SH3-domain mutations selectively disrupt Csk homodimerization or PTPN22 binding

The kinase Csk is the primary negative regulator of the Src-family kinases (SFKs, e.g., Lck, Fyn, Lyn, Hck, Fgr, Blk, Yes), phosphorylating a tyrosine on the SFK C-terminal tail that mediates autoinhibition. Csk also binds phosphatases, including PTPN12 (PTP-PEST) and immune-cell PTPN22 (LYP/Pep), which dephosphorylate the SFK activation loop to promote autoinhibition. Csk-binding proteins (e.g., CBP/PAG1) oligomerize within membrane microdomains, and high local concentration promotes Csk function. Purified Csk homodimerizes in solution through an interface that overlaps the phosphatase binding footprint. Here we demonstrate that Csk can homodimerize in Jurkat T cells, in competition with PTPN22 binding. We designed SH3-domain mutations in Csk that selectively impair homodimerization (H21I) or PTPN22 binding (K43D) and verified their kinase activity in solution. Disruption of either interaction in cells, however, decreased the negative-regulatory function of Csk. Csk W47A, a substitution previously reported to block PTPN22 binding, had a secondary effect of impairing homodimerization. Csk H21I and K43D will be useful tools for dissecting the protein-specific drivers of autoimmunity mediated by the human polymorphism PTPN22 R620W, which impairs interaction with Csk and with the E3 ubiquitin ligase TRAF3. Future investigations of Csk homodimer activity and phosphatase interactions may reveal new facets of SFK regulation in hematopoietic and non-hematopoietic cells.


Results
Multiple molecules of epitope-tagged Csk co-immunoprecipitate from Jurkat-cell lysates. We investigated Csk self-association in Jurkat T cells by co-transfecting HA-tagged Csk (Csk HA ) and Myc-tagged Csk (Csk Myc ) and subjecting cell lysates to anti-Myc immunoprecipitation. Csk HA was detected in Csk Myc immunoprecipitates, suggesting interaction of Csk HA and Csk Myc proteins (Fig. 1a, Supplementary Fig. S1). Blots of whole-cell and immunodepleted lysates demonstrated the specificity of the HA and Myc antibodies and the efficiency of Myc immunoprecipitation (Fig. 1b). Lack of detectable Csk HA depletion from Myc-immunodepleted lysates (i.e., substoichiometric binding of Csk HA to Csk Myc ) is consistent with the moderately low affinity and fast on/off kinetics of dimerization in solution 43 . Other factors that could limit Csk HA co-immunoprecipitation include competition of homodimerization with binding of endogenous PTPN22 and signal interference from other dimer pairings, such as Csk Myc /Csk Myc , Csk HA /Csk HA , and Csk Myc /endogenous Csk.
Although the footprints of these interfaces overlap substantially, we identified unique interactions via structural alignment of the dimer and extended PTPN22 binding surfaces from published crystal and NMR structures 13,44,51 . A single residue, H21, was situated in the symmetrical homodimer interface and outside the PTPN22 binding footprint (Fig. 2b). Analysis with PyMol software suggested that most amino acids could adopt non-clashing rotamers if swapped into this position; β-branched isoleucine was an exception. We therefore generated H21I variants of Csk Myc and Csk HA and tested their effect on Csk co-immunoprecipitation.
We transiently transfected Jurkat cells with pairs of Csk Myc and Csk HA constructs, both either WT or H21I. Protein expression was not altered by H21I substitution, and the Myc-tagged construct was effectively immunodepleted (Fig. 2c, Supplementary Fig. S2). H21I substitution disrupted Csk HA co-immunoprecipitation with Csk Myc (60 ± 20% less HA-tagged H21I than WT) (Fig. 2d,e). This deficit was accompanied by increased coimmunoprecipitation of endogenous PTPN22 (80 ± 20% more in H21I-transfected cells than in WT-transfected cells). Together, these data suggest that Csk dimerization can compete with PTPN22 binding. This competition also reveals that the uniquely extended SH3-binding motif in the PTPN22 peptide 10 cannot bind stably to the SH3 domain of Csk in the absence of canonical PXXP-motif docking.
Although a previous study probed the SH3 homodimer interface using point mutations 43 , the authors did not test the effect of H21I substitution on dimerization. We purified recombinant Csk WT and H21I from bacteria and assessed their mobility on a size exclusion column. As expected, purified Csk WT and H21I were each visible as a single SDS PAGE band of ~ 50 kDa, consistent with the 50.6 kDa molecular weight of full-length Csk (Fig. 3a inset). Csk H21I, however, was retained longer than WT on a size exclusion column (Fig. 3a) www.nature.com/scientificreports/ a decrease in apparent molecular weight 44,53,54 . The Csk H21I retention volume was consistent with a monomeric species, confirming disruption of the homodimer interface ( Fig. 3b) 43,52,53 . The more rapid migration of WT Csk through the column corresponded to a molecular weight of ~ 70 kDa, smaller than a stable dimeric species of 101 kDa. This lower apparent molecular weight is consistent with a rapidly exchanging population of monomers and dimers after dilution of the sample in the column 43 . Together the co-immunoprecipitation of Csk HA with Csk Myc and the observation that a structure-guided point mutation selectively disrupts this interaction, in cells and in purified protein, lead us to conclude that Csk homodimerizes in Jurkat T cells.
An amino-acid substitution in the Csk SH3 domain selectively impairs PTPN22 binding. The homodimer interface in the Csk SH3 domain overlaps completely with the canonical PXXP-binding site shared by PTPN12 and PTPN22. A second point of interaction with Csk, however, is unique to PTPN22 (LYP in humans, Pep in mice) 43 . Structural modeling of this extended interaction suggested that K43 in the Csk SH3 domain uniquely participates in PTPN22 binding but not homodimer formation (Fig. 4a). Since this positively charged residue forms a salt bridge with a nearby aspartate in PTPN22, we flipped the charge and minimized the degrees of rotameric freedom with a K43D substitution. Endogenous PTPN22 co-immunoprecipitated poorly with Csk K43D (70% ± 30% less than WT), without a significant secondary defect in Csk HA co-immunoprecipitation (Fig. 4b,c, Supplementary Fig. S3). The selectivity of the H21I and K43D substitutions for impairing Csk or PTPN22 co-immunoprecipitation, respectively, also demonstrate the adequate stringency of the co-immunoprecipitation protocol for detecting specific protein-protein interactions.
We also generated Myc-and HA-tagged constructs of Csk W47A, which has been used previously to ablate the binding of PXXP-containing ligands, including PTPN22, to the SH3 domain of Csk [46][47][48][49][50] . Residue W47 forms one of the proline binding pockets for the canonical PXXP interaction and lies within the homodimerization footprint (Fig. 4a). Co-immunoprecipitation experiments confirmed that the Csk W47A substitution impaired both Csk/Csk and Csk/PTPN22 interactions (Fig. 4d,e). This suggests that a defect in Csk homodimer formation could be a complicating factor in previous studies of PTPN22 and other SH3-binding proteins; going forward, the K43D mutation could be a more specific tool for disrupting Csk binding to PTPN22.
H21I and K43D substitutions do not impair the kinase activity of Csk in solution but may alter cellular function. Full-length Csk WT, H21I, K43D, and K222R (a kinase-impaired lysine-to-arginine mutant) 3 proteins were purified and tested for activity against an optimal substrate peptide 54 . Activity was measured using a continuous spectrophotometric assay coupling ATP hydrolysis to NADH oxidation and decreasing absorbance at 340 nm (Fig. 5a) 55 . Neither the H21I substitution nor K43D impaired kinase activity in solution (Fig. 5b). As expected, Csk K222R was nearly kinase dead (94% ± 5% loss of activity compared to WT). www.nature.com/scientificreports/ As an initial test of homodimer function in cell signaling, we measured the ability of Csk H21I to suppress Lck-dependent TCR triggering 16 . Signaling was initiated by treating Csk Myc transient transfectants with the C305 antibody, which ligates the Jurkat TCR 56 . After 2 min, we quenched signaling and performed intracellular staining for Myc (Csk-construct expression) and phosphorylated Erk1/2 threonine 202/tyrosine 204 (pErk, reflecting TCR-pathway activation 16 ). We then performed flow cytometry to probe the effect of Csk expression on TCR signaling. In the heterogenous pool of transient transfectants, we observed a complete loss of pErk positivity in cells with the highest expression of WT Csk. (Fig. 6a). As induction of pErk downstream of the TCR in a given cell is an all-or-none response 57 , quantification of the fractional content of pErk-positive ( + ) cells at each Csk dose was well described by a sigmoidal function. We therefore fit each data set to a dose-response curve to generate an apparent IC 50 value, a relative measure of the Csk dose that reduced the frequency of pErk + cells by 50% (Fig. 6b). Of the constructs tested, WT Csk suppressed TCR signaling most efficiently (Fig. 6c). At the other extreme, Csk K222R was unable to fully suppress TCR signaling, even at the highest expression levels. Csk H21I and Csk K43D both had more subtle but still significant functional defects, with apparent IC 50 values increased relative to WT (120% ± 10% and 130% ± 20%, respectively) ( Fig. 6d). Together, these data suggest that blocking homodimerization or PTPN22 binding decreases the immunosuppressive function of Csk in cells, an effect that cannot be attributed to a simple loss of intrinsic catalytic function.

Discussion
We present the first cellular evidence that Csk can form homodimers that compete with PTPN22 (LYP/Pep) binding. We used structural analysis to design single-amino-acid substitutions in the SH3 domain of Csk that disrupt either homodimer formation (Csk H21I) or PTPN22 binding (K43D) without impairing kinase activity in solution. We also tested the substitution commonly used to block SH3-domain PXXP binding (Csk W47A) and found a secondary effect of impairing Csk homodimer formation. The PTPN22-specific Csk K43D construct will therefore be a useful tool for selective studies of homodimer vs. phosphatase interactions in cell signaling. Together, the H21I and K43D mutations demonstrate that PTPN22 binding and Csk homodimer formation compete for access to the Csk SH3 domain. This may indicate that multivalent recruitment of Csk to adaptors and other binding partners at the plasma membrane controls the residency of phosphatases alongside their membrane-localized substrates. www.nature.com/scientificreports/ Future studies of Csk at physiological expression levels and in primary (non-cancer) cells will be necessary for quantitative analysis of homodimer formation. However, even at physiological expression levels Csk is likely at sufficiently high local concentration to drive dimerization 58 . CBP/PAG1 oligomerizes upon phosphorylation 1,4-6 , and the signaling complexes in the early TCR signalosome are highly oligomeric and bridged. In our experiments, moreover, endogenously expressed proteins would recruit and concentrate Csk at the membrane, limiting the effect of Csk overexpression on access to substrate.
The effect of homodimerization on Csk function remains an interesting question. We found no evidence that Csk H21I had impaired kinase activity, but our analysis was performed at a concentration below the likely K d of homodimer formation. Complex allosteric networks regulate many tyrosine kinases, including Csk, in which the phosphopeptide binding to the SH2 domain is activating 1,5,45 . SH3-SH3 homodimer interactions could similarly modulate Csk kinase activity.
Conversely, loss of either homodimerization or PTPN22 binding decreased the ability of overexpressed Csk to suppress Lck/TCR signaling. We do not yet know whether Csk H21I can form homodimers in the highestexpressing transfectants or whether a single molecule of Csk H21I may pair with endogenous Csk (with loss of only one of the two histidine residues in the symmetrical interface). Future studies without competition from endogenous Csk will be necessary to define these properties and elucidate the physiological function of Csk homodimer formation. The role of Csk-bound PTPN22 is debated, and we cannot definitively address this question in this study. However, our observations that loss of homodimerization (with increased PTPN22/Csk interaction) or loss of PTPN22 binding may impair Csk function in T cells does not support a Csk-centered gain-of-function model for PTPN22 R620W.
The dual requirement for homodimerization and PTPN22 binding is particularly interesting because the homodimer and PTPN22 are likely incapable of forming a ternary complex. This suggests that different pools of Csk or a kinetic process of swapping one binding partner for the other is required for optimal Csk function. The more severe defect in K222R function and kinase activity relative to Csk H21I and K43D reaffirms that catalytic activity is the dominant requirement for the inhibitory function of Csk. However, homodimer and PTPN22 interactions with Csk may coordinate to maximize suppression of Lck and the TCR pathway.
We speculate that Csk homodimerization could be a key step in cell signaling. At equilibrium and relatively low concentration, the Csk SH3 domain could bind preferentially to phosphatase. In T cells, half of all PTPN22 molecules are associated with Csk, so this would be a substantial proportion of the available phosphatase  www.nature.com/scientificreports/ activity 19 . Upon recruitment to phosphorylated CBP/PAG1 and the TCR signaling complex, the local concentration of Csk would increase due to enrichment in microdomains and CBP/PAG1 oligomerization. Csk homodimers might then outcompete PTPN22 for access to the SH3 domain. Released near its substrates, PTPN22 would then be free to dephosphorylate and block signaling through Lck, ITAMs, and Zap70. In this gain-of-function mode, the catalytically efficient phosphatase could act on many more substrate molecules after dissociation from the longer-lived Csk/CBP/Lck complex. Alternatively, in a loss-of-function mode, PTPN22 might bind an www.nature.com/scientificreports/ alternative ligand (for instance, TRAF3, Fak or Pyk2 kinase 6 , or a Dok-family adaptor protein 7 ) and lose access to substrates. It is also possible that a secondary effect of the autoimmunity risk allele PTPN22 R620W could be to increase Csk homodimer formation via loss of PTPN22 competition for SH3-domain access. Future studies will be required to address these questions. Our structural observations also predict a functional difference between PTPN12 and PTPN22 phosphatases, possibly leading to differential regulation of SFK signaling in non-hematopoietic and hematopoietic cells. The PXXP-only PTPN12/Csk interaction, for instance, might be easily outcompeted by homodimer formation at a modest local concentration of Csk. PTPN12 binds to Csk only 1-10-fold more tightly than would another Csk molecule, and the PTPN12 binding site is completely occluded by the Csk/Csk interface, preventing rebinding. Hematopoietic-cell PTPN22, in contrast, might be more difficult to outcompete. PTPN22 binds to Csk 10-100fold more tightly than would another molecule of Csk, and part of the PTPN22 binding site remains exposed even when Csk homodimer is formed. This could facilitate PTPN22 rebinding to Csk and require higher local concentrations of Csk for phosphatase release. These dynamics, in turn, could regulate the signaling thresholds or feedback regulation of the SFKs by cell type, by local concentrations of Csk-binding proteins at the plasma membrane, and by Csk expression itself.
In summary, we report that Csk can homodimerize in Jurkat T cells, in competition with the phosphatase PTPN22. We also present the Csk variants H21I and K43D as new tools for uncoupling these binding processes. Csk expression levels, activity, and localization are all important regulators of signaling 42,59-66 and can be disrupted in autoimmune disease 25 and cancer [22][23][24] . The effect of these variables on homodimerization and PTPN22 binding will be interesting avenues for future mechanistic analysis. Size exclusion chromatography. Size exclusion chromatography was performed by loading 1 ml (20 mg, 400 µM) purified Csk (WT or H21I) onto a HiLoad 16/60 Superdex 200 column. Retention volumes were obtained by fitting absorbance (280 nm) peaks to Gaussian curves (absorbance = amplitude × exp(− 0.5 × ((volume − mean retention volume)/standard deviation) 2 )). Mean retention volumes were used to calculate partition coefficients from the function (retention volume − void volume)/(column volume − void volume). The elution of blue dextran marked the void volume, and elution of high-salt buffer (measured by conductance) marked the column volume. To calculate apparent molecular weights, partition coefficients of standard proteins from the high-molecular-weight gel filtration calibration kit (GE Healthcare #28403842) were plotted against Log(MW) and fit to a linear function using GraphPad Prism software.

Materials and methods
Kinase activity assay. The activity of purified Csk was measured using a continuous spectrometric assay in which ATP hydrolysis is coupled via pyruvate kinase and lactate dehydrogenase to NADH oxidation, which results in a decrease in absorbance at 340 nm 55,70  Cell stimulation and flow cytometry. We quantified Erk phosphorylation as a readout of Jurkat-cell signaling downstream of the TCR and Lck, which is an all-or-none response in each cell 57  www.nature.com/scientificreports/ and Alexa488-Goat-anti-Mouse IgG (H + L) (Life Technologies A-11029). After washing and resuspension in FACS buffer, data were collected on a BD Fortessa flow cytometer. Compensation was performed using FAC-SDiva software with unlabeled cells, cells treated with Pacific Blue or Pacific Orange, and cells labeled with anti-human CD45 (APC, 2D1, eBioscience, 17-9459 and A488, H130, BioLegend, 304019). TCR expression on transfected Jurkat cells was assessed by staining with phycoerythrin (PE)-conjugated mouse anti-human CD3 (BD Pharmingen 555333); TCR expression was not affected by Csk overexpression.
Apparent IC 50 of Csk. Flow cytometric data were analyzed with FlowJo software (Tree Star, Inc.). The live cell population was selected by forward/side scatter, and cells expressing Csk Myc were gated using Alexa488 fluorescence. Populations positive and negative for pErk were gated using APC fluorescence, and a histogram was generated with the number of pErk + and pErkcells within each Csk expressionl bin. Microsoft Excel was used to calculate % pErk + cells in each bin, and the results were plotted using Graphpad Prism software. Data  www.nature.com/scientificreports/