Growth Factor Receptor-bound Protein 2 SH2/SH3 Domain Binding to CD28 and Its Role in Co-signaling*

The co-stimulatory antigen CD28 has been shown to bind to several intracellular proteins including phosphatidylinositol 3-kinase, growth factor receptor-bound protein 2 (Grb2), and ITK. Paradoxically, Grb2 and phosphatidylinositol 3-kinase binding has been mapped to a similar pYMNM motif within the CD28 cytoplasmic tail. Given the importance of CD28 co-signaling to T cell function, questions exist regarding the mechanism by which Grb2 binds to CD28, and whether the interaction plays a role in co-stimulation. To biochemically characterize Grb2/CD28 binding, we initially utilized glutathioneS-transferase-Grb2 fusion proteins carrying inactivating mutations within the SH2 and SH3 domains of Grb2, and assessed their ability to bind to CD28. In vitro binding experiments indicated that the Grb2 SH2 domain is critical for the association, while the SH3 domain plays an additional role in facilitating optimal binding. Enhanced binding via the SH3 domains was not observed when the C-terminal PXXP motif within CD28 was disrupted, thereby indicating that both SH2 and SH3 domains contribute to CD28 binding. Mutations that alter Grb2 binding were found to block the CD28-dependent interleukin-2 production. Further, tyrosine phosphorylation of Vav and the costimulation-dependent activation of Jun N-terminal kinase was blocked in cells defective in CD28/Grb2 binding. These results provide evidence for an alternate CD28-mediated signaling process involving Grb2 binding to the co-receptor.

Optimal activation of resting T cells requires a second CD28mediated co-signal in addition to T cell receptor/CD3 ligation (reviewed in Refs. 1 and 2). Two ligands termed CD80 (B7-1) and CD86 (B7-2) have been found to bind to CD28 (3). CD28 ligation regulates lymphokine production including interleukin 2 (IL-2) 1 (4), prevents T cell anergy (5) and apoptosis (6), and may regulate the balance between Th1 and Th2 responses (7). Although extensively studied, the manner by which CD28 generates co-signals remains to be fully understood. CD28 ligation leads to phosphorylation of the nonenzymatic cytoplasmic do-main on tyrosine residues (8 -10). Co-expression studies have identified Src-related kinases Lck and Fyn as tyrosine kinases capable of phosphorylating this receptor (11,12). Phosphorylation at the cytoplasmic motif pYMNM allows for binding to PI 3-kinase (8,13,14) and Grb2 (15). ITK can also bind CD28 (16,17), but at a site distinct from the pYMNM motif (11). CD28 has also been linked to the activation of pp70 S6 kinase (18), Ras-MAPK/Erk signaling pathway (19), and sphingomyelinase (20,21). Further, stimulation of CD28 in the presence of T cell receptor ligation was shown to cause the activation of JNK/ stress-activated protein kinase (22). In fibroblasts, Vav was found, when tyrosine phosphorylated, to function as a guanine nucleotide exchange factor for Rac and result in the activation of JNK (23,24). Thus the mechanism of CD28-dependent JNK activation may involve the phosphorylation of Vav and subsequent activation of Rac.
Grb2 is an adapter protein, composed of two SH3 domains and an intervening SH2 domain (25,26). This protein plays a key role in the regulation of p21 ras activation by recruiting SOS, a guanine nucleotide exchange factor, to growth factor receptors (26,27). In this case, specificity is determined by SH2 domain binding to a phosphorylated tyrosine residues within the motif pYXNX (28). It has also been reported to bind cytoplasmic signaling molecules such as Shc (29), Cbl (30), protein tyrosine phosphatase 1C (31,32), PTP1D (33), SH2 domaincontaining leukocyte protein 76 (34), p85 of PI 3-kinase (35), and Vav (36). In certain of these instances such as with Grb2/ Vav binding, the SH3 domain of Grb2 binds to the proline based motif PXXP (37). In the case of Grb2 binding to receptor protein-tyrosine phosphatase ␣, tandem SH2/SH3 domain binding has been reported, although the SH3 domain appears to bind indirectly (38).
Given this background, Grb2 binding to CD28 could connect this co-receptor to a number of downstream signaling events (15). A potential difficulty with the interaction concerns the fact that Grb2 shows some constitutive binding to CD28, and the SH2 domain binds to the pYMNM motif, but with some 10 -100-fold lower affinity than PI 3-kinase (15). Despite this difference, the binding affinity of Grb2 for CD28 is comparable to other biologically relevant interactions such as between Grb2 and Shc (15). In the present study, we have further characterized the molecular basis of Grb2/CD28 binding by providing data that Grb2 uses both its SH2 and SH3 domains in binding to CD28 at pYMNM (residues 191-194) and PXX-PXR (residues 208 -213), respectively. Moreover, the potential importance of Grb2 binding is underlined by the blocking effect of the N193Q mutation on CD28-induced enhancement of IL-2 production. The same mutation was also found to abolish tyrosine phosphorylation of Vav and activation of JNK, raising the possibility that the requirement of Grb2 binding is linked to activation of Vav and JNK in costimulatory activation in some T-cells.
GST Precipitation, Immunoprecipitation, and Western Blotting-Sf21 cells were harvested 40 -64-h postinfection, lysed in the Triton lysis buffer (1% Triton X-100, 40 mM Tris (pH 8.0), 150 mM NaCl, 5 mM EDTA, 2 mM sodium orthovanadate, 2 mM sodium fluoride, 2 mM phenylmethylsulfonyl fluoride) and centrifuged for 10 min at 10,000 ϫ g. The supernatants were incubated with either GST fusion protein (1 g) or anti-CD28 (4B10) and rabbit anti-mouse for 1 h at 4°C. Glutathione-Sepharose or protein A-Sepharose was added and rocked for another hour at 4°C. The precipitates were washed five times with the lysis buffer, subjected to SDS-PAGE, transferred to a nitrocellulose membrane, and immunoblotted with indicated antibodies as described (41).
In Vitro Kinase Assay and Reprecipitation Analysis-The precipitates were subjected to kinase reaction as described (40), resolved by SDS-PAGE, and then transferred to a nitrocellulose membrane. The band corresponding to CD28 was excised from the membrane and resolved in 100 l of dimethyl sulfoxide. 900 l of the Triton lysis buffer were added, and vigorous vortexing was followed. The supernatant was collected after centrifugation for 15 min at 10,000 ϫ g and incubated with GST fusion proteins for reprecipitation. The reprecipitated pellet was washed three times in the lysis buffer, separated by SDS-PAGE, and autoradiographed. For Fig. 1E, Western blotting with anti-Tyr(P) replaced the in vitro labeling.
JNK Activity Assays-25 ϫ 10 6 cells were electroporated with 40 g of pEBG-JNK1 at 260 V and 960 F (Bio-Rad). 20 -40 h after transfection, cells were stimulated and disrupted in the Triton lysis buffer. Cleared lysates were then precipitated with glutathione-Sepharose beads. The precipitates were washed five times in the lysis buffer and subjected to kinase reaction with 3-5 g of purified GST-c-Jun as a substrate for 20 min at RT. Phosphorylation of c-Jun was visualized by autoradiography.

RESULTS AND DISCUSSION
We have previously reported that CD28 associates with Grb2 in human T cells (15). Further analysis using various CD28 mutants identified the Tyr 191 residue within the YMNM motif as crucial to this interaction (15,39). However, in reconstitution studies, substitution of Tyr 191 did not completely abolished Grb2 binding, suggesting that mechanism(s) other than SH2 domain binding may facilitate the interaction (39). In an effort to investigate this issue further, GST-Grb2 fusion proteins containing single residue mutations in the SH2 (GST-Grb2-mSH2; R86K), SH3 (GST-Grb2-mSH3; P49L and P206L), or both domains (GST-Grb2-mSH2/3; R86K, P49L, and P206L) were used in precipitation studies. Lysates from insect cells co-infected with baculoviruses carrying the human CD28 and the Lck cDNAs were precipitated and examined for a CD28 association with Grb2. We previously showed that p56 lck can phosphorylate CD28 at pYMNM and induce binding to PI 3-kinase (11). Under these conditions, wild-type GST-Grb2 precipitated a broad ϳ33 kDa band, as detected by antiphosphotyrosine immunoblotting, which co-migrated with a polypeptide precipitated by anti-CD28 monoclonal antibody (Fig. 1A, lanes 5 and 6). Western blotting with anti-CD28 monoclonal antibody confirmed the identity of this ϳ33-kDa protein as CD28 (Fig. 1B). As expected, Grb2-WT was considerably more efficient in precipitating CD28 than GST-Grb2-mSH2 (Fig. 1A, lanes 2 and 5). Nevertheless, the SH2 domain mutant did show some precipitation of Grb2 (lane 2), and mutations within the SH3 domains significantly decreased the precipitation of CD28 (lanes 3 and 5). Each of the GST preparations were routinely monitored for degradation, and equivalent amounts of protein was used in each precipitation. Densitometric analysis showed that CD28 recognition was reduced by 2.0-and 9.0-fold by inactivating the Grb2-SH3 and Grb2-SH2 domains, respectively (lower histogram). Neither GST nor GST-Grb2-mSH2/3 bound CD28 (lanes 1 and 4). As an additional negative control, neither Lck nor wild type baculovirus infection of cells allowed for precipitation of CD28 (lane 7 and data not shown).
The ability of mutations in both the SH2 and SH3 domains to reduce precipitation efficacy suggested that both domains might play a role in recognizing CD28. It was possible that the CD28/Grb2 association was mediated by a third molecule endogenous to insect cells. To exclude this latter possibility, and as a further control for the specificity of Grb2 recognition, [␥-32 P]ATP-labeled CD28 was eluted from the nitrocellulose membrane and subjected to reprecipitation. Under these conditions, GST-Grb2 reprecipitated CD28 (Fig. 1C, lane 5), and moreover, differential binding efficiency of the various Grb2 proteins was preserved (Fig. 1C, lanes 2-5, upper and lower panels). This result indicates that the Grb2 can bind directly to CD28, and that the full-length Grb2 protein binds more efficiently than the SH2 domain alone.
To verify the specificity of the Grb2 SH2 domain for CD28 binding, SH2 domains from other signaling molecules were added to the analysis. The Grb2 SH2 domain (Grb2-mSH3) showed greater binding than the SH2 domains of phospholipase C␥ N-terminal, Fyn, and Lck (Fig. 1D). By contrast, the C-terminal SH2 domain of p85 bound CD28 more efficiently than Grb2, which is consistent with the results of our previous peptide binding studies (14,15).
The direct binding of Grb2 to CD28 was also examined using in vivo tyrosine phosphorylated CD28. For this assay, anti-CD28 precipitates from CD28/Lck infected cell lysates were resolved by SDS-PAGE, and the CD28 band was eluted and reprecipitated with GST-Grb2. Lane 1 in Fig. 1E shows a portion of the eluted CD28. Western blotting with anti-phosphotyrosine revealed reprecipitation of CD28 under these conditions (Fig. 1E, lane 3). Reprobing the membrane with anti-CD28 confirmed the identity of the reprecipitate as CD28 (data not shown). These data indicate that the Grb2 SH2 could recognize in vivo phosphorylated CD28 in a direct and specific manner.
The greater binding of wild type Grb2 relative to Grb2 with mutated SH3 domains suggested an involvement of the SH3 domain(s) in the Grb2/CD28 association. In this context, anal-ysis of the CD28 cytoplasmic tail revealed two PXXP motifs (residues 196 -199 and 208 -211), the minimal consensus sequence for SH3 domain binding ( Fig. 2A). As the first PXXP motif is not conserved in rat CD28 and the conserved second PXXP motif resembles more closely one of the two classes of proline-rich motifs (PXXPXR) (37), we tested the possibility that the second PXXP motif of CD28 interacts with the Grb2 SH3 domain. For this, alanine residues were substituted for prolines at residues 211 and 212 (CD28-Ala 211 Ala 212 ). Binding of CD28-P211A/P212A to the various GST-Grb2 fusion proteins was then compared with that of CD28-WT. As shown in Fig.  2B, binding of Grb2-WT to the CD28-P211A/P212A mutant was greatly reduced relative to wild type CD28 (lanes 5 versus 10). This indicated that the second PXXP motif contributed to the efficacy of Grb2/CD28 binding. Consistent with this, the SH2 domain mutant showed similar binding levels for both CD28-WT and CD28-Ala 211 Ala 212 proteins (lanes 3 and 8). Hence, the differential binding observed between Grb2-WT and Grb2-mSH3 for wild type CD28 was abolished when CD28 was mutated at the two proline residues (Fig. 2B, lanes 3 and 5  versus 8 and 10).
This difference was not due to a difference in the levels of was eluted as described under "Materials and Methods" and reprecipitated with indicated fusion proteins. E, reprecipitation experiment was carried out as above except that probing with anti-Tyr(P) was performed instead of in vitro labeling. protein expression, or in the levels of tyrosine phosphorylation between the wild type and mutant CD28, as shown by immunoblotting with anti-CD28 and anti-Tyr(P) (Fig. 2B, top right  panel). Furthermore, the binding of the Grb2 SH3 domain (Grb2-mSH2) to CD28-WT was again detected in the absence of the SH2 domain-mediated binding (Fig. 2B, lane 2, and Fig. 1A,  lane 2). Under the same experimental conditions, CD28-Ala 211 Ala 212 was not precipitated by GST-Grb2-mSH2 (Fig. 2B,  lane 7). This difference in SH3 domain recognition was also observed using anti-CD28 in blotting (Fig. 2B, lower right  panel). These data indicate that there is an involvement of the PXXP motif/SH3 domain interaction in the CD28/Grb2 association.
To examine the in vivo association of CD28 and Grb2, various CD28 mutants were expressed in the Sf21 cells. The expression levels of the wild type, Phe 191 , Cys 194 , and Ala 211 -Ala 212 were similar whereas Gln 193 had a lower level of expression (Fig. 3A, left panel). Tyrosine phosphorylation levels of Cys 194 and Ala 211 Ala 212 were a little higher than the wild type (Fig. 3A, lanes 9 and 10 versus 6), suggesting that these two mutants might serve as better substrates for Lck in this system. The efficiency of tyrosine phosphorylation of Gln 193 appeared similar to that of the wild type as the relative intensities of the wild type and Gln 193 signals on the ␣-CD28 and ␣-Tyr(P) blots (Fig. 3A, lanes 1 and 6 versus lanes 3 and 8) were similar when analyzed by densitometry (data not shown). As expected, substitution of phenylalanine for tyrosine at 191 (CD28-Phe 191 ) abrogated phosphorylation (Fig. 3A, lane 7).
Loss of tyrosine in the YXNM motif of CD28 (CD28-Phe 191 ) resulted in a great reduction of Grb2 binding (Fig. 3B, lane 9  versus 8), although, as expected, residual Grb2 binding was still observed. Grb2 binding was also substantially decreased when the asparagine residue (Gln 193 ) was altered (Fig. 3B, lane  10). The apparent weaker signal of Gln 193 relative to Phe 191 in the Grb2 blot (Fig. 3B, lanes 9 and 10) was due to the lower expression level of Gln 193 compared with other CD28 constructs (Fig. 3A, left panel) as densitometric analyses showed similar intensities of Gln 193 relative to WT in the CD28 (26%) and Grb2 (33%) blots. Significantly, mutation in the PXXP motif (CD28-Ala 211 Ala 212 ) also diminished Grb2 binding (Fig.  3B, lane 8 versus 12, 32% reduction indicated by densitometry) even though the level of tyrosine phosphorylation of Ala 211 Ala 212 was similar to or greater than the wild type (Fig.  3A, right panel). These observations indicate that Grb2 SH3 domain recognition of the PXXPXR motif (residues 208 -213) within CD28 occurred in vivo, and further that this recognition occurs independent of tyrosine phosphorylation. In contrast, mutation of Met 194 of the YMNM motif, the residue specifically required for the p85 SH2 but not for the Grb2 SH2 domain, did not affect Grb2 binding (Fig. 3B, lane 11). Similar expression levels of Grb2 was found among the cells with a exception of Gln 193 cells that had a little lower level of expression (Fig. 3B,  lanes 1-6).
The capability of the wild type and the mutants of CD28 to bind p85 was also compared. As expected, mutations of the residues critical for the p85 SH2 domain binding (Phe 191 and Cys 194 ) almost completely eliminated the p85 association (Fig.  3C, lanes 8 and 10). The CD28-Ala 211 Ala 212 mutant bound p85 as efficiently as the wild type (Fig. 3C, lanes 7 and 11). Binding of Gln 193 was similar to wild type when the difference in the CD28 expression level was taken account (Fig. 3C, lanes 7 and  9; Fig. 3A, lanes 1 and 3). The fact that the Ala 211 Ala 212 and Gln 193 mutants could bind p85 as efficiently as the wild type indicates that these mutants were not nonspecifically defective in an ability to bind intracellular proteins.
Given our evidence for the binding of SH2 and SH3 domains of Grb2 to CD28, we studied the role of this interaction in CD28-mediated co-stimulation of IL-2 production. Stable DC27.10 cell lines transfected with mutant human CD28 cDNAs carrying substitutions in the YXNX motif were generated as described (39). Flow cytometric analyses showed that the surface expression levels of CD3 and the endogenous mCD28 proteins were similar among the CD28 transfectants and that hCD28 was expressed in the Gln 193 clones at a higher level (Fig. 4A). Further, the level of hCD28 was severalfold higher than that of mCD28 (Fig. 4A). Cells were stimulated with anti-CD3 and either CHO or CHO cells ectopically expressing CD86 (B7-2, the ligand for CD28), and the culture medium was assayed for IL-2. As previously reported (39), co-stimulation of the wild type CD28 transfectants with CD86 in combination with a suboptimal anti-CD3 concentration enhanced IL-2 production (Fig. 4B). Further, mutation of the Tyr 191 reduced IL-2 production by 50 -80%. Significantly, mutation of the N193Q in the YXNX motif also greatly reduced this co-stimulatory activity (Fig. 4B). Two separate transfectants showed the same phenotype with inhibition of 50 -70%. These observations implicate the residues needed for Grb2 binding in the generation of co-signals needed for IL-2 production.
In the course of examining the signaling capacity of the wild type-and Gln 193 -CD28 in T cells, we observed a distinct 95-kDa protein that was prominently phosphorylated on tyrosine residues. The phosphorylation of this protein was strongly enhanced upon stimulation of the wild type cells with B7-2, whereas stimulation of Gln 193 did not lead to any increase of phosphorylation of this band (Fig. 5A, lanes 1-4). Stripping and reprobing with an anti-Vav antibody identified the protein as Vav (Fig. 5A, lanes 5-8). Recently, tyrosine-phosphorylated Vav was reported to function as a guanine nucleotide exchange factor for Rac, an upstream activator of JNK/stress-activated protein kinase (23,24). Therefore, we examined the activation of JNK in the wild type and Gln 193 cells. For JNK activity assay, JNK was precipitated from the wild type and Gln 193 cells transiently transfected with JNK expression plasmid and subjected to in vitro kinase assay with GST-Jun as a substrate. In wild type-CD28 cells, some JNK activation was detected upon CD3 ligation with antibody and this activation was further enhanced by co-ligation of CD28 (Fig. 5B, lanes 1-4); a result consistent with a previous report (22). The activation of JNK was also present in CD3-ligated Gln 193 cells, however, importantly, further activation by the co-ligation of CD28 was not observed (Fig. 5B, lanes 5-8). These results implicate Grb2 binding to CD28 in the tyrosine phosphorylation of Vav and consequently in the activation of JNK in T cells. To rule out the possibility that the Gln 193 mutant cells lacked signaling capacity in general, we examined the CD28/p85 binding in these cells. Upon stimulation of wild type, Phe 191 , and Gln 193 cells with B7-2, comparable binding of p85 to the Gln 193 mutant and wild type CD28 was detected whereas Phe 191 did not bind p85 (Fig. 5C). These results indicate that the lack of co-stimulation of the Gln 193 cells in IL-2 production was not due to the inability of Gln 193 -CD28 to be phosphorylated on the site necessary for recruiting the PI 3-kinase signaling component and support that specific transduction pathway(s) involved in CD28 signaling was blocked in the Gln 193 mutant cells rather than the cells were aberrantly defective in signaling.
In summary, our observations provide strong evidence in  1-4). The membrane was stripped and reprobed with anti- Vav (lanes 5-8). B, cells were transfected with pEBG-JNK1 as described under "Materials and Methods." Two days post-transfection, cells were stimulated with combinations of 2C11 and CHO/B7-2 or control CHO and lysed. Cell lysates were precipitated with glutathione-Sepharose beads and the washed precipitates were subjected to kinase assay with purified GST-c-Jun as a substrate. C, cells were treated as in A and cell lysates were immunoprecipitated with anti-CD28. The immunoprecipitates together with aliquots of cell lysates were blotted for p85. support of Grb2 SH2 and SH3 domain binding to CD28. In several experiments, the Grb2 SH3 domain precipitated small, but reproducible amounts of CD28 in the absence of an intact SH2 domain (Figs. 1 and 2). Further, the CD28 molecule with a Phe 191 mutation (a mutation that blocks SH2 domain binding) consistently co-precipitated Grb2 (Fig. 3). Taken together, the results indicate that the Grb2 SH3 domain(s) can directly recognize CD28 independent of SH2 domain binding. As shown with the use of the CD28-Ala 211 Ala 212 mutant, the SH3 recognition required the intact PXXPXR site (residues 208 -213). A similar PXXPXR site is recognized within dynamin and SOS (27,42). Therefore, Grb2 SH3/CD28 binding could account for the low level of constitutive Grb2/CD28 binding observed in resting T cells (15).
The Grb2 SH2 domain mutation reduced CD28 binding about 40 -60% compared with the wild type Grb2 (Fig. 1A). This contribution of the SH3 domain was abolished when the PXXPXR site was disrupted in CD28 (Fig. 2B) and was greater than the amount of CD28 precipitated directly by the Grb2 SH3 domain alone (Figs. 1A and 2B). This result suggests that the SH2 and SH3 domains work cooperatively in recognition of CD28. Cooperative binding could occur upon activation and the induction of tyrosine phosphorylation, when residual SH3 domain binding is supplemented by additional SH2 domain binding. Alternatively, de novo supplemental SH3 domain binding could occur following the initiation of the SH2 domain binding. Together, the cooperative binding could enhance the avidity of Grb2 in binding to CD28 under circumstances where it may compete with PI 3-kinase recognition of the same YMNM site. Further structural studies should provide more detailed information on the nature of the SH2/SH3 binding to CD28.
A potential role for the Grb2 SH2 binding in CD28-mediated co-stimulation was shown by the inhibitory effect of the asparagine-193 mutation on IL-2 production induced by anti-CD3 plus CD86 stimulation (Fig. 4). Although we can not exclude the possibility that another protein may also require the presence of asparagine residue, or that the mutation caused an indirect conformation effect, no effect on p85 binding was observed (Fig. 3). Under the conditions, IL-2 production was reduced by some 50 -80% (Fig. 4). Unfortunately, transfection with the CD28-Ala 211 Ala 212 failed to show a stable phenotype (data not shown). Therefore, in addition to PI 3-kinase (9,39,43), mutations that alter Grb2 binding also influence IL-2 production. The C-terminal SH3 domain of Grb2 can bind to the Vav protein through the PXXP motif (36). Upon CD28 ligation, Vav in a complex with Grb2 may be recruited to the proximity of CD28 and its associated protein tyrosine kinase, and become tyrosine phosphorylated by the CD28-associated kinase. The phosphorylated Vav may then activate the downstream target molecule Rac that in turn engages the JNK activation cascade. As the activation of JNK has been shown to be essential for CD28 co-stimulation dependent IL-2 production (22), it can be postulated that the effect of Gln 193 mutation of CD28 on IL-2 production is due to the defect in tyrosine phosphorylation of Vav and subsequent activation of the JNK pathway (Fig. 5, A and B).