Crystal Structures and Thermodynamic Analysis Reveal Distinct Mechanisms of CD28 Phosphopeptide Binding to the Src Homology 2 (SH2) Domains of Three Adaptor Proteins*

Full activation of T cells and differentiation into effector T cells are essential for many immune responses and require co-stimulatory signaling via the CD28 receptor. Extracellular ligand binding to CD28 recruits protein-tyrosine kinases to its cytoplasmic tail, which contains a YMNM motif. Following phosphorylation of the tyrosine, the proteins growth factor receptor-bound protein 2 (Grb2), Grb2-related adaptor downstream of Shc (Gads), and p85 subunit of phosphoinositide 3-kinase may bind to pYMNM (where pY is phosphotyrosine) via their Src homology 2 (SH2) domains, leading to downstream signaling to distinct immune pathways. These three adaptor proteins bind to the same site on CD28 with variable affinity, and all are important for CD28-mediated co-stimulatory function. However, the mechanism of how these proteins recognize and compete for CD28 is unclear. To visualize their interactions with CD28, we have determined the crystal structures of Gads SH2 and two p85 SH2 domains in complex with a CD28-derived phosphopeptide. The high resolution structures obtained revealed that, whereas the CD28 phosphopeptide bound to Gads SH2 is in a bent conformation similar to that when bound to Grb2 SH2, it adopts a more extended conformation when bound to the N- and C-terminal SH2 domains of p85. These differences observed in the peptide-protein interactions correlated well with the affinity and other thermodynamic parameters for each interaction determined by isothermal titration calorimetry. The detailed insight into these interactions reported here may inform the development of compounds that specifically inhibit the association of CD28 with these adaptor proteins to suppress excessive T cell responses, such as in allergies and autoimmune diseases.

For complete activation of T cells and differentiation into effector T cells, co-stimulatory signals via CD28 receptors are required in addition to signals via T cell receptors (1,2). Binding of ligands such as CD80 (B7-1) and CD86 (B7-2) to CD28 on its extracellular surface recruits protein-tyrosine kinases to the CD28 cytoplasmic tail, which consists of 41 amino acid residues and a phosphorylated Tyr 189 within a YMNM motif (3). Following phosphorylation, CD28 recruits growth factor receptorbound protein 2 (Grb2), 4 Grb2-related adaptor downstream of Shc (Gads), and p85 subunit of phosphoinositide 3-kinase (PI3K) (4,5). These adaptor molecules primarily bind to the sequence pYMNM (where pY is phosphotyrosine) in CD28 via their SH2 domains. As one of the downstream signals of CD28mediated co-stimulatory activity, Grb2 stimulates the CD28mediated IL-2 promoter activation through the Vav1-NFAT/ AP-1 signaling pathway (6). The binding of Gads to CD28 is essential for the CD28-mediated NF-B activation (7)(8)(9). The regulatory subunit of PI3K, p85, has both positive and negative effects on CD28-mediated co-stimulation (10).
Both Grb2 and Gads contain a central SH2 domain flanked by two SH3 domains with Gads having an extra Pro/Glu-rich sequence between the SH2 domain and the C-terminal SH3 domain (11). Grb2 is expressed ubiquitously in mammalian cells, whereas Gads is expressed predominantly in lymphoid tissue and hematopoietic cells, particularly T cells (12)(13)(14). The p85 subunit of PI3K consists of an SH3 domain, a RhoGAP domain, and two SH2 domains, which are designated as nSH2 and cSH2 for the N-and C-terminal SH2 domains, respectively. The consensus SH2-binding sequence of Grb2 and Gads is pYXNX, whereas that of p85, either nSH2 or cSH2, is pYXXM. Although these three adaptor molecules activate different signaling pathways downstream of CD28 as mentioned above, the SH2 domains of all the molecules can bind to CD28 at the same site, YMNM, after the phosphorylation of its first Tyr. Because our previous report showed that both nSH2 and cSH2 of p85 have a higher affinity to pYMNM than Grb2 and Gads (15), it is likely that the high affinity of p85 SH2 to CD28 results in a sequestration of CD28 from other SH2 domains, although we need to know the relative concentration of the proteins to draw such a conclusion. However, both Grb2-CD28 and Gads-CD28 interactions are also important for CD28-mediated co-stimulation. It remains unclear how these three adaptor molecules structurally bind to CD28 and elicit the proper co-stimulatory function of CD28.
We have recently reported the crystal structure of Grb2 SH2 complexed with a phosphotyrosine-containing peptide derived from CD28 (16) and that the peptide adopted the twisted U-type conformation. Here, we report the crystal structures and the thermodynamic analyses of Gads SH2, p85 nSH2, and cSH2 in complex with the phosphotyrosine-containing peptide derived from CD28. The structures reveal the bent conformation of CD28-derived peptide in Gads SH2, whereas the extended conformation is observed in both p85 nSH2 and cSH2. The conformational variations of the peptide together with the detailed interaction manner between the peptide and SH2s confirm the results of the thermodynamic analyses. High resolution structures of 1.20, 0.90, and 1.10 Å for Gads SH2, p85 nSH2, and cSH2, respectively, reported here should also be good templates for further computational analysis, such as molecular dynamics simulations and rational design of new ligands.

Results
Protein Expression and Crystallization-All the SH2 domains analyzed in this study were expressed as glutathione S-transferase (GST) fusion proteins but purified in GST-free form after the thrombin digestion. The sequence homology between Gads SH2 and Grb2 SH2 is 55.9%, and that between p85 nSH2 and cSH2 is 35.2% (Fig. 1). Our first construct for the Gads SH2 (Trp 58 -His 155 ) expression has an additional 5 residues, GSPEF, at the N terminus that is a cloning artifact and remained attached even after digestion by thrombin, similar to the construct for Grb2 SH2 whose crystal structure was previously determined at 1.35-Å resolution (16). Although the protein was successfully crystallized in complex with CD28-derived 8-residue phosphopeptide, SDpYMNMTP, designated as OctP, its diffractions were limited to a relatively low resolution (ϳ3 Å).  To improve the resolution, we expressed Gads SH2 with the C-terminal 6 residues (Arg 150 -His 155 ) deleted, but only the free form of Gads SH2 was crystallized (data not shown). Our final construct has shorter N-terminal additional residues, Gly-Ser, and the original C terminus, His 155 , which produced high quality crystals in complex with OctP. The additional sequence, Gly-Ser, was necessary because of thrombin digestion. The modification of N and C termini of Gads SH2 described above had little effect on its CD28 binding (data not shown). Consequently, a similar approach was taken for both p85 nSH2 (Asn 325 -Lys 430 ) and cSH2 (Glu 614 -Ala 720 ) where the proteins were expressed with two additional residues, Gly-Ser, at the N terminus and co-crystallized with Oct-P.
Crystal Structures of CD28-SH2 Complexes-The crystal structure of Gads SH2 complexed with OctP was determined at 1.20-Å resolution ( Fig. 2A and Table 1). The overall structure of Gads SH2 consists of a central antiparallel ␤-sheet flanked by two ␣-helices, which is similar to that of Grb2 and other SH2 domains ( Fig. 2B) reported previously (17). The crystal structures of p85 nSH2 and cSH2 complexed with OctP were also determined at 0.90-and 1.10-Å resolution, respectively ( Fig. 2, C and D, and Table 1). These complexes also show similar structures to that of Gads SH2 (Fig. 2, E and F). The root mean square deviations of main-chain structures between Grb2 SH2 and Gads SH2, Gads SH2 and p85 nSH2, and p85 nSH2 and cSH2 are 0.428, 1.320, and 0.875 Å, respectively. Despite the high similarity of the overall structures, the recognition mechanism of the phosphopeptide is quite different between Gads SH2 and p85 SH2 domains, probably reflecting the differences in consensus amino acid sequences, as described below.
The recognition mode of Gads SH2 for CD28 phosphopeptide is very similar to that of Grb2 SH2. The side chains of Arg 65 , Arg 83 , Ser 87 , and Ser 93 together with the main chain of Gln 86 are involved in a salt bridge and hydrogen bonding with the phosphate group (Fig. 3A). The main chains of Lys 106 and Leu 117 interact with Asn at pYϩ2 via hydrogen bonding (Fig.  3B). The bound OctP adopts a bent conformation (Fig. 3C), similar to a type I ␤-turn but different from the canonical conformation, as reported previously in Grb2 SH2-peptide complex (16). The bulky side chain of Trp 121 in Grb2 SH2 is considered to force the bound peptide into the bent conformation. Because this Trp residue is conserved as Trp 118 in Gads SH2 and its side chain orientation is stabilized by the cation-inter- is the observed structure factor amplitude and F c is the calculated structure factor amplitude. c R free is the same as R but was calculated using a random set containing 5% of the data that were excluded during refinement.
action with another conserved residue, Arg 139 , OctP is also forced to bend in the complex with Gads SH2 to avoid steric hindrance.
In contrast, OctP is in a more extended conformation in the crystal structures of p85 nSH2 and cSH2 complexes as we modeled previously (16). The residue Asn at pYϩ2 is exposed to the solvent and does not interact with any residues in either p85 nSH2 or cSH2 (Fig. 4, A and B). Instead, the side chain of Met at pYϩ3, which is oriented toward the solvent in the structure of Gads SH2 complexed with OctP, sticks into the hydrophobic pocket of both p85 nSH2 and p85 cSH2 (Fig. 4, A and B). The pockets are formed by Ile 381 , Phe 392 , Leu 413 , Tyr 416 , and Leu 420 in p85 nSH2 and corresponding residues of Cys 670 , Phe 681 , Leu 703 , His 706 , and Leu 710 in p85 cSH2, respectively. Interestingly, Met at pYϩ3 occupies almost the same position for Trp 118 of Gads SH2. There is neither a Trp nor a bulky residue at this position in both p85 nSH2 and cSH2 as the corresponding residue is Ser 393 and Ala 682 for p85 nSH2 and cSH2, respectively. The lack of bulky residue and slight conformational changes in this region allow Met at pYϩ3 to reach the hydrophobic pocket (Fig. 4, C and D). The absence of a bulky residue in p85 nSH2 and cSH2 also allows OctP to adopt extended conformations. The two SH2 domains in p85 show a similar molecular recognition toward the phosphate group of the phosphotyrosine; the salt bridge and hydrogen bonding to the phosphate group are formed with the side chains of Arg 340 , Arg 358 , Ser 361 , and Thr 369 in p85 nSH2 (Fig. 4E) and with the side chains of Arg 631 , Arg 649 , Ser 651 , and Ser 652 in p85 cSH2. However, p85 cSH2 has an additional unique cationinteraction between the side chain of Arg 631 and aromatic ring of the phos-photyrosine (Fig. 4F), which is not observed in p85 nSH2. This additional interaction seems to contribute to binding of the phosphotyrosine residue more tightly than in p85 nSH2.
Binding Thermodynamics of CD28 Interaction with SH2 Domains-The interactions of CD28-derived 12-amino acid phosphopeptide SDpYMNMTPRRP, designated as DdcP, with SH2 domains of Gads, Grb2, and p85 were analyzed using isothermal titration calorimetry (ITC). Because of the low solubility of OctP, more soluble DdcP was used for the ITC experiments. Fig. 5, A-D, show the typical ITC profile of CD28 binding to SH2 domains, and the thermodynamic parameters are summarized in Table 2. All interactions between CD28 and SH2 domains showed exothermic reactions. The binding affin-  Shown are the interactions (salt bridges, hydrogen bonds, and hydrophobic contacts) of the conserved Met (pYϩ3) of CD28-derived phosphopeptide with p85 nSH2 (A) and p85 cSH2 (B). The superposition of Gads SH2 and p85 nSH2 (C) and p85 nSH2 and p85 cSH2 (D) are shown. The residues located near the entrance of hydrophobic core are shown in stick representations (see text). The interactions of the phosphotyrosine residue of CD28-derived phosphopeptide with p85 nSH2 (E) and p85 cSH2 (F) are shown. The SH2 domain is shown in ribbon and surface representation, and CD28-derived phosphopeptide is shown in stick representation. The black dashed lines indicate the salt bridges and hydrogen bonds. The blue dashed line indicates cation-interaction between CD28-derived phosphopeptide and SH2. The backbone structures of Gads SH2, p85 nSH2, and p85 cSH2 are colored in green, cyan, and pink, respectively. ity of Gads SH2 was similar to that of Grb2 SH2, and their binding thermodynamics were comparable. The binding affinities of p85 nSH2 and cSH2 were higher than those of Grb2 SH2 and Gads SH2 mainly because of the favorable enthalpy change (⌬H). In comparison with the two SH2 domains of p85, the binding affinity of cSH2 was about 1 order of magnitude higher than that of nSH2 mainly because of the favorable ⌬H. The heat capacity changes (⌬C p ) for binding were successfully obtained from temperature-dependent ⌬H changes in the range of 15-30°C and are summarized in Table 2. The p85 cSH2 domain had the largest ⌬C p value of the group.
The residues Ser 393 in nSH2 and Ala 628 in cSH2, corresponding to Trp 118 in Gads, were mutated to Trp, and the binding thermodynamics of mutants to DdcP were analyzed using ITC ( Table 2). Upon mutation, the binding affinities were slightly lower than those of wild types but still higher than those of Grb2 SH2 and Gads SH2.

Discussion
The crystal structure analyses of three SH2 domains, Gads SH2, p85 nSH2, and p85 cSH2, in complex with a CD28-derived peptide showed details of their peptide recognition, and the structures correlate well with their binding thermodynamics. The most striking result of the study is that the same CD28peptide is recognized in a different manner, especially in the conformation of the bound peptide, despite the fact that they share high sequence similarity. In the case of p85 nSH2 and cSH2, the bound peptide showed an extended conformation, and Met at pYϩ3 position makes hydrophobic interactions with the protein (Fig. 6B). In contrast, Gads SH2 has a series of bulky residues, Phe 105 , Trp 118 , and Arg 139 , to force the phosphopeptide to adopt a bent conformation (Fig. 6A), similar to the conformation seen in Grb2 SH2 (16). As a result, Met at pYϩ3 loses contact with the protein, and Asn at pYϩ2, needed for the specific interaction, forms hydrogen bonds with the protein. Thr and Pro at pYϩ4 and pYϩ5 also make no contact with Grb2 SH2. The bent conformation and fewer contacts with less hydrophobic interaction between OctP peptide and Grb2/Gads SH2 would be the reason for the lower binding affinity than those in p85 nSH2 and cSH2 as determined by our ITC experiments ( Table 2). The larger ⌬C p values in p85 nSH2 and cSH2 also support the larger hydrophobic interactions (18,19). Similar to the previous mutational study in Grb2 (20), our mutational study also supports the notion that the Trp residue, corresponding to Trp 118 in Gads, is not the determining factor for the conformation of bound phosphopeptide ( Table 2).
The binding affinity of p85 cSH2 was about 10-fold higher than that of p85 nSH2. This can be explained by the different recognition for the phosphotyrosine of CD28. In p85 cSH2, the side chain of Arg 631 is located at the closed site of the phosphotyrosine side chain forming cationinteraction (Fig. 4F). In p85 nSH2, the corresponding residue, Arg 340 , is located far from the phosphotyrosine (Fig. 4E). This additional interaction in cSH2 would contribute to the favorable binding enthalpy change. It should also be noted that the binding enthalpy change of p85 cSH2 is a significantly larger value compared with the other SH2 domains analyzed in this study and has been reported previously (21). The favorable ⌬H is partially compensated by the unfavorable entropy change (⌬S), which is a general phenomenon of biomolecular interactions (22).
The binding thermodynamics of Grb2 SH2, p85 nSH2, and p85 cSH2 to phosphopeptides different from the CD28 peptide have been reported previously (21,23,24). The binding affinity of Grb2 SH2 to the Shc-derived phosphopeptide SpYVNVQ is about 10 times higher than that to DdcP02, SDpYMNMT-PRRPG (24). The NMR structure of Grb2 SH2 in complex with Shc-derived peptide showed that the Shc peptide forms a ␤-turn (25), which is slightly different from the twisted conformation of CD28-peptide bound to Grb2 SH2. This would be mainly due to the residue at pYϩ3, Val of Shc peptide and Met of CD28 peptide. The side chain of Met would prevent the bound peptide from forming a canonical ␤-turn, resulting in the decreased binding affinity. The binding affinity of p85 nSH2 to platelet-derived growth factor receptor (PDGFR)-derived phosphopeptide SVDpYVDMSK is similar to that of the CD28 peptide (23), but the thermodynamic contribution is different from each other, which is also seen in case of p85 cSH2. In comparison with the binding of cSH2 to the PDGFR-derived phosphopeptide GGpYMDMSKDESVDYVPML, both enthalpy and entropy changes for the binding to the CD28 peptide are more negative, more favorable enthalpy change and less favorable entropy change (23). Within the consensus SH2-binding sequence, the residue at pYϩ2 is different in these peptides: Asp for both PDGFR peptides and Asn for CD28 peptide. Because the residue is exposed to the solvent in both PDGFRand CD28-complexed structures (26), it can be excluded that their different thermodynamics is due to the difference in the interaction at the residue at pYϩ2. Instead, the difference in hydration might be because of the different numbers of water molecules located on the surface of SH2 and/or the peptide.
Based on binding affinities of respective SH2 domains to CD28, it is likely that interactions of p85 SH2 domains with CD28 could inhibit the binding of Grb2 and Gads to CD28.  SH2 (A) and p85 nSH2 and cSH2 (B). However, both interactions of Grb2 and Gads with CD28 are also important for the CD28-mediated co-stimulation. We recently showed that SH3 domains of Grb2 and Gads also contribute to CD28 binding, whereas that of p85 does not (15). In addition, we also should consider the possibility of formation of a CD28 dimer in which multivalent binding would increase the affinity relative to monovalent binding due to the effect of avidity (27). Nevertheless, the interaction of SH2 itself mainly contributes to CD28 binding, and therefore the high resolution structure information presented here will help to generate compounds to selectively inhibit the respective CD28 interactions (28). Such compounds would specifically modulate CD28 co-stimulatory signaling without affecting the function of other co-stimulatory molecules, such as inducible co-stimulator, CTLA-4, and PD-1 (29 -31). It is also possible that they would lead to improvement in the treatment for immune diseases involving excessive T cell responses, such as allergy and autoimmune disease.

Experimental Procedures
Expression and Purification of Gads SH2, PI3K p85 nSH2, and cSH2-DNA fragments encompassing Gads SH2 domain (Trp 58 -His 155 ), PI3K p85 nSH2 (Asn 325 -Lys 430 ), and p85 cSH2 (Glu 614 -Ala 720 ) domains were amplified by PCR using Gads and PI3K p85 full-length cDNA plasmids as the template and two synthetic primers (Table 3) to generate a BamHI site and an EcoRI site at the 5Ј-and the 3Ј-ends of the amplified fragment, respectively. After digestion with BamHI and EcoRI, the DNA fragments were cloned into plasmid pGEX-4T-2. The expression plasmid of Grb2 SH2 was constructed previously (16). Escherichia coli BL21(DE3) were transformed with the Grb2 SH2, Gads SH2, PI3K p85 nSH2, and cSH2 plasmids. Freshly precultivated cells were inoculated into growth medium containing 100 g/ml ampicillin and grown at 37°C. When the culture reached an A 600 of about 0.6, isopropyl 1-thio-␤-D-galactopyranoside was added to a final concentration of 0.1 mM. The cells were incubated at 22°C for another 20 h. All SH2 domains were expressed as GST fusion proteins. The harvested cells were suspended in PBS (pH 7.4) and lysed by sonication at 4°C. After the cell debris was removed by centrifugation, the supernatant was loaded to a glutathione column (glutathione-Sepharose 4B fast flow, GE Healthcare). To isolate the SH2 domains from GST, the thrombin protease was added to eluted fractions and incubated at 4°C for 12 h followed by application onto a benzamidine-Sepharose column (HiTrap benzamidine fast flow, GE Healthcare) to remove thrombin. The purified fractions were pooled and loaded to a gel filtration column (HiPrep Sephacryl S-100 column, GE Healthcare), and the buffer was changed to PBS (pH 7.4) for ITC experiments and 10 mM potassium phosphate buffer (pH 7.5) for all crystallizations. The protein concentrations were determined from UV absorbance at 280 nm and were calculated by using the molar absorption coefficients of 1.43 ϫ 10 4 , 1.58 ϫ 10 4 , 1.99 ϫ 10 4 , and 1.53 ϫ 10 4 M Ϫ1 cm Ϫ1 for Grb2 SH2, Gads SH2, PI3K p85 nSH2, and cSH2, respectively.
Synthesis of CD28-derived Phosphopeptide-The 8-and 12-residue phosphopeptides OctP and DdcP were synthesized by Fmoc solid-phase method as described previously (16). The C terminus is a carboxamide group prepared with Fmoc-NH-SAL-PEG resin (Watanabe Chemicals), and the N terminus is an amino group.
Crystallization, Data Collection, Model Building, and Refinement-Crystallization conditions were screened by the sparse matrix method using commercially available screening kits (Hampton Research) with the hanging drop vapor diffusion method at 20°C (32). Prior to data collection, all the crystals were soaked in cryoprotectant solutions containing 20% (v/v) glycerol along with their respective reservoir buffers and flash frozen using nitrogen gas stream at Ϫ183°C.
X-ray diffraction experiments were performed at the beamlines BL-17A at High Energy Accelerator Research Organization-Photon Factory (KEK-PF) and BL41XU at SPring-8. All data were processed and scaled using XDS (33) and truncated by the CCP4 program suite (34). For further refinement, 5% of independent, randomly selected reflections were set apart as a test set to calculate the R free values.
The structure of the SH2 domain of Gads complexed with a phosphopeptide (Protein Data Bank code 1R1Q) was used as the search model to determine the initial phases for Gads-OctP by the molecular replacement method using Phaser (35,36). Several cycles of manual model rebuilding and refinement were performed using Coot (37) and PHENIX (38), respectively. The structure of cSH2-OctP was determined as described above by using the structure of the cSH2 complexed with PDGFRderived peptide (Protein Data Bank code 1H9O) as a search model (39). The partially refined structure of cSH2-OctP was used as a search model for nSH2-OctP. The refined models were validated by MolProbity (40). The statistics for data collection and refinement are summarized in Table 1. The figures were prepared using PyMOL software. The coordinates and structural data for the complexes have been deposited in the Protein Data Bank (codes 5GJH, 5GJI, and 5AUL for Gads SH2-OctP, p85 nSH2-OctP, and p85 cSH2-OctP, respectively).
ITC Experiments-ITC experiments were performed on an MCS-ITC instrument (Malvern). The CD28-derived 12-amino acid peptide DdcP (100 -200 M) was titrated into the protein solution (10 -20 M) using a 250-l syringe. Each titration consisted of a preliminary 5-l injection followed by 19 subsequent 13-l additions. The heat for each injection was subtracted from the heat of dilution of the injectant, which was measured by injecting the CD28-derived peptide solution into the experimental buffer. Each corrected heat was divided by the molar concentration of CD28-derived peptide injected and was analyzed on the basis of a "one set of sites" model with MicroCal Origin 5.0 software supplied by the manufacturer. The ⌬C p was calculated from the slope of the regression line of linear fit of Gads-SH2-F CGGGATCCTGGTTTCACGAAGGCCTCTCT Gads-SH2-R GGAATTCTTAGTGACCCTGGTCTTCTCG p85-nSH2-F CGGGATCCAATATGTCCTTACAAAATGCT p85-nSH2-R GGAATTCTCATTTGGATACTGGATAAAGTA p85-cSH2-F CGGGATCCGAAGATTTGCCCCATCATG p85-cSH2-R GGAATTCTCATGCATATACTGGGTAGG