Progressive enhancement of kinetic proofreading in T cell antigen discrimination from receptor activation to DAG generation
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
In this manuscript, a light-gated receptor system (LOV2) linked to T cell receptor signaling machinery is enhanced by addition of an adhesion system enabling robust operation over a larger range of kinetic parameters. This system enables an exploration of how kinetic proofreading processes executed in seconds relate to T cell activation program involving reactions taking minutes to hours.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Immunology and Inflammation (eLife)
Abstract
T cells use kinetic proofreading to discriminate antigens by converting small changes in antigen-binding lifetime into large differences in cell activation, but where in the signaling cascade this computation is performed is unknown. Previously, we developed a light-gated immune receptor to probe the role of ligand kinetics in T cell antigen signaling. We found significant kinetic proofreading at the level of the signaling lipid diacylglycerol (DAG) but lacked the ability to determine where the multiple signaling steps required for kinetic discrimination originate in the upstream signaling cascade (Tiseher and Weiner, 2019). Here, we uncover where kinetic proofreading is executed by adapting our optogenetic system for robust activation of early signaling events. We find the strength of kinetic proofreading progressively increases from Zap70 recruitment to LAT clustering to downstream DAG generation. Leveraging the ability of our system to rapidly disengage ligand binding, we also measure slower reset rates for downstream signaling events. These data suggest a distributed kinetic proofreading mechanism, with proofreading steps both at the receptor and at slower resetting downstream signaling complexes that could help balance antigen sensitivity and discrimination.
Article activity feed
-
-
Author Response
Reviewer #2 (Public Review):
- “…it was important that the output response was intimately linked to the bound state of the receptor, in this case the TCR, with ligand unbinding rapidly reversing all proofreading steps. This means that dissociation of a single TCR should disrupt signaling, and implicitly assumes a direct physical connection between the bound receptor and the KP modifications. However, this mechanism becomes much harder to argue when the KP steps are physically uncoupled from bound TCR, such as in LAT microclusters or DAG production.”
We agree that signaling events in the kinetic proofreading chain must be linked to ligand unbinding. We have added discussion to the paragraphs beginning on page 20 line 440 of recent work from Yi et al. 2019 and Lo et al. 2018 suggesting a physical link between bound TCRs …
Author Response
Reviewer #2 (Public Review):
- “…it was important that the output response was intimately linked to the bound state of the receptor, in this case the TCR, with ligand unbinding rapidly reversing all proofreading steps. This means that dissociation of a single TCR should disrupt signaling, and implicitly assumes a direct physical connection between the bound receptor and the KP modifications. However, this mechanism becomes much harder to argue when the KP steps are physically uncoupled from bound TCR, such as in LAT microclusters or DAG production.”
We agree that signaling events in the kinetic proofreading chain must be linked to ligand unbinding. We have added discussion to the paragraphs beginning on page 20 line 440 of recent work from Yi et al. 2019 and Lo et al. 2018 suggesting a physical link between bound TCRs and LAT clusters. The full paragraphs are reproduced below.
“The kinetic proofreading model requires all intermediate steps to reset upon unbinding of the ligand (Fig. 1A). This means that information about the receptor’s binding state must be communicated to all proofreading steps. If kinetic proofreading steps exist beyond the T cell receptor, how is unbinding information conveyed to these effectors? Importantly, there is evidence of physical proximity of LAT with the receptor. While TCR/Zap-70 and LAT/PLCγ microclusters form spatially segregated domains, these domains remain adjacent to one another (Yi et al., 2019). Lo et al. demonstrated that the protein Lck binds Zap-70 with its SH2 domain and LAT with its SH3 domain, potentially bridging the two signaling domains together and propagating binding information (Lo et al., 2018).
An attractive reset mechanism is the segregation of CD45 away from bound receptors, creating spatial regions in which TCR and LAT associated activating events can occur (S. J. Davis & van der Merwe, 2006). Super-resolution microscopy by Razvag et al. measured TCR/CD45 segregated regions within seconds of antigen contact at the tips of T cell microvilli (Razvag et al., 2018). Upon unbinding, these regions of phosphatase exclusion collapse, allowing CD45 to dephosphorylate receptor ITAMs and LAT clusters. However, the rate of dephosphorylation for LAT and receptor ITAMs could differ. LAT clusters exclude CD45 in reconstituted bilayer systems, potentially limiting the dephosphorylation to LAT molecules at the edges of the cluster thus slowing reset (Su et al., 2016). The kinetics of multivalent protein-protein interactions within TCR and LAT clusters can also influence dephosphorylation and dissociation rates (Goyette et al., 2022).
A CD45-mediated reset mechanism would restrict proofreading to membrane-bound signaling events occurring within a CD45-depleted region. Downstream events that dissociate away from the membrane or diffuse out of the segregated region could not directly participate in the proofreading chain, as the collapse of a CD45 segregated region could not reset signaling entities released into the cytosol (e.g. release of IP3 in the cleavage of PIP2 to DAG).”
- …The data clearly demonstrate a time delay between receptor binding and the measured outputs, but it is not so surprising that this lag would exist in propagating the signal through the intracellular network.
We apologize for this point of confusion in our methodology. We are unable to measure the time lag between receptor binding and signal propagation through the network because our system is terminated by blue light. Binding is stochastically initiated much like native ligand/receptor interactions. The time values reported in our dataset are the average ligand binding half-lives of the LOV2 ligand under various intensities of constant blue-light illumination, as measured by separate in vitro kinetic washout experiments. Our model is fit to the steady-state signaling output achieved after a 3 minute exposure of cells to LOV2 ligands of an average ligand binding half-life enforced by constant blue light illumination. We clarify this point by including the following paragraphs beginning on page 8 line 170.
“We are unable to control when binding events start since our optogenetic system is inhibited by blue-light, as opposed to being activated by blue-light. The initiation of binding after blue-light inhibition is a function of both the stochastic relaxation of inhibited LOV2 back into the binding-state as well as the diffusion of binding-state LOV2 from outside the previously illuminated area. Without temporal control over the start of binding, it is difficult to measure the time delay between ligand binding and a downstream signaling event (Yi et al., 2019). Such studies typically require careful single-molecule imaging of numerous stochastic binding events (Lin et al., 2019).
To overcome this technical limitation of our system, we chose instead to measure the steady-state output of the antigen signaling cascade achieved several minutes after ligand binding. Kinetic proofreading systems behave differently than non-proofreading systems at steady-state. A non-proofreading system’s steady-state output is set by the number of ligand-bound receptors and not the binding half-lives of those ligands (Fig. 3D, left). In contrast, a kinetic proofreading system can produce different steady-state outputs in response to ligands of different binding half-lives, even when ligand densities are adjusted to achieve equivalent occupancy (Daniels et al., 2006) (Fig. 3D, right). Signaling events take varying amounts of time to occur after ligand binding (Lin et al., 2019; Yi et al., 2019). However, the temporal delays between steps are on the order of tens of seconds. By imaging the cells after minutes of constant exposure to a set ligand binding half-life, we measure the steady state output achieved at a signaling event in the cascade on a longer timescale than these delays (Tischer & Weiner, 2019).”
- The authors use a simple equation for KP to fit their datasets in Figure 4, equivalently to their previous work. However, no goodness-of-fit metric is provided for these fits, and by manual inspection it is hard to see the defining curves of their KP model in the datasets, especially not for LAT and DAG, where the datasets look much more like vertical bars. The estimated values of steps (n) may well be the best fit to the data, but they are not necessarily a 'good' fit.
To assist readers in assessing how well our models fit our datasets, we have included heatmaps of the residuals from each model fit (Fig 4S3) on page 52, along with discussion (reproduced below) of the residual plots of regions where our models imperfectly capture our dataset on page 13 line 283.
“To assess our model fits, we evaluated the residuals of each model subtracted from their respective dataset. For Zap70 recruitment, our model underestimates the degree of activation at moderate binding half-lives and receptor occupancies, as indicated by the positive region in the center of the heatmap. It is possible that Zap70 recruitment reaches saturation at shorter ligand binding half-lives than our model predicts (Fig. 4S3 A). For both LAT clustering and DAG generation, our models performed poorest in the region of lowest occupancy and shortest half-life (Fig. 4S3 B&C). In this region of our dataset, the fluorescent signal from bound LOV2 above the background fluorescence of unbound LOV2 is smallest. To compensate for fluorescence of unbound LOV2, we subtract off the local background fluorescence of unbound LOV2 around each cell. In doing so we may be underestimating the amount of LOV2 bound to each cell, leading to an underestimation of signaling output by the models. Future studies at LOV2 densities approaching single molecule would better capture this regime of receptor occupancy, but cell-to-cell variation in activation would be too high to be compatible with our current steady-state analysis (Lin et al., 2019).”
- The values of n are also very high, which would imply that the kp rate constant might be very fast to compensate; no estimates of this value are presented. Recent data from the Dushek lab (Pettmann et al, eLife 2021) measured n to be ~3, which seems much more physically realistic. Furthermore, in their previous published work, Tischer & Weiner measured n to be 2.7 for DAG production but in the present study it is now n=11.3, using the same equation
We are unable estimate the kp rate constant, as our datasets are at steady state and do not provide temporal information. To assess the plausibility of our higher n value fits, we explored the steady-state model presented in Ganti et al. PNAS 2020, which defines a kp rate of 0.1 s-1. This model predicts the minimum number of signaling steps required to achieve a defined Hopfield error rate at defined cognate-ligand/self-ligand concentration and half-life ratios. Our exploration of this model is detailed in Fig. 4S4 on page 53 and detailed in discussion on page 14 line 299
“In our previous work our model fit fewer (N=2.7) steps to DAG generation. We now fit a higher number of steps (N=11.3) to DAG generation. This change could be due to the incorporation of ICAM into our current study, which has been shown to potentiate ligand discrimination (Pettmann et al., 2021). Furthermore, our previous antibody-based adhesion may have short-circuited some proofreading steps by irreversibly holding the cell membrane close to the supported lipid bilayer. To evaluate if our higher value fits are indeed the best fit values for our datasets, we fit our model to each dataset while holding the value of N constant in the range of zero to fourteen steps, and evaluated the average residual value for each model fit (Fig 4S3 D). For all signaling steps, the fit value of N was near the minima of average residual and had a lower average residual value than a model with 3 proofreading steps.
To assess the plausibility of a larger number of proofreading steps, we implemented the steady state kinetic proofreading model from Ganti et al. (Ganti et al., 2020). The model estimates the minimum number of proofreading steps required to discriminate between cognate-ligands and self-ligands with different binding half-lives present at a given concentration ratios at a given Hopfield error-rate (Hopfield, 1974). First, we evaluated what combinations of ligand half-lives and concentration ratios an 11-step kinetic proofreading network could discriminate at an error rate less than 10-3 (Fig 4S4 A). We chose the error rate of 10-3, as it is an order of magnitude less than the theorized 10-4 upper limit error rate of the native TCR (Ganti et al., 2020). At moderate half-life ratios, an 11-step network can discriminate cognate peptides present in small concentrations (e.g. 1 cognate-ligand per 1000 self-ligands at a half-life ratio of 6).
In our optogenetic system, the ratio of the average ligand binding half-life between the longest suppressive half-life and the shortest fully activated half-life is about 2. However, an 11-step network is insufficient to discriminate between ligands with a half-life ratio of 2, even at the high ligand ratio of 1 (equal concentrations of cognate- and self-ligand). This suggests our cells are unlikely to be detecting the average ligand binding half-life of each blue-light condition, but are more likely detecting longer-lived binding events from the underlying distribution of half-lives. Another possibility is that our in vitro washout measurements, which measure average ligand binding half-lives of soluble ligands diffusing in three dimensions, differ from the half-lives of ligand-receptor interactions between the cell’s plasma membrane and the supported lipid bilayer diffusing in two dimensions (J. Huang et al., 2010).
To better explore the kinetic proofreading model space, we generated heatmaps reporting the required number of steps to discriminate combinations of ligand and half-life ratios at an error rate of 10-3 (Fig 4S4 B). To discriminate between ligands with a half-life ratio of two, at least 14 steps are needed when the ligands are at equal concentrations, and more than 25 steps are needed if cognate-ligands are 1 per 1000 self-ligands. The required number of proofreading steps decreases rapidly as the half-life ratio increases, reaching a minimum of 8-steps needed for a concentration ratio of 1/1000 and a half-life ratio of 10, which is more in line with physiological half-life ratios between agonist and non-agonist peptides (M. M. Davis et al., 1998).
After comparing our results with the Ganti model, this analysis suggest that our number of fit proofreading steps may be somewhat inflated as a function of our use the average ligand binding half-lives of three dimensional washout experiments in place of the two dimensional single molecule information T cells use to make activation decisions. However, the higher fit N values are more consistent with the required number of steps to discriminate ligands under more physiological conditions than our previous measurements of ~3 steps, which would not be expected to discriminate ligands with half-life ratio of 10 even at a ligand ratio of 1 (Fig 4S4 B, right).”
- If the fitted value of n provides no realistic insight into the KP mechanism, it should not be discussed as though it does.
The many assumptions of our simplistic model likely results in error in determining the absolute number of fit proofreading steps. We feel the strength of our model lies in capturing the relative increase in the strength of proofreading as signal propagates through the cascade, and not determining the absolute number of proofreading steps, though it is comforting that our values are broadly consistent with the expectations of Ganti et al. To highlight the point that relative values are the most important feature of our experiments, we are open to normalizing our n fit values by the fit n of Zap70 for all discussion of our results and the proofreading strength increase shown in Fig 4D if the reviewers think this will better highlight the relative increase in proofreading strength.
- While it is good to confirm it, the result that downstream signaling complexes reset more slowly than distal ones is surely to be expected, given the increased number of steps over which ligand unbinding must traverse, as in their Erlang distribution. You would not expect ERK phosphorylation to decrease at the same rate as LAT cluster dissociation for this same reason. However, the fact that the lifetime of LAT clustering (14.2s) or ZAP70 (9.6s) is so different to LOV2 (3.3s) provides good evidence that it is not proofreading, as by definition the measured outputs should rapidly return to the 'unbound' state in line with ligand unbinding. At least for LAT, there must be a 'memory' from previous signalling lasting several seconds, which means the system has not reset, as required for true KP.
Slower resetting of downstream signaling events in a kinetic proofreading cascade is not a given, as it could be the case that all events reset at the same rate. One requirement for kinetic proofreading is that events in the chain be irreversible on the timescale of the ligand binding half-life. The steps are reset through an orthogonal pathway, opposed to traversing back down a chain of reversible reactions. Both the TCR and LAT are dephosphorylated by the phosphatase CD45, and it would be possible for CD45 to dephosphorylate both proteins at the same rate (or even dephosphorylate LAT faster than the TCR). To clarify this point, we have expanded discussion on possible reset mechanism on page 21 line 451 as reproduced below
“An attractive reset mechanism is the segregation of CD45 away from bound receptors, creating spatial regions in which TCR and LAT associated activating events can occur (S. J. Davis & van der Merwe, 2006). Super-resolution microscopy by Razvag et al. measured TCR/CD45 segregated regions within seconds of antigen contact at the tips of T cell microvilli (Razvag et al., 2018). Upon unbinding these regions of phosphatase exclusion collapse, allowing CD45 to dephosphorylate receptor ITAMs and LAT clusters. However, the rate of dephosphorylation for LAT and receptor ITAMs could differ. LAT clusters exclude CD45 in reconstituted bilayer systems, potentially limiting the dephosphorylation to LAT molecules at the edges of the cluster thus slowing reset (Su et al., 2016). The kinetics of multivalent protein-protein interactions within TCR and LAT clusters can also influence dephosphorylation and dissociation rates (Goyette et al., 2022).
A CD45-mediated reset mechanism would restrict proofreading to membrane-bound signaling events occurring within a CD45-depleted region. Downstream events that dissociate away from the membrane or diffuse out of the segregated region could not directly participate in the proofreading chain, as the collapse of a CD45 segregated region could not reset signaling entities released into the cytosol (e.g. release of IP3 in the cleavage of PIP2 to DAG).”
We also added discussion of recent work from Harris et al. quantifying the slower timescale of Ca++ and ERK reset upon TCR signal termination on Page 23 line 498 as reproduced below.
“Recently Harris et al. quantified the reset rate of the downstream signaling events Ca++ release and ERK phosphorylation upon signal inhibition to be 29 seconds and 3 minutes respectively (Harris et al., 2021). They showed both Ca++ and ERK levels can persist across short inhibitions of signaling. What makes LAT clusters different than these persistent downstream events? The dissolution of LAT clusters is directly triggered by the unbinding of ligand from the TCR, and both the TCR and LAT are de-phosphorylated by CD45. To our knowledge, the rate of ERK dephosphorylation or cytosolic Ca++ depletion are not accelerated by TCR unbinding, and are turned over through constant rather than agonist-gated degradation. A useful future line of inquiry would be to quantify the reset rate for signaling steps throughout the cascade upon ligand unbinding versus orthogonal signal inhibition (e.g. kinase inhibition).”
-
Evaluation Summary:
In this manuscript, a light-gated receptor system (LOV2) linked to T cell receptor signaling machinery is enhanced by addition of an adhesion system enabling robust operation over a larger range of kinetic parameters. This system enables an exploration of how kinetic proofreading processes executed in seconds relate to T cell activation program involving reactions taking minutes to hours.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)
-
Reviewer #1 (Public Review):
Kinetic proofreading is the canonical mechanism that is posited to enable biological systems to discriminate far more finely between inputs than equilibrium considerations alone would suggest. It has also long been studied in the context of ligand discrimination by T cells. A few years ago, Weiner and co-workers published a paper in which they described the development of the light-gated receptor system employed in this paper (LOV2) that enabled studying signaling with controlled receptor-ligand half-lives. However, they had some puzzling results therein, such as the level of proofreading observed at ZAP70. Now, the authors have improved their experimental system using adhesion molecules to stabilize the synaptic junction and obtained the results described in this paper. Their main results are: 1) they can …
Reviewer #1 (Public Review):
Kinetic proofreading is the canonical mechanism that is posited to enable biological systems to discriminate far more finely between inputs than equilibrium considerations alone would suggest. It has also long been studied in the context of ligand discrimination by T cells. A few years ago, Weiner and co-workers published a paper in which they described the development of the light-gated receptor system employed in this paper (LOV2) that enabled studying signaling with controlled receptor-ligand half-lives. However, they had some puzzling results therein, such as the level of proofreading observed at ZAP70. Now, the authors have improved their experimental system using adhesion molecules to stabilize the synaptic junction and obtained the results described in this paper. Their main results are: 1) they can now observe how proofreading steps increase as signaling progresses down the signaling pathway; 2) their results suggest that the signaling pathway resets more slowly as you traverse down the pathway. The paper is clearly written and easy to understand.
-
Reviewer #2 (Public Review):
The manuscript from Britain & Weiner revisit previous experiments from the Weiner group, using optogenetics to gauge the kinetic proofreading (KP) capability of the T cell antigen receptor (TCR) at different parts of the signaling network. By using ICAM1-functionalised lipid bilayers, they are now able to observe some form of proofreading/time delay in the recruitment of ZAP70 to the TCR, which was not the conclusion from their previous work. They then proceed to investigate proofreading at the level of LAT clustering and DAG production by PLCγ1, finding this is indeed the case. A simple mathematical model of KP is then used to fit their datasets to extract an estimate of the number of steps (n) in the KP pathway.
Strengths:
The use of ICAM1-functionalised bilayers has clearly significantly improved the …
Reviewer #2 (Public Review):
The manuscript from Britain & Weiner revisit previous experiments from the Weiner group, using optogenetics to gauge the kinetic proofreading (KP) capability of the T cell antigen receptor (TCR) at different parts of the signaling network. By using ICAM1-functionalised lipid bilayers, they are now able to observe some form of proofreading/time delay in the recruitment of ZAP70 to the TCR, which was not the conclusion from their previous work. They then proceed to investigate proofreading at the level of LAT clustering and DAG production by PLCγ1, finding this is indeed the case. A simple mathematical model of KP is then used to fit their datasets to extract an estimate of the number of steps (n) in the KP pathway.
Strengths:
The use of ICAM1-functionalised bilayers has clearly significantly improved the power of the authors' experiments to study more proximal signaling events in TCR triggering. The result that the LFA-1/ICAM1 interaction is important for efficient T cell engagement to bilayers has been known for some time, though. By essentially redoing many of the same experiments from the previous paper with this new strategy, some form of proofreading can now be observed at the very proximal events of ZAP70 recruitment and corrects the technical deficiency in the previous work. This is a helpful result, and it is good that the authors have revisited this question, as the opposite conclusion drawn from their previous work was rather counter-intuitive.
The data presented also provides good evidence that there are readily measurable time delays between receptor triggering and different downstream outputs within the TCR signal transduction network, which has been hard to measure by other approaches.
Weaknesses:
In the KP concept originally proposed by Hopfield (oddly not cited), it was important that the output response was intimately linked to the bound state of the receptor, in this case the TCR, with ligand unbinding rapidly reversing all proofreading steps. This means that dissociation of a single TCR should disrupt signaling, and implicitly assumes a direct physical connection between the bound receptor and the KP modifications. However, this mechanism becomes much harder to argue when the KP steps are physically uncoupled from bound TCR, such as in LAT microclusters or DAG production. This is because it is possible (and likely) that multiple bound TCRs contribute to the LAT phosphorylation or PLCγ1 activation to produce DAG. The data clearly demonstrate a time delay between receptor binding and the measured outputs, but it is not so surprising that this lag would exist in propagating the signal through the intracellular network. If the authors had measured ERK activation as another readout of TCR engagement, for instance, they almost certainly would have found evidence for proofreading at this point in the network with n>11, but it would be very difficult to reconcile this step as KP proper. Signalling from multiple TCRs can and does lead to ERK activation, which would invalidate proofreading at this step; the same logic surely applies to PLCγ1 activity.
The authors use a simple equation for KP to fit their datasets in Figure 4, equivalently to their previous work. However, no goodness-of-fit metric is provided for these fits, and by manual inspection it is hard to see the defining curves of their KP model in the datasets, especially not for LAT and DAG, where the datasets look much more like vertical bars. The estimated values of steps (n) may well be the best fit to the data, but they are not necessarily a 'good' fit. The values of n are also very high, which would imply that the kp rate constant might be very fast to compensate; no estimates of this value are presented. Recent data from the Dushek lab (Pettmann et al, eLife 2021) measured n to be ~3, which seems much more physically realistic. Furthermore, in their previous published work, Tischer & Weiner measured n to be 2.7 for DAG production but in the present study it is now n=11.3, using the same equation. If the fitted value of n provides no realistic insight into the KP mechanism, it should not be discussed as though it does.
While it is good to confirm it, the result that downstream signaling complexes reset more slowly than distal ones is surely to be expected, given the increased number of steps over which ligand unbinding must traverse, as in their Erlang distribution. You would not expect ERK phosphorylation to decrease at the same rate as LAT cluster dissociation for this same reason. However, the fact that the lifetime of LAT clustering (14.2s) or ZAP70 (9.6s) is so different to LOV2 (3.3s) provides good evidence that it is not proofreading, as by definition the measured outputs should rapidly return to the 'unbound' state in line with ligand unbinding. At least for LAT, there must be a 'memory' from previous signalling lasting several seconds, which means the system has not reset, as required for true KP.
-
