Asymmetric framework motion of TCRαβ controls load-dependent peptide discrimination

  1. Ana C Chang-Gonzalez
  2. Robert J Mallis
  3. Matthew J Lang
  4. Ellis L Reinherz
  5. Wonmuk Hwang

  • Curated by eLife

    eLife logo

    eLife assessment

    Using extensive atomistic molecular dynamics simulations, the authors analyzed the TCR/pMHC interface with different peptide sequences and protein constructs. The results provide important insights into the catch-bond phenomenon in the context of T-cell activation. In particular, the analysis points to convincing evidence that supports the role of force in further discriminating different peptides during the activation process beyond structural considerations.

Reviewed by

Read the full article

Listed in

Log in to save this article

Abstract

Mechanical force is critical for the interaction between an αβ T cell receptor (TCR) and a peptide-bound major histocompatibility complex (pMHC) molecule to initiate productive T-cell activation. However, the underlying mechanism remains unclear. We use all-atom molecular dynamics simulations to examine the A6 TCR bound to HLA-A*02:01 presenting agonist or antagonist peptides under different extensions to simulate the effects of applied load on the complex, elucidating their divergent biological responses. We found that TCR α and β chains move asymmetrically, which impacts the interface with pMHC, in particular the peptide-sensing CDR3 loops. For the wild-type agonist, the complex stabilizes in a load-dependent manner while antagonists destabilize it. Simulations of the Cβ FG-loop deletion, which reduces the catch bond response, and simulations with in silico mutant peptides further support the observed behaviors. The present results highlight the combined role of interdomain motion, fluctuating forces, and interfacial contacts in determining the mechanical response and fine peptide discrimination by a TCR, thereby resolving the conundrum of nearly identical crystal structures of TCRαβ-pMHC agonist and antagonist complexes.

Article activity feed

  1. Version published to 10.7554/elife.91881 on eLife
  2. eLife

    Author Response

    Reviewer #1 (Public Review):

    The authors present a detailed analysis of a set of molecular dynamics computer simulations of several variants of a T-cell receptor (TCR) in isolation and bound to a Major Histocompatibility Complex with peptide (pMHC), with the aim of improving our understanding of the mechanism T cell activation in immunity. By analyzing simulations of peptide mutants and partially truncated TCRs, the authors find that native peptide agonists lead to a so-called catch-bond response, whereby tensile force applied in the direction of separation between TCR/pMHC appears to strengthen the TCR/pMHC interface, whereas mutated peptides exhibit the more common slip-bond response, in which applied force destabilizes the binding interface. Using various computational metrics and simulation statistics, the authors propose a model in which tensile force preferentially suppresses thermal fluctuations in the variable α domain of the TCR (vs the β domain) in a peptide-dependent manner, which orders and strengthens the binding interface by bringing together the complementarity-determining regions (CDRs) in the TCR variable chains, but only if the peptide is correctly matched to the TCR.

    R1-0. The study is detailed and written clearly, and conclusions appear convincing and are supported by the simulation data. However, the actual motions at the molecular or amino-acid level of how the catch-bond vs slip bond response originates remain somewhat unclear, and will probably warrant further investigations. Specific hypotheses that could be testable in experiments, such as predictions of which peptide (or TCR) mutations or which peptides could generate a catch-vs-slip response or activation, would have especially strengthened this study.

    Catch bonds have been observed in different αβ TCRs that differ in sequence when paired with their matching pMHC. Thus, there should be a general principle that apply irrespective of particular TCR sequences, as summarized in Fig. 8. The predictive capacity of this model in terms of understanding experiments is explained in our reply R0-3. Here, we discuss about designing specific point mutations to TCR that have not been studied previously. In our simulations, we can identify high-occupancy contacts that are present mainly in the high-load case as target for altering the catch bond behavior. An example is V7-G100 between the peptide and Vβ (Fig. 2C, bottom panel). The V7R mutant peptide is a modified agonist that we have already studied, where R7 forms hydrogen bonds and nonpolar contacts with residues other than βG100, albeit with lower occupancy (page 11, lines 280–282 and page 32, Fig. 5–figure supplement 2B). Instead of the V7R mutation to the peptide, mutating βG100 to other residues may lead to different effects. For example, compared to G100A, mutation to a bulkier residue such as G100F may cause opposing effects: It may induce steric mismatch that destabilizes the interface. Conversely, a stronger hydrophobic effect might increase the baseline bond lifetime. Also, mutating G100 to a polar residue may have even greater effect, leading to a slip bond or absence of measurable binding.

    As the reviewer suggested in R1-5, it will also be interesting to crosslink Vα and Cα by a disulfide bond to suppress its motion. Again, there are different possible outcomes. The lack of Vα-Cα motion could stabilize the interface with pMHC, resulting in a longer bond lifetime. Conversely, if the disulfide bond alters the V-C angle, it would have an opposite effect of destabilizing the interface by tilting it relative to the loading direction, similar to the dFG mutant in Appendix 1 (page 24).

    To make better predictions, simulations of such mutants should to be performed under different conditions and analyzed, which would be beyond the scope of the present study.

    Change made:

    • Page 14, Concluding Discussion, lines 395–402: We added a discussion about using simulations for designing and testing point mutants.

    Reviewer #2 (Public Review):

    In this work, Chang-Gonzalez and co-workers investigate the role of force in peptide recognition by T-cells using a model T-cell/peptide recognition complex. By applying forces through a harmonic restraint on distances, the authors probe the role of mechanical pulling on peptide binding specificity. They point to a role for force in distinguishing the different roles played by agonist and antagonist peptides for which the bound configuration is not clearly distinguishable. Overall, I would consider this work to be extensive and carefully done, and noteworthy for the number of mutant peptides and conditions probed. From the text, I’m not sure how specific these conclusions are to this particular complex, but I do not think this diminishes the specific studies.

    I have a couple of specific comments on the methodology and analysis that the authors could consider:

    R2-1. 1) It is not explained what is the origin of force on the peptide-MHC complex. Although I do know a bit about this, it’s not clear to me how the force ends up applied across the complex (e.g. is it directional in any way, on what subdomains/residues do we expect it to be applied), and is it constant or stochastic. I think it would be important to add some discussion of this and how it translates into the way the force is applied here (on terminal residues of the complex).

    As explained in our reply R0-1, force on the TCRαβ-pMHC complex arises during immune surveillance where the T-cell moves over APC. Generated by the cellular machinery such as actin retrograde flow and actomyosin motility, the applied force fluctuates, which would be on top of spontaneous fluctuation in force by thermal motion. This has been directly measured for the T-cell using a pMHC-coated bead via optical tweezers (see Feng et al., 2017, Fig. 1) and by DNA tension sensors (Liu, et al., 2016, Fig. 4; already cited in the manuscript). The direction of force also fluctuates that is longitudinal on average (see R1-6). How force distributes across the molecule is a great question, for which we plan to develop a computational method to quantify.

    Changes made.

    • Pages 3–4, newly added Results section ‘Applying loads to TCRαβ-pMHC complexes:’ We included the origin of force and its fluctuating nature, and the question of how loads are distributed across the molecule.

    • The reference (Feng et al., 2017) has been added in the above section.

    R2-2. 2) In terms of application of the force, I find the use of a harmonic restraint and then determining a distance at which the force has a certain value to be indirect and a bit unphysical. As just mentioned, since the origin of the force is not a harmonic trap, it would be more straightforward to apply a pulling force which has the form -F*d, which would correspond to a constant force (see for example comment articles 10.1021/acs.jpcb.1c10715,10.1021/acs.jpcb.1c06330). While application of a constant force will result in a new average distance, for small forces it does so in a way that does not change the variance of the distance whereas a harmonic force pollutes the variance (see e.g. 10.1021/ct300112v in a different context). A constant force could also shift the system into a different state not commensurate with the original distance, so by applying a harmonic trap, one could be keeping ones’ self from exploring this, which could be important, as in the case of certain catch bond mechanisms. While I certainly wouldn’t expect the authors to redo these extensive simulations, I think they could at least acknowledge this caveat, and they may be interested in considering a comparison of the two ways of applying a force in the future.

    Thanks for the suggestions and references. The paper by Stirnemann (2022) is a review including different computational methods of applying forces, mainly constant force and constant pulling velocity (steered molecular dynamics; SMD). The second one by Gomez et al., (2021) is a rather broad review of mechanosensing where discussion about computer simulation was mainly on SMD. In the third one by Pitera and Chodera (2012), potential limitations of using harmonic potentials in sampling nonlinear potential of mean force (PMF) are discussed.

    In the above references, loads or restraints are used to study conformational transitions or to sample the PMF, which are different from the use of positional restraints in our work. As explained in R0-1, positional restraint better mimics reality where the terminal ends of TCR and pMHC are anchored on the membranes of respective cells. Also, the concern raised by the reviewer about ruling out different states would be applicable to the case when there are multiple conformational states with local free energy minima at different extensions. Here, we are probing changes in the conformational dynamics (deformation and conformational fluctuation), rather than transitions between well-defined states.

    In Pitera and Chodera (2012) and also in other approaches such as umbrella sampling, the spring constant of the harmonic potential should be chosen sufficiently soft so that sampling around the neighborhood of the center of the potential can be made. On the other hand, if the harmonic potential is much stiffer than the local curvature of the PMF, although sampling may suffer, local gradient of the PMF, i.e, the force about the center of the potential, can be made. This has been studied earlier by one of us in Hwang (2007), which forms the basis for using a stiff harmonic potential for measuring the load on the TCRαβ-pMHC complex. The 1-kcal/(mol·˚A2) spring constant used in our study (page 17, line 540) was selected such that the thermally driven positional fluctuation is on the order of 0.8 ˚A. Hence, it is sufficiently stiff considering the much larger size of the TCRαβ-pMHC complex and the flexible added strands.

    Changes made:

    • Page 4, lines 117–119, newly added Results section ‘Applying loads to TCRαβ-pMHC complexes:’ The above explanation about the use of stiff harmonic restraint for measuring forces is added.

    • The 4 references mentioned above have been added to the above section.

    R2-3. 3) For the PCA analysis, I believe the authors learn separate PC vectors from different simulations and then take the dot product of those two vectors. Although this might be justified based on the simplified coordinate upon which the PCA is applied, in general, I am not a big fan of running PCA on separate data sets and then comparing the outputs, as the meaning seems opaque to me. To compare the biggest differences between many simulations, it would make more sense to me to perform PCA on all of the data combined, and see if there are certain combinations of quantities that distinguish the different simulations. Alternatively and probably better, one could perform linear discriminant analysis, which is appropriate in this case because one already knows that different simulations are in different states, and hence the LDA will directly give the linear coordinate that best distinguishes classes.

    As explained in R0-2, triads and BOC models are assigned to the same TCR across different simulations in identical ways. For the purpose of examining the relative Vα-Vβ and V-C motions, we believe comparing them across different simulations is a valid approach. When the motions are very distinct, it would be possible to combine all data and perform PCA or LDA to classify them. However, when behaviors differ subtly, analysis on the combined data may not capture individual behaviors. By analogy, consider two sets of 2-dimensional data obtained for the same system under different conditions. If each set forms an elliptical shape with the major axis differing slightly in direction, performing PCA separately on the two sets and comparing the angle between the major axes informs the difference between the two sets. If PCA were performed on the combined data (superposition of two ellipses forming an angle), it will be difficult to find the difference. LDA would likewise be difficult to apply without a very clear separation of behaviors.

    As also explained in R0-2, PCA is just one of multiple analyses we carried out to establish a coherent picture. The main use of PCA to this end was to compare directions of motion and relative amplitude of the motion among the subdomains.

    Changes made:

    • Page 6, lines 171–175 and page 8, lines 226–227: The rationale for applying PCA on triads and BOC models in different simulations are explained.

  3. eLife

    eLife assessment

    Using extensive atomistic molecular dynamics simulations, the authors analyzed the TCR/pMHC interface with different peptide sequences and protein constructs. The results provide important insights into the catch-bond phenomenon in the context of T-cell activation. In particular, the analysis points to convincing evidence that supports the role of force in further discriminating different peptides during the activation process beyond structural considerations.

  4. eLife

    Reviewer #1 (Public Review):

    The authors present a detailed analysis of a set of molecular dynamics computer simulations of several variants of a T-cell receptor (TCR) in isolation and bound to a Major Histocompatibility Complex with peptide (pMHC), with the aim of improving our understanding of the mechanism T cell activation in immunity. By analyzing simulations of peptide mutants and partially truncated TCRs, the authors find that native peptide agonists lead to a so-called catch-bond response, whereby tensile force applied in the direction of separation between TCR/pMHC appears to strengthen the TCR/pMHC interface, whereas mutated peptides exhibit the more common slip-bond response, in which applied force destabilizes the binding interface.

    Using various computational metrics and simulation statistics, the authors propose a model in which tensile force preferentially suppresses thermal fluctuations in the variable α domain of the TCR (vs the β domain) in a peptide-dependent manner, which orders and strengthens the binding interface by bringing together the complementarity-determining regions (CDRs) in the TCR variable chains, but only if the peptide is correctly matched to the TCR.

    The study is detailed and written clearly, and conclusions appear convincing and are supported by the simulation data. However, the actual motions at the molecular or amino-acid level of how the catch-bond vs slip bond response originates remain somewhat unclear, and will probably warrant further investigations. Specific hypotheses that could be testable in experiments, such as predictions of which peptide (or TCR) mutations or which peptides could generate a catch-vs-slip response or activation, would have especially strengthened this study.

  5. eLife

    Reviewer #2 (Public Review):

    In this work, Chang-Gonzalez and co-workers investigate the role of force in peptide recognition by T-cells using a model T-cell/peptide recognition complex. By applying forces through a harmonic restraint on distances, the authors probe the role of mechanical pulling on peptide binding specificity. They point to a role for force in distinguishing the different roles played by agonist and antagonist peptides for which the bound configuration is not clearly distinguishable. Overall, I would consider this work to be extensive and carefully done, and noteworthy for the number of mutant peptides and conditions probed. From the text, I'm not sure how specific these conclusions are to this particular complex, but I do not think this diminishes the specific studies.

    I have a couple of specific comments on the methodology and analysis that the authors could consider:

    1. It is not explained what is the origin of force on the peptide-MHC complex. Although I do know a bit about this, it's not clear to me how the force ends up applied across the complex (e.g. is it directional in any way, on what subdomains/residues do we expect it to be applied), and is it constant or stochastic. I think it would be important to add some discussion of this and how it translates into the way the force is applied here (on terminal residues of the complex).

    2. In terms of application of the force, I find the use of a harmonic restraint and then determining a distance at which the force has a certain value to be indirect and a bit unphysical. As just mentioned, since the origin of the force is not a harmonic trap, it would be more straightforward to apply a pulling force which has the form -F*d, which would correspond to a constant force (see for example comment articles 10.1021/acs.jpcb.1c10715, 10.1021/acs.jpcb.1c06330). While application of a constant force will result in a new average distance, for small forces it does so in a way that does not change the variance of the distance whereas a harmonic force pollutes the variance (see e.g. 10.1021/ct300112v in a different context). A constant force could also shift the system into a different state not commensurate with the original distance, so by applying a harmonic trap, one could be keeping ones' self from exploring this, which could be important, as in the case of certain catch bond mechanisms. While I certainly wouldn't expect the authors to redo these extensive simulations, I think they could at least acknowledge this caveat, and they may be interested in considering a comparison of the two ways of applying a force in the future.

    3. For the PCA analysis, I believe the authors learn separate PC vectors from different simulations and then take the dot product of those two vectors. Although this might be justified based on the simplified coordinate upon which the PCA is applied, in general, I am not a big fan of running PCA on separate data sets and then comparing the outputs, as the meaning seems opaque to me. To compare the biggest differences between many simulations, it would make more sense to me to perform PCA on all of the data combined, and see if there are certain combinations of quantities that distinguish the different simulations. Alternatively and probably better, one could perform linear discriminant analysis, which is appropriate in this case because one already knows that different simulations are in different states, and hence the LDA will directly give the linear coordinate that best distinguishes classes.

  6. eLife

    Reviewer #3 (Public Review):

    This simulation study presents a valuable finding on the load-dependence (i.e., dependence on a pulling force) of the recognition of a peptide-bound major histocompatibility complex (pMHC) antigen by a T cell receptor (TCR). The evidence supporting the claims of the authors is solid, although inclusion of a larger number of simulations would have strengthened the study. The work will be of interest to computational structural biologists and immunologists.

  7. Version published to 10.1101/2023.09.10.557064v1 on bioRxiv