta chain gene recombination. And then showed that a subset of them is highly clustered also having high generation probability.

   They show that these T cells can bind multiple antigens, both via the use of public antigen-specific datasets as well as …

   Author Response

   Reviewer #1 (Public Review):

   In this paper, Quiniou and colleagues show via orthogonal methods that human thymopoiesis releases a large population of CD8+ T cells harboring a/b paired TCRs that (i) have high generation probabilities, (ii) have a preferential usage of some V and J genes, (iii) are shared between individuals and (iv) can each recognize and be activated by multiple unrelated viral peptides, notably from EBV, CMV and influenza.

   Major strengths of the paper

   Quiniou et al. generated single-cell sequencing datasets of the earliest stages of TCR beta chain gene recombination. And then showed that a subset of them is highly clustered also having high generation probability.

   They show that these T cells can bind multiple antigens, both via the use of public antigen-specific datasets as well as corroborating experimental TCR expression and binding essays.

   Minor weaknesses

   To what extent is TCR clustering and high Pgen and cross-individual sharing correlated? What is the Pgen of the sequences clustered with the high Pgen cells? Can you comment on the correlation between these three phenomena?

   Indeed, there is a significant positive correlation between the Pgen and the number of connections among the clustered TCRs, as was reported in Fig.1F of the original manuscript. Furthermore, this correlation is true for both private and public TCRs, as was reported in figure 2B of the original manuscript.

   To show the link between the three phenomena, we now have added two supplementary figures showing a high positive correlation between Pgen and the number of connections, and between cross-individual sharing and the number of connections, and to a lesser extent between Pgen and cross-individual sharing (Figure 2-figure supplement 4C and D in the manuscript supplementary information).

   However, we would like to emphasize that the difference in the mean Pgen of the clustered and dispersed TCRs is of about 20-fold. This is a high difference for a biological process (and highly statistically significant), but a small one compared to the 10-log10 span of the Pgens of the two populations. Factually, what we observed is not that clustered sequences have a high Pgen, but that they have a higher Pgen than the non-clustered sequences. Yet, many CDR3s with high Pgen do not cluster, and vice versa, indicating that a high Pgen is not the only (nor most important) driver of clustering. We have now added these as Figure 1-figure supplement 3E-F of our revised manuscript.

   In other words, to what extent is this surprising to see that highly clustered TCRs have higher Pgen and are more shared?

   That for a given CDR3 there is a correlation between having a high Pgen and being public is not surprising as both suggest a positive selection during evolution. What is more surprising is that there are CDR3s forming large clusters that occupy over 20% of the repertoire and that co-cluster between individuals with different HLA, “indicating a convergence of specificities between individuals’ clustered repertoires”. This suggests a surprising selection process that could depend less on HLA than the “classical” selection.

   These points are now better emphasized in the revised manuscript.

   Potential Impact of the paper

   This work highlights an intrinsic property of the adaptive immune response: to generate TCRs with high generation probability that can efficiently bind multiple antigens. This finding has, therefore important impact on drug discovery and vaccine design.

   We thank the reviewer for his appreciation.

   Reviewer #2 (Public Review):

   This study analyses the T cell receptor (TCR) repertoire of double positive human thymocytes, and compares this to mature single positive CD8 cells. The first major finding is that the repertoire post-selection is enriched for groups of TCRs with high generation probabilitites, similar sequences, and for TCRs previously annotated for viral specificity. This data is clearly presented and convincing. The extent of analysis of the human thymocyte repertoire is still very limited, and the paper adds significantly to this important question.

   We thank the reviewer for his appreciation.

   The second major finding is much more controversial. The authors first investigate the publicly available databases and show that there is a substantial proportion of TCRs which have been annotated to multiple viral specificities, a fact which is well-known to the specialists in the field, but not previously addressed.

   Indeed, we are not aware of reports disclosing “a substantial proportion of TCRs which have been annotated to multiple viral specificities”. Actually, one could wonder why “a fact which is well-known to the specialists in the field” is not mentioned and discussed in published articles? To us, it reveals that this point has been overlooked by immunologists as recently in Zhang et al, 2021 where authors aiming at identifying highly specific T cell clones with a new modelling approach, excluded all clones binding more than 1 peptide. Thus, it makes it important to report it, as we do. Furthermore, we would also like to emphasize that we do more than just reporting that some TCR have “been annotated to multiple viral specificities”. We show from a manual curation of public databases that (i) some TCR have been reported to bind to tetramers presenting peptides from unrelated viruses; (ii) that such TCRs co-cluster using Levenshtein distance or GLIPH2 based clustering method; and (iii) that some of these TCRs indeed recognize different, unrelated peptides without significant sequence homology upon re-expression in carrier T cells.

   The authors acknowledge that this in silico analysis is mostly based on unpaired alpha/beta sequence data, and that the chain pairing may influence specificity. They, therefore, perform a number of functional assays, demonstrating examples of T cells which respond by interferon gamma production to more than one peptide.

   We thank the reviewer for pointing to the fact that, beyond tetramer binding, we performed cumbersome functional studies to document polyreactivity.

   The paper is mostly very clearly written and presented and provides some fascinating novel perspectives on T cell cross-reactivity.

   We thank the reviewer for his appreciation

   The findings will surely be of interest to a broad readership - indeed anyone interested in how adaptive immunity works.

   The link between the different sections of the paper is the weakest aspect. The relationship between thymic selection and polyspecificity, and also the real relationship between in silico "cross-reactivity" as evidenced by multiple annotations and the functional polyspecific T cells remains unclear.

   Our flow of reasoning/analyzing was as follow. As we were studying the thymic selection of TCR repertoires, (1) we discovered a massive clustering within these repertoires. As for thymocytes this cannot be accounted for by a history of immune responses, this triggered our attention and led us to analyze the properties of these TCRs. This led us (2) to discover in these thymic repertoires “TCRs which have been annotated to multiple viral specificities”, that we were not aware of. We were so much intrigued by these observations that we wanted to substantiate them using datasets of paired  TCRs. As (3) we could confirm these observations in such datasets, this led us (4) to investigate these TCRs in functional studies. This is the link for the 1-to-4 sections.

   To make this link clearer, we have reworked the titles of the different Results’ sections such as to emphasize the switch from thymocyte bulk sequencing studies to that of single peripheral cell sequencing studies.

   The mechanistic molecular details underlying polyspecificity also remain unclear.

   Indeed, we believe that solving the structure of polyreactive TCRs interacting with different peptides will be needed for a molecular understanding of polyreactivity, but that it falls beyond the present work.

   But overall, lots of interesting new data, and some very intriguing hypotheses for the community to follow up on.

   We thank the reviewer for his overall comment

   Reviewer #3 (Public Review):

   In this manuscript, the authors propose that there is a special, previously unrecognized, high-frequency population of a/b TCRs that are shared between people, have high generation probabilities, and react to many unrelated viral epitopes. Here is the main flow of the results, with comments on the strengths of the conclusions:

   "Thymopoiesis selects a large and diverse set of clustered CDR3s with high generation probabilities" -- this seems correct and has been noted in earlier work by Mora and Walczak and others.

   So far, Mora and Walczak selection models in humans are based on studying PBMCs (our ref n° 27 in the revised version), not thymic DP and SP sorted cells, even in the mouse derived models for which they used the total thymic cells (our ref n° 27).

   Selection leads to a focusing of the CDR3 length which likely increases the degree of clustering and increases Pgen.

   To address this question, we compared the CDR3 length distribution between DP CD3+ cells and CD8 SP cells from our thymic dataset. We did not observe major changes. The distribution and the mean of CDR3 length for the two cell populations remained identical. We only observed a small shifting in the CDR3 length distribution towards shorter sequences post-selection. This is now reported in the new Figure 1-figure supplement 3C in the revised manuscript.

   "Clustered CDR3s are enriched for publicness " This also seems correct and again it makes sense: publicness is equivalent to having been independently rearranged (and sequenced) in another individual, which is determined by Pgen, and clustering is also determined to a large extent by Pgen (the factors that contribute to Pgen, shorter CDR3s for example, are largely shared between neighbor TCRs).

   We agree that theory could have indeed predicted that. In any case, to our knowledge, this is the first report of large clusters of just selected thymocytes’ CDR3s that moreover co-cluster between individuals with different HLA.

   "Clustered public CDR3s are enriched in viral specificities" -- This claim is not justified by the data, which comes from sequence matching against literature-derived databases. Rather, what is true is that "Clustered public CDR3s are enriched in public viral specificities".

   We changed “CDR3s are enriched in viral specificities” for “clustered public CDR3s are enriched in public viral specificities".

   But this might be a simple consequence of the previous observation, that "clustered CDR3s are enriched for publicness". One would need experimental specificity data on the very same datasets to make a conclusion about viral specificities in general.

   We based our interpretation on experimental data.

   Indeed, we manually curated databases to identify CDR3s that bind specific tetramers/dextramers. This type of “experimental specificity data” is for immunologists a paradigmatic and yet unchallenged mean to define specificity.

   We make the observation that there are more CDR3s from a TCR that does bind tetramers/dextramers presenting viral peptides in clustered than in dispersed CDR3s. This is a highly statistically significant fact, that we now report as a fact that we leave open to discussion/challenge by our community.

   "Identification of polyspecific TCRs" -- In this section, the authors report that some of the CDR3 clusters contain CDR3 sequences from literature-derived TCRs with multiple specificities. They conclude that these must represent polyspecific TCRs. The problem with this conclusion is that even having the same CDR3beta, let alone similar CDR3beta sequences, does not imply the same specificity. One can see the problem if one imagines a very deeply sequenced dataset, and focuses on a short CDR3 length with high frequency. With sufficient sampling, one will be able to navigate from nearly any single CDR3beta to any other CDR3beta of the same or similar length by jumping between single-mismatch variants. But this doesn't imply that all the TCRs from which these CDR3s were sampled, which likely have many different Vbeta genes and completely different TCRalpha sequences, must all bind the same thing.

   We will first point to the fact that we did not analyze “a very deeply sequenced dataset”, but only the 18 000 most abundant sequences per sample. Singletons were excluded. In addition, we did not mean to say that all the connected TCRs have the same specificities, regardless of their position in the cluster. Clustering algorithms, whether LV distance of GLIPH2 for example, are now commonly used to infer specificity of clusters and it is admitted that the closer the TCR sequences are, the more they share their specificities.

   That said, it is precisely because we acknowledge the limitation of bulk sequencing for inferring specificities that we turned to also analyze single-cell datasets.

   We made this more apparent by the new sections of the results that more clearly indicate the shift from unpaired bulk thymocyte sequencing and paired single peripheral cell sequencing.

   "Binding properties of polyspecific TCRs" -- Here the authors look to validate these results with paired TCR sequences. They analyze a public dataset made available by 10X genomics, featuring single-cell gene expression, TCR sequencing, and dextramer UMI counts for ~150,000 T cells. This is an amazing dataset with lots of interesting features, but, like any large high-throughput dataset, it needs to be analyzed with care.

   We can assure the reviewer that we were always very careful. Actually, we even started by carefully reviewing the 10X proposed methodology, in which we identified major biases. This led us to explore this dataset cautiously and without preconceived ideas.

   The authors claim to see evidence for large-scale cross-reactivity. This comes mainly from a set of dextramers for A03 and A11-restricted peptides. But these dextramers appear to be binding in a uniquely non-specific manner (by comparison with the other dextramers) and non-TCR-dependent manner in this experiment. One can see this, for example, by comparing the consistency of binding within expanded clonotypes: for a specific dextramer like A*02-GIL(Flu), positive binding for one cell in a clonotype greatly increases the likelihood of binding for other cells in the clonotype, suggesting that the binding is mediated by the TCR.

   This is not true for the A03 and A11 dextramers (except for a few expanded clonotypes in an A*11 donor). TCR sequence doesn't appear to be the determining factor for binding to these dextramers; rather it may be expression of KIR genes or other surface proteins that can interact with MHC.

   These are indeed striking binding patterns that are remarkably similar for a single CDR3 beta associated with more than 40 different CDR3s alpha (and moreover from two donors). The first attitude of immunologists would indeed be of discarding this observation for non-fitting the paradigms. We would like to rather propose an agnostic view at these results.

   These results show that a series of five A03 and A11 dextramers loaded with various peptides bind to cells that express a given CDR3 beta associated with a multitude of CDR3alpha. If it would be an MHC to KIR binding, then such dextramers should bind to most cells, independently of their TCRs. We have added two supplementary figures (Figure 4-figure supplement 8B-C) to show that this is not the case, and that further show very different binding patterns.

   If it would be a binding to “other surface proteins”, it would likely be the same.

   We identified a CDR3 from donor 3 which binds preferentially to A03 and A11 dextramers. However, it binds to only 4 out of 5 of these. If the binding is non-specific and non-TCR-dependent, a binding for the A0301 RIAAWMATY BCL2L1 dextramer should also have been observed. Moreover, we identified this same CDR3beta in two other cells from donor 1 and 4, and that were associated with a different CDR3alpha. Except for only one binding, these TCRs didn’t show binding to the A03 and A11 dextramers.

   Moreover, we identify another CDR3 from donor 1 that is associated with a strong binding to one A1101 dextramer presenting an EBV peptide when associated to many different CDR3alpha. The binding to the other A03 and A011 dextramer is weaker and seem to depend more on the CD3alpha.

   If the binding of A03 and A011 dextramers is non-specific and non-TCR-dependent, why is there such a difference between the binding of A1101 IVTDFSVIK and A1101 AVFDRSDAK dextramers?

   "Polyspecific T cells are activated in vitro by multiple viral peptides" Here the authors explore polyspecificity experimentally. First they report that polyclonal populations of T cells, sorted for binding to one dextramer, can also produce IFN gamma upon stimulation with a distinct peptide, albeit more weakly than for the cognate peptide.

   This is indeed true for CMV+ sorted cells that respond better to CMV peptides than to EBV ones, but not true for EBV+ sorted cells that also respond better to CMV peptides than to EBV ones.

   But it's not clear that the concentrations of the peptides are appropriate for stringently detecting cross-reactivity.

   We wonder what does mean “stringently”? It is possible that stringently mainly means defining the conditions that eliminates what does not fit the current paradigm?

   More factually, the peptide concentration used for these experiments, presented in Fig. 5A-B, was 1 µg/mL, i.e. ~1 µM for a 9-10 aa-long peptide. This is clearly a physiological concentration for viral peptides, routinely used in in-vitro recall assays. We can thus rule out that the observed cross-reactivity is simply due to an excess peptide stimulation.

   Then the authors actually synthesize and characterize individual TCRs. Here what is seen is consistent with expectation and does not seem to support the idea of substantial fuzzy cross-reactivity: binding to the cognate peptide is 3-4 orders of magnitude stronger than to the alternative peptides.

   We respectfully disagree. First, as shown in Fig. 5C TCR#35-13 (cognate peptide HLA-A2-restricted Flu MP 58-66) indeed recognizes the alternative HLA-A2-restricted CMV IE1 184-192 peptide with a 3-4 higher log EC50; yet, the EC50 of this TCR is approx. 10e-6 M, i.e. 1 µM, which remains a physiological concentration. Second, this is not the case for TCR#36-150 (same cognate peptide HLA-A2-restricted Flu MP 58-66), which actually recognizes the alternative HLA-A2-restricted EBV BMLF1 280-288 peptide with a 4-fold lower EC50.

   The only exception is the GAD 114-122 TCR, where the different peptides appear to be closer in binding strength. But in this case, the authors state that they "analyzed their response to a set of peptides comprising their cognate peptide and peptides with no significant structural commonalities, selected by testing combinatorial peptide libraries". If the competitor peptides came from peptide library screening then the observation of strong binding to alternative peptides does not seem as surprising as a TCR that binds well to a Flu peptide, say, and also a CMV peptide, selected from a smallish set of possibilities.

   As explained above, this TCR does not stand as an exception compared to Flu-reactive TCRs. Moreover, it should be noted that this GAD 114-122 TCR recognizes its cognate peptide in a similar or even lower concentration range compared to the Flu-reactive TCR #36-150. It should also be pointed out that, contrary to the Flu-reactive TCRs, here we did not have any reference dextramer binding data to guide our peptide selection, which is why we resorted to combinatorial peptide libraries. Thus, although different strategies were used, peptide selection was “guided” in both instances.

   It is pretty well established that TCRs are cross-reactive, both for nearby peptides and also for sequence-dissimilar peptides.

   We agree and had notably quoted the landmark paper by Don Mason estimating that each TCR may respond to over 106 different peptides from an estimated repertoire of > 1010 peptides. Based on the Don Mason estimate of cross reactivity, the chance to find a cross reactive peptide at random would be around 10-4.

   Here, we just tested a few peptides from different viruses. If Don Mason’s estimates are correct, for a given TCR, the chance to find even just 1 cross-reactive peptide among these few peptides would be at most of 10-3, the chance to find 2 cross reactive peptides would be of 10-6 and that to find 3 or more cross reactive peptides would have be infinitesimal.

   Thus, if the polyreactivity that we described is part of this general cross reactivity, our results are at least highlight a major previously unreported bias in the selection of these cells.

   The question is whether widespread, functionally relevant (not just dextramer binding at some concentration) poly-reactivity to diverse viral peptides is a defining feature of a large fraction of the TCR repertoire. The paper does not appear to present sufficiently strong evidence to support this claim.

   We agree with the reviewer that more work is needed to “fully” appreciate the role of polyreactive cells!

   Read the original source
3. 

   eLife

   Sep 9, 2022

   eLife assessment

   This paper reports on important observations regarding human CD8 T cells that express shared T cell receptors amongst individuals and exhibit poly-specificity directed mainly to several unrelated viral antigens. Although the majority of the claims are convincingly supported by results from both in silico and experimental approaches, mechanistic molecular details underlying poly-specificity remain incomplete. The results from these studies will enhance the ongoing debate on T cell specificity and potentially, will impact fields related to immunology, for example, immunoparasitology, cell biology, and vaccine development.

   Read the original source
4. 

   eLife

   Sep 9, 2022

   Reviewer #1 (Public Review):

   In this paper, Quiniou and colleagues show via orthogonal methods human thymopoiesis releases a large population of CD8+ T cells harboring a/b paired TCRs that (i) have high generation probabilities and (ii) a preferential usage of some V and J genes, (iii) are shared between individuals and (iv) can each recognize and be activated by multiple unrelated viral peptides, notably from EBV, CMV and influenza.

   Major strengths of the paper:

   Quiniou et al. generated single-cell sequencing datasets of the earliest stages of TCR beta chain gene recombination. And then showed that a subset of them is highly clustered also having high generation probability.

   They show that these T cells can bind multiple antigens, both via the use of public antigen-specific datasets as well as corroborating experimental TCR expression …

   Reviewer #1 (Public Review):

   In this paper, Quiniou and colleagues show via orthogonal methods human thymopoiesis releases a large population of CD8+ T cells harboring a/b paired TCRs that (i) have high generation probabilities and (ii) a preferential usage of some V and J genes, (iii) are shared between individuals and (iv) can each recognize and be activated by multiple unrelated viral peptides, notably from EBV, CMV and influenza.

   Major strengths of the paper:

   Quiniou et al. generated single-cell sequencing datasets of the earliest stages of TCR beta chain gene recombination. And then showed that a subset of them is highly clustered also having high generation probability.

   They show that these T cells can bind multiple antigens, both via the use of public antigen-specific datasets as well as corroborating experimental TCR expression and binding essays.

   Minor weaknesses:

   To what extent is TCR clustering and high pgen and cross-individual sharing correlated? What is the pgen of the sequences clustered with the high pgen cells? Can you comment on the correlation between these three phenomena? In other words, to what extent is this surprising to see that highly clustered TCRs have higher pgen and are more shared?

   Potential Impact of the paper:

   This work highlights an intrinsic property of the adaptive immune response: to generate TCRs with high generation probability that can efficiently bind multiple antigens. This finding has, therefore important impact on drug discovery and vaccine design.

   Read the original source
5. 

   eLife

   Sep 9, 2022

   Reviewer #2 (Public Review):

   This study analyses the T cell receptor (TCR) repertoire of double positive human thymocytes, and compares this to mature single positive CD8 cells. The first major finding is that the repertoire post-selection is enriched for groups of TCRs with high generation probabilitites, similar sequences, and for TCRs previously annotated for viral specificity. This data is clearly presented and convincing. The extent of analysis of the human thymocyte repertoire is still very limited, and the paper adds significantly to this important question.

   The second major finding is much more controversial. The authors first investigate the publicly available databases and show that there is a substantial proportion of TCRs which have been annotated to multiple viral specificities, a fact which is well-known to the specialists …

   Reviewer #2 (Public Review):

   This study analyses the T cell receptor (TCR) repertoire of double positive human thymocytes, and compares this to mature single positive CD8 cells. The first major finding is that the repertoire post-selection is enriched for groups of TCRs with high generation probabilitites, similar sequences, and for TCRs previously annotated for viral specificity. This data is clearly presented and convincing. The extent of analysis of the human thymocyte repertoire is still very limited, and the paper adds significantly to this important question.

   The second major finding is much more controversial. The authors first investigate the publicly available databases and show that there is a substantial proportion of TCRs which have been annotated to multiple viral specificities, a fact which is well-known to the specialists in the field, but not previously addressed. The authors acknowledge that this in silico analysis is mostly based on unpaired alpha/beta sequence data, and that the chain pairing may influence specificity. They, therefore, perform a number of functional assays, demonstrating examples of T cells which respond by interferon gamma production to more than one peptide.

   The paper is mostly very clearly written and presented and provides some fascinating novel perspectives on T cell cross-reactivity. The findings will surely be of interest to a broad readership - indeed anyone interested in how adaptive immunity works. The link between the different sections of the paper is the weakest aspect. The relationship between thymic selection and polyspecificity, and also the real relationship between in silico "cross-reactivity" as evidenced by multiple annotations and the functional polyspecific T cells remains unclear. The mechanistic molecular details underlying polyspecificity also remain unclear. But overall, lots of interesting new data, and some very intriguing hypotheses for the community to follow up on.

   Read the original source
6. 

   eLife

   Sep 9, 2022

   Reviewer #3 (Public Review):

   In this manuscript, the authors propose that there is a special, previously unrecognized, high-frequency population of a/b TCRs that are shared between people, have high generation probabilities, and react to many unrelated viral epitopes. Here is the main flow of the results, with comments on the strengths of the conclusions:

   "Thymopoiesis selects a large and diverse set of clustered CDR3s with high generation probabilities" – this seems correct and has been noted in earlier work by Mora and Walczak and others. Selection leads to a focusing of the CDR3 length which likely increases the degree of clustering and increases Pgen.

   "Clustered CDR3s are enriched for publicness" This also seems correct and again it makes sense: publicness is equivalent to having been independently rearranged (and sequenced) in …

   Reviewer #3 (Public Review):

   In this manuscript, the authors propose that there is a special, previously unrecognized, high-frequency population of a/b TCRs that are shared between people, have high generation probabilities, and react to many unrelated viral epitopes. Here is the main flow of the results, with comments on the strengths of the conclusions:

   "Thymopoiesis selects a large and diverse set of clustered CDR3s with high generation probabilities" – this seems correct and has been noted in earlier work by Mora and Walczak and others. Selection leads to a focusing of the CDR3 length which likely increases the degree of clustering and increases Pgen.

   "Clustered CDR3s are enriched for publicness" This also seems correct and again it makes sense: publicness is equivalent to having been independently rearranged (and sequenced) in another individual, which is determined by Pgen, and clustering is also determined to a large extent by Pgen (the factors that contribute to Pgen, shorter CDR3s for example, are largely shared between neighbor TCRs).

   "Clustered public CDR3s are enriched in viral specificities" – This claim is not justified by the data, which comes from sequence matching against literature-derived databases. Rather, what is true is that "Clustered public CDR3s are enriched in public viral specificities". But this might be a simple consequence of the previous observation, that "clustered CDR3s are enriched for publicness". One would need experimental specificity data on the very same datasets to make a conclusion about viral specificities in general.

   "Identification of polyspecific TCRs" – In this section, the authors report that some of the CDR3 clusters contain CDR3 sequences from literature-derived TCRs with multiple specificities. They conclude that these must represent polyspecific TCRs. The problem with this conclusion is that even having the same CDR3beta, let alone similar CDR3beta sequences, does not imply the same specificity. One can see the problem if one imagines a very deeply sequenced dataset, and focuses on a short CDR3 length with high frequency. WIth sufficient sampling, one will be able to navigate from nearly any single CDR3beta to any other CDR3beta of the same or similar length by jumping between single-mismatch variants. But this doesn't imply that all the TCRs from which these CDR3s were sampled, which likely have many different Vbeta genes and completely different TCRalpha sequences, must all bind the same thing.

   "Binding properties of polyspecific TCRs" – Here the authors look to validate these results with paired TCR sequences. They analyze a public dataset made available by 10X genomics, featuring single-cell gene expression, TCR sequencing, and dextramer UMI counts for ~150,000 T cells. This is an amazing dataset with lots of interesting features, but, like any large high-throughput dataset, it needs to be analyzed with care. The authors claim to see evidence for large-scale cross-reactivity. This comes mainly from a set of dextramers for A*03 and A*11-restricted peptides. But these dextramers appear to be binding in a uniquely non-specific manner (by comparison with the other dextramers) and non-TCR-dependent manner in this experiment. One can see this, for example, by comparing the consistency of binding within expanded clonotypes: for a specific dextramer like A*02-GIL(Flu), positive binding for one cell in a clonotype greatly increases the likelihood of binding for other cells in the clonotype, suggesting that the binding is mediated by the TCR. This is not true for the A*03 and A*11 dextramers (except for a few expanded clonotypes in an A*11 donor). TCR sequence doesn't appear to be the determining factor for binding to these dextramers; rather it may be expression of KIR genes or other surface proteins that can interact with MHC.

   "Polyspecific T cells are activated in vitro by multiple viral peptides" Here the authors explore polyspecificity experimentally. First they report that polyclonal populations of T cells, sorted for binding to one dextramer, can also produce IFNgamma upon stimulation with a distinct peptide, albeit more weakly than for the cognate peptide. But it's not clear that the concentrations of the peptides are appropriate for stringently detecting cross-reactivity. Then the authors actually synthesize and characterize individual TCRs. Here what is seen is consistent with expectation and does not seem to support the idea of substantial fuzzy cross-reactivity: binding to the cognate peptide is 3-4 orders of magnitude stronger than to the alternative peptides. The only exception is the GAD 114-122 TCR, where the different peptides appear to be closer in binding strength. But in this case, the authors state that they "analyzed their response to a set of peptides comprising their cognate peptide and peptides with no significant structural commonalities, selected by testing combinatorial peptide libraries". If the competitor peptides came from peptide library screening then the observation of strong binding to alternative peptides does not seem as surprising as a TCR that binds well to a Flu peptide, say, and also a CMV peptide, selected from a smallish set of possibilities.

   It is pretty well established that TCRs are cross-reactive, both for nearby peptides and also for sequence-dissimilar peptides. The question is whether widespread, functionally relevant (not just dextramer binding at some concentration) poly-reactivity to diverse viral peptides is a defining feature of a large fraction of the TCR repertoire. The paper does not appear to present sufficiently strong evidence to support this claim.

   Read the original source
7. 

   Version published to 10.1101/2020.07.27.223354v2 on bioRxiv

   Jun 21, 2022
8. 

   Version published to 10.1101/2020.07.27.223354v1 on bioRxiv

   Jul 29, 2020