Phosphorylation Site Dynamics of Early T-cell Receptor Signaling

In adaptive immune responses, T-cell receptor (TCR) signaling impacts multiple cellular processes and results in T-cell differentiation, proliferation, and cytokine production. Although individual protein–protein interactions and phosphorylation events have been studied extensively, we lack a systems-level understanding of how these components cooperate to control signaling dynamics, especially during the crucial first seconds of stimulation. Here, we used quantitative proteomics to characterize reshaping of the T-cell phosphoproteome in response to TCR/CD28 co-stimulation, and found that diverse dynamic patterns emerge within seconds. We detected phosphorylation dynamics as early as 5 s and observed widespread regulation of key TCR signaling proteins by 30 s. Development of a computational model pointed to the presence of novel regulatory mechanisms controlling phosphorylation of sites with central roles in TCR signaling. The model was used to generate predictions suggesting unexpected roles for the phosphatase PTPN6 (SHP-1) and shortcut recruitment of the actin regulator WAS. Predictions were validated experimentally. This integration of proteomics and modeling illustrates a novel, generalizable framework for solidifying quantitative understanding of a signaling network and for elucidating missing links.


Modeling method
To match the level of detail present in the data (population-averaged phosphorylation levels of individual amino acid residues), reactions must factor in specific protein sites, which increases the number of species that must be considered. For example, if one considers three sites of phosphorylation in the kinase LCK (Y192, Y394, and Y505, all of which are considered in our model), the molecule has 2 3 = 8 possible phosphorylation states. When the multiple binding and phosphorylation states of all proteins of interest are considered, the large number of possible species is an obstacle to formulation of a model via traditional means, such as coupled ordinary differential equations. An alternative means of model specification that captures chemical kinetics like a traditional model but is better suited for modeling with site-specific resolution is rulebased modeling. In this approach, a domain-specific language is used to define local rules for biomolecular interactions. The modeling language that we have used is the BioNetGen language 1. Crosslinking reagents Jurkat T cells were stimulated with anti-CD3, anti-CD28, and secondary antibodies, which are commonly used reagents for experimental stimulation of T cells [91] because antibody-mediated crosslinking recapitulates intracellular signaling events that lead to cellular activation [23].
These antibodies trigger signaling through clustering of TCR and CD28 molecules, which mimics the outcome of interactions between a T cell and an antigen-presenting cell. These antibodies have the potential to induce large receptor/co-receptor oligomers, but as a simplification that captures the essential action of these stimuli, we model them as three virtual ligands that induce CD28 homodimers (Ligand 1), CD28/TCR heterodimers (Ligand 2), and TCR homodimers (Ligand 3). Molecule type definitions: aCD28,aCD3) c. Lig3(aCD3,aCD3) The components aCD3 and aCD28 are considered to be binding sites that interact with CD3 and CD28, respectively. 2. TCR/CD3 complex.

Interactions in Model
Receptor stimulation 1. CD28 cross-linking. Ligand-induced clustering of CD28 complexes was modeled as follows: The first rule represents initial binding and the second rule represents cross-linking.
3. CD3/CD28 cross-linking. Ligand-induced co-clustering of TCR and CD28 was modeled as follows: The first rule represents interactions between NCK and a TCR that is unphosphorylated at Y188 of CD3E. The second rule represents dissociation of NCK induced by phosphorylation of Y188 [36]. We assume that phosphorylation of Y188 takes place even when NCK is bound (i.e., phosphorylation is not inhibited) because the phosphorylation dynamics of this site is comparable to that of other ITAMs in our data (see Supplementary Fig. S6).  S6). To capture fast recruitment of PTPN6 to the membrane, we assume that it binds phosphotyrosines in the TCR δ and γ ITAMs. We note that interaction of PTPN6 with these sites has not been demonstrated. However, given the large number of phosphosites in the TCR, we take these sites to represent phosphotyrosines that may be bound by PTPN6 without creating competition with ZAP70. PTPN6 may be recruited through other interactions. For example, PTPN6 has recently been found to interact with CRKL [182]. It can also bind the SH2 domain of LCK (see Interaction 24 below); however, this interaction requires phosphorylation of PTPN6 Y566 and therefore cannot account for recruitment of unphosphorylated PTPN6. Recruitment of unphosphorylated PTPN6 is needed for the level of phosphorylated PTPN6 to increase quickly upon stimulation of TCR/CD28 signaling. A role for PTPN6 is supported by a direct test of a model prediction via RNAi knockdown (see Fig. 3) as well as our finding that Y192 in LCK is a direct substrate of PTPN6, at least in vitro (Fig. S9).
Interactions among receptor-proximal signaling proteins 20. LCK trans-phosphorylates LCK. LCK can be activated by autophosphorylation of Y394 in the kinase domain [193,194,195], which occurs after CD28 binding [196]. Mutations of this site reduce kinase activity [197,198,199,200 The first rule characterizes phosphorylation by autoinhibited LCK (phosphorylated on the inhibitory tyrosine 505), and the second rule characterizes phosphorylation by activated LCK (unphosphorylated on tyrosine 505). Rules a and c characterize phosphorylation by autoinhibited LCK (phosphorylated on the inhibitory tyrosine 505), and Rules b and d characterize phosphorylation by activated LCK (unphosphorylated on tyrosine 505). Rules a and b differ from Rules c and d in that PTPN6 is taken to be associated with CD3D in the former rules and with CD3G in the latter rules.
24. LCK binds PTPN6. Y566 is located in a consensus sequence for LCK binding. Mutation of this site to phenylalanine abolishes interactions between LCK and PTPN6 [32].
31. PTPN6 dephosphorylates PAG1 Y163. PAG1 is observed to be rapidly dephosphorylated following TCR/CD28 co-stimulation. We hypothesize that PAG1 is a substrate of PTPN6. We found that PTPN6 knockdown (KD) resulted in a decrease in phosphorylation of ZAP70 Y493, which is a target of LCK, supporting a a positive role for PTPN6 in early TCR signaling. In the model, the positive effect of PTPN6 KD arises partly because dephosphorylation of PAG1 limits colocalization of LCK and CSK on PAG1 and CSK-mediated phosphorylation of LCK 505 (and LCK autoinhibition). LCK autoinhibition could also be limited if PTPN6 acted directly on LCK pY505. However, it seems that PTPN6-mediated dephosphorylation of pY505 is limited because of intramolecular interaction of pY505 with the LCK SH2 domain [39]. We note that PTPN11 (SHP-2), which is closely related to PTPN6, has been found to act on sites in PAG1 32. PTPN6 dephosphorylates DOK1 pY449. DOK1 pY449 follows a similar time course as other sites of dephosphorylation (see Fig. 2). Thus, we hypothesize that it is a substrate of PTPN6, although other phosphatases may be involved (see notes on Interaction 12). DOK1 has been found to be dephosphorylated by PTPN6 [212], but Y449 has not been specifically identified as a substrate.