Structural Insight into the Interaction between Platelet Integrin αIIbβ3 and Cytoskeletal Protein Skelemin*

Skelemin is a large cytoskeletal protein critical for cell morphology. Previous studies have suggested that its two-tandem immunoglobulin C2-like repeats (SkIgC4 and SkIgC5) are involved in binding to integrin β3 cytoplasmic tail (CT), providing a mechanism for skelemin to regulate integrin-mediated signaling and cell spreading. Using NMR spectroscopy, we have studied the molecular details of the skelemin IgC45 interaction with the cytoplasmic face of integrin αIIbβ3. Here, we show that skelemin IgC45 domains form a complex not only with integrin β3 CT but also, surprisingly, with the integrin αIIb CT. Chemical shift mapping experiments demonstrate that both membrane-proximal regions of αIIb and β3 CTs are involved in binding to skelemin. NMR structural determinations, combined with homology modeling, revealed that SkIgC4 and SkIgC5 both exhibited a conserved Ig-fold and both repeats were required for effective binding to and attenuation of αIIbβ3 cytoplasmic complex. These data provide the first molecular insight into how skelemin may interact with integrins and regulate integrin-mediated signaling and cell spreading.

The connection between extracellular matrix and the actin cytoskeleton is essential for cell extracellular matrix adhesion. This connection is mediated by a network of protein-protein interactions involving integrins and many cytoskeletal proteins (1,2). Spatiotemporal regulation of such connection leads to many dynamic cell adhesive processes such as cell shape change (cell spreading), cell migration, and cell survival. Skelemin, a muscle M-line cytoskeletal protein, has been known to play a critical role in mediating the extracellular matrix-actin connection during the early stages of cell spreading by directly interacting with the integrin cytoplasmic face (3)(4)(5). Skelemin is a 195-kDa member of a family of cytoskeletal proteins that con-tains five fibronectin (FN) 2 type III-like domains and seven immunoglobulin C2-like (IgC) domains (4) (Fig. 1A). Other members of this family, all associated with myosin thick filaments in skeletal and cardiac muscle, include titin (6), twitchin (7), progectin (8), myomesin-II (9), and myosin light-chain kinase (10). Based upon yeast two-hybrid, in vitro deletion/ competitive binding and in vivo transient expression studies, the integrin/skelemin interface was proposed to involve the skelemin IgC domains 4 and 5 (4,11).
Ig-folds are widely used as scaffolds for protein-protein interactions and are frequently involved in creating networks within the cells as well as bringing adjacent cells together (12). Ig-like domains have been implicated to have versatile binding modes. For example, for titin to complex with telethonin (13), the ␤-strands from the telethonin Ig-fold align in an anti-parallel arrangement with the last ␤-strands of the titin Ig-fold, resulting in an Ig-heterotetramer. In case of filamin, Ig-fold bound to the C-terminal part of ␤ integrin cytoplasmic tail (14), the integrin tail forms a ␤-strand that is aligned anti-parallel to the third and parallel to the fourth ␤-strand of Ig-fold of filamin on the other side of the ␤-sandwich. Although skelemin IgC domains 4 -5 also bind to integrins, the molecular details for this binding have not been characterized.
In this work, we have undertaken a detailed structural characterization of the skelemin/integrin interaction using NMR as a primary technique. We show that skelemin IgC repeats 4 and 5 exhibit a conserved Ig-fold, and we have characterized their interactions with integrin ␣ IIb ␤ 3 cytoplasmic tails. Specific regions of contact between the IgC of skelemin and the integrin ␤ 3 CT have been mapped. Our NMR data confirmed that the primary skelemin binding site resides in the membrane proximal region of the ␤ 3 CT (4), while suggesting an additional lower affinity contact within the vicinity of the NPLY motif. Furthermore, surprisingly, we have detected a skelemin binding site in the membrane-proximal region of ␣ IIb CT. Overall, our data provide a molecular template for understanding how SkIgC45 may serve as a modulator of the ␣ IIb ␤ 3 complex formation and a regulator of integrin clustering.
NMR Spectroscopy-HSQC (heteronuclear single quantum correlation) titration experiments (summarized in Fig. 2) were performed on 0.1 mM 15 N-labeled ␤ 3 CT or ␣ IIb CT mixed at 1:1, 1:3, 1:5 molar ratios of unlabeled SkIgC4, SkIgC5, and SkIgC45 at 25°C in 0.25ϫ PBS. Sequence-specific assignments of integrin tails are described elsewhere (15). Transferred NOESY experiments for different peptides were performed at 25°C in 0.25ϫ PBS with mixing times of 100, 200, 300, and 400 ms to analyze NOE build-up to eliminate spin-diffusion artifacts. Different ratios of the peptides to the binding partner were investigated to find the optimal range for NOE transfer for each particular analysis. The resonance assignments of unlabeled peptides were made using conventional two-dimensional 1 H TOCSY and NOESY spectra (16).
Structure Calculation-All heteronuclear NMR experiments were performed on 1 mM 15 N/ 13 C samples of SkIgC4 in 1ϫ PBS adjusted to pH 7.0, and 5% D 2 O as described in Clore and Gronenborn (17) on Varian Inova 600 MHz equipped with inverse triple-resonance cryoprobe or Bruker Avance 500 MHz with TXI probe spectrometers. The resonance assignments of double-labeled SkIgC4 were made using standard triple resonance experiments at two temperatures, 25°C and 40°C, to resolve the differential line-broadening for certain regions of the protein and to overcome degeneracy of peak frequencies. Briefly, HNCACB and HNCA experiments were used for the backbone and H(CC)(CO)HN, (H)CC(CO)HN, and HCCH-TOCSY experiments for side-chain assignments. All the spectra were processed with nmrPipe (18) or TopSpin (Bruker Inc.) and visualized with PIPP (19) or AUREMOL (Bruker Inc.). 2536 NOE distance restraints were obtained from 13 C-and 15 N-edited three-dimensional NOESY. Dihedral angle restraints were obtained from TALOS (20) and 3 J HNHA -coupling constants. Structure calculations were performed by CYANA 2.1 (21). Twenty best conformers with lowest target energy functions were subjected to molecular dynamics simulation in explicit water (22) using CNS (23). The protocol for water refinement has been adopted from Jung et al. (24). Structures were validated by PROCHECK-NMR (25) and visualized by VMD-XPLOR (26).

RESULTS
Skelemin Binding Site on the Integrin ␤ 3 CT-It has been previously suggested that SkIgC45 binds to integrin ␤ 3 cytoplasmic tail (4). This notion is based on yeast two-hybrid and GST pull-down experiments. To confirm and further characterize this interaction at the molecular level, we performed chemical shift mapping experiments by NMR. Unlabeled SkIgC45 was titrated into the solution of 15 N-labeled ␤ 3 CT, and associated chemical shift perturbations were monitored ( Fig. 2A). Chemical shifts changes, plotted as a function of the residue number in ␤ 3 CT (Fig. 2B), clearly show that the membrane-proximal region is most perturbed, confirming that primary binding site resides in this region (4). It should be noted that the extent of the chemical shift perturbations is rather small, which may be because of rapid reversibility and/or low affinity of the interaction. However, the perturbations are reproducible and concentration-dependent. Increase in SkIgC45 to ␤ 3 CT ratio of 5:1 resulted in significant line-broadening and disappearance of some peaks that correspond to the most perturbed N-terminal membrane proximal region of ␤ 3 CT (the Lys 716 -Arg 724 region; Fig. 2B).
To further characterize the binding interface and to distinguish which of the two Ig domains is essential for this interaction, we titrated unlabeled SkIgC4 and SkIgC5 separately into 15 N-labeled ␤ 3 CT (Fig. 2C). SkIgC5 appears to be primarily responsible for binding to the membrane-proximal region ending at Asp 723 . SkIgC4 causes perturbations mostly around Asn 743 -Thr 755 . SkIgC45 caused perturbations in both regions albeit in a smaller extent. Interestingly, the Asn 743 -Thr 755 binding site overlaps with the filamin Ig domain binding site found in the crystal structure of integrin ␤ 7 in complex with Ig-FLN21 (14) suggesting that SkIgC4 may bind to integrin ␤ 3 tail in a similar manner as filamin. Several additional residues, Ala 735 , Lys 738 , and Asp 740 , are also affected by binding to SkIgC4.
To confirm the above binding data, we employed another method by performing transferred NOE (trNOE) experiments (27). This method is especially well suited to characterize weak interactions (28). Briefly, it is expected that unstructured or only a partially folded peptide assumes a stable conformation upon binding to the target protein, and this process is manifested by appearance of additional peaks in its NOESY spectra. We used short synthetic ␤ 3 peptides composed of either the 24 N-terminal (N␤ 3 ) or 25 C-terminal (C␤ 3 ) residues. IgC4 and IgC5 were fused to glutathione S-transferase (GST), respectively, where fusion of GST increased molecular weight of the binding targets to allow favorable NOE transfer (28). N␤ 3 mixed with GST-IgC5 exhibited several additional peaks (Fig.  3A, blue, highlighted by arrows) compared with the free unstructured peptide, whereas N␤ 3 mixed with GST-IgC4 did not (3A, red). These additional peaks were also observed when N␤ 3 was mixed with SkIgC45 (data not shown). In comparison, C␤ 3 displayed almost no additional peaks when mixed with GST-IgC5 (Fig. 3B, blue) but exhibited several additional peaks when mixed with GST-IgC4 (3B, red, highlighted by arrows). Interestingly, these peaks are more pronounced and form a slightly different pattern when C␤ 3 was mixed with SkIgC45 ( Fig. 6 in supplemental data), suggesting that the loop connecting domains 4 and 5 may help to define the proper mutual orientation between integrin ␤ 3 CT and/or tandem skelemin IgC repeats. To conclude, our trNOE data independently validate the above HSQC-based skelemin binding model to integrin ␤ 3 CT, i.e. SkIgC5 binds to the N terminus of ␤ 3 CT, whereas SkIgC4 is involved in interaction with C terminus of ␤ 3 CT. The number of additional peaks was limited in both cases, which precluded us from total structure determination of the bound peptides. However, the NOE pattern did suggest that NPLY motif of ␤ 3 forms inverse turn when bound to SkIgC4.
Skelemin Binding Site on the ␣ IIb CT-Having mapped the specific interaction of skelemin with integrin ␤ 3 CT, we performed a control experiment examining whether IgC45 binds to ␣ IIb . To our surprise, the 1 H-15 N HSQC spectrum of ␣ IIb CT also underwent dose-dependent chemical shift changes upon addition of unlabeled SkIgC45 (Fig. 2E). However, no changes were associated with titration of SkIgC4 (data not shown), suggesting that SkIgC5 is a major binding partner for the integrin ␣ IIb subunit. Several peaks, such as the ones corresponding to residues Gly 991 , Phe 992 , Lys 994 , and Arg 997 , are significantly broadened and shifted slightly, whereas others, like Val 990 , Phe 993 , Arg 995 , and Asn 996 , are just shifted. All perturbations are clustered within the membrane-proximal region of ␣ IIb . This is the region involved in the maintenance of the receptor in a latent state by interacting with the membrane-proximal region of ␤ 3 . Interestingly, the most significantly perturbed res-idues are Phe 993 , Lys 994 , and Asn 996 , which are located on the opposite side of ␣ IIb interface with ␤ 3 CT (15). This novel finding may have significant impact on our understanding of how skelemin might associate with the integrin cytoplasmic face.
Integrin ␤ 3 Binding Site on Skelemin-We next attempted to investigate the interaction from the skelemin side using the same 1 H-15 N HSQC method. We do see changes in the SkIgC45 spectrum upon addition of full-length ␤ 3 CT, mostly manifested by differential line broadening, but also by some small chemical shifts perturbations (Fig. 2D). However, the overall quality of this spectrum prevents us from making full (or even backbone) resonance assignments of this construct. The molecular weight of SkIgC45 is about 29, which is within modern solution NMR limits, but the protein appears to undergo a significant degree of conformational exchange, and neither deuteration nor TROSY-based experiments at high magnetic field improved the spectral quality. Thus we decided to investigate SkIgC4 and SkIgC5 separately. We found that these two highly similar (about 20% sequence identity, Fig. 1B) domains of skelemin, composing essentially the same fold, behaved very differently in solution. SkIgC4 is a monomer and is characterized by a well dispersed, homogeneous spectrum, which allows not only full resonance assignments but also high-resolution structure determination (see below). However, SkIgC4 1 H-15 N HSQC spectrum shows no chemical shifts perturbations upon addition of full-length ␤ 3 CT at ratios of up to 1:3 (data not shown). Further increase of this ratio is limited by the insolubility of ␤ 3 CT, which prefers low salt and pH below 6.1, whereas SkIgC4 requires some salt and a pH ϳ 7.0 to prevent aggregation. Thus, we again used the short synthetic ␤ 3 peptides, N␤ 3 or C␤ 3 , to examine interactions. We see no changes upon addition of N␤ 3 in SkIgC4 1 H-15 N HSQC spectrum even with high excess of the peptide (ratios of up to 20:1). Very small changes upon addition of C␤ 3 in the SkIgC4 1 H-15 N HSQC spectrum (ratios of up to 40:1) were masked by random changes in frequencies of highly sensitive surface exposed residues, which did not allow us to map unambiguously integrin ␤ 3 binding surface on skelemin IgC4. Addition of ␣ IIb CT has no effect on SkIgC4 HSQC spectrum, confirming the above titration results for 15 N-labeled ␣ IIb subunit. SkIgC5 tends to oligomerize in solution, therefore we do not observe its 1 H-15 N HSQC spectrum because of its slow, on the NMR time scale, tumbling rate. Interestingly, the SkIgC45 construct seemed to behave better than SkIgC5 alone, which allowed us to obtain HSQC despite some conformational exchange problems (Fig. 2D). This finding indicates that SkIgC4 plays an important structural/stabilizing role in the solution behavior of SkIgC45.
NMR Structure of SkIgC4 and Homology Modeling of SkIgC5-To verify fold and characterize unique features of skelemin IgC domains 4 and 5, we performed detailed structural analysis. First, we determined the high-resolution structure of SkIgC4 by modern triple resonance NMR methods as described in the "Structure Calculation" section of the "Experimental Procedures" (assignments are presented in Table 2 of supplemental data). As expected, SkIgC4 adopts a very well known Ig-fold and contains seven ␤-strands in two ␤-sheets, which form a ␤-sandwich. Apart from these ␤-strands, however, SkIgC4 contains unique features, namely two alpha heli-ces, referred to as N-terminal H1 and C-terminal H2. Fig. 4A maps SkIgC4 secondary structural elements, and 4B shows a schematic representation of their arrangement. In accordance with the IgC-fold, ␤-strands A, B, E, and F form one face of the ␤-sandwich and strands C, F, and G form the other face (Fig.  4C). Interestingly, there is one short ␤-strand (DЈ), which runs almost perpendicular to strand C. This feature is not observed often in IgC-folds. The N-terminal H1 is made up of six amino acid residues and is followed by a long flexible region (13 residues), which connects it to ␤-strand A. The C-terminal H2 is composed of 10 residues and is attached to ␤-strand G by a short loop. Additionally, there is a helix-like turn, which connects ␤-strands E and F. Fig. 4D presents the backbone superposition of 20 calculated structures with the lowest target  . Transferred NOEs evidence for integrin ␤ 3 /skelemin and ␣ IIb /skelemin interactions. All shown trNOE experiments were performed at 25°C (400 ms mixing time) in 0.25ϫ PBS buffer (pH 6.2; 5 mM Ca 2ϩ was added to the samples used for experiments presented in panels C and D to stabilize ␣ IIb CT as described earlier (15,32)). Superposition of NOESY spectra are presented as follows: A, N␤ 3 alone (black) and in complex with skelemin GST-IgC5 (blue, some additional peaks are highlighted by arrows) or with GST-IgC4 (red) at the ratios of 20:1; B, C␤ 3 alone (black) and in complex with skelemin GST-IgC5 (blue) or with GST-IgC4 (red, some additional peaks are highlighted by arrows) at the ratios of 20:1; C, ␣ IIb alone (black), in the presence of MBP-␤ 3 (red) at the ratio of 100:1 showing additional peaks, a manifestation of trNOE effect reported previously (15) and in the presence of both MBP-␤ 3 and SkIgC45 (blue) at the ratio of 100:1:3, showing reduced number of attenuated additional peaks; D, ␣ IIb alone (black) and in the presence of SkIgC45 (red) at the ratio 100:1, showing additional peaks, but the pattern is somewhat different from the one seen in C. energy functions. This ensemble has been deposited to Protein Data Bank (access code 2jtd). Statistics of this ensemble are presented in Table 1 demonstrate a very high quality structure (see also Ramachandran plot, Fig. 7 in supplemental data).
The poor behavior of SkIgC5 in solution precludes its detailed structure determination by NMR. We, therefore, used the three-dimensional JIGSAW program (29) to build a threedimensional model for SkIgC5 (Fig. 4E) based on the homology to known structures, NCAM Ig-3 (12) and titin Ig-Z2 (13). As expected, it also adopts well defined Ig-C2-fold.

Influence of Skelemin on the Integrin ␣ IIb ␤ 3 Cytoplasmic
Complex and Its Implication on Integrin Signaling-The CTs of ␣ IIb and ␤ 3 form a clasp to maintain the integrin in an inactive conformation, and talin can activate the integrin by unclasping this complex (reviewed in Ref. 30). Our discovery that skelemin can bind to both ␣ IIb and ␤ 3 CTs has prompted us to investigate its influence on the CT complex. To find out if skelemin disrupts this clasp, we applied the same approach we used previously for studying integrin/talin interactions (15). In the case of talin/integrin interaction, talin competes with the ␣ IIb CT for binding to the ␤ 3 CT fused to MBP (MBP-␤ 3 CT) and abolished the trNOE induced by the ␣ IIb CT. However, when we added SkIgC45 to ␣ IIb CT/MBP-␤ 3 CT mixture (Fig. 3C), we saw attenuation but not total elimination of the trNOE peaks. Some trNOE peaks do disappear but many remain. Because the ␣ IIb CT also binds to SkIgC45, some of the trNOE peaks could arise from this interaction. Indeed, when we added SkIgC45 into free ␣ IIb , we saw a number of trNOE (Fig. 3D). The additional peaks observed are consistent with our HSQC binding data (Fig. 2C) implicating the membrane-proximal region of the ␣ IIb CT in the interaction. However, this latter spectral pattern is different from one of the ␣ IIb CT in the presence of both MBP-␤ 3 CT and SkIgC45 (Fig. 3C). This observation suggests that SkIgC45 attenuates the ␣ IIb ␤ 3 CT clasp; however, complexity of the system and limits of the experimental approach do not allow us to conclude that this complex is fully disrupted by skelemin.

DISCUSSION
In this study, we have performed detailed NMR experiments to understand the nature of the skelemin/integrin interaction. Our data not only confirm skelemin IgC45 interaction with integrin ␤ 3 CT but also show for the first time that it binds to integrin ␣ IIb CT. Detailed chemical shift mapping and transferred NOE analysis allowed us to pinpoint the specific regions within both ␣ IIb and ␤ 3 CTs involved in interaction with skelemin. In particular, our studies revealed that the membrane proximal regions of both the ␣ IIb and ␤ 3 CT are involved in binding to skelemin. The structural analysis demonstrates that skelemin IgC repeats 4 and 5 do adopt canonical Ig-folds but recognize the integrin in different ways. SkIgC4 appears to interact weakly with C terminus of ␤ 3 CT, whereas SkIgC5 binds to both N-terminal membrane-proximal regions of ␣ IIb and ␤ 3 CTs. Whether skelemin interacts with ␣ IIb ␤ 3 before or after integrin activation remains unknown although skelemin has been shown to be critical for cell spreading (an outside-in signaling event) suggesting that skelemin is at least involved in binding to integrin after integrin activation. Our NMR data suggest that skelemin exerts a different effect on the structure of the ␣ IIb ␤ 3 CT complex as compared with talin. Although the final conclusion awaits the structure determination of skelemin/integrin complex, the transferred NOE and the HSQC binding data suggest two major possibilities (Fig. 5). Skelemin may disrupt the ␣ IIb ␤ 3 cytoplasmic complex but may sit in between ␣ IIb and ␤ 3 CTs by binding to both subunits (Fig. 5, I). Skelemin may bind to the individual ␣ IIb and ␤ 3 CTs FIGURE 5. Model of ␣ IIb ␤ 3 /skelemin tertiary complex. Skelemin interaction with integrin cytoplasmic complex might initiate and/or stabilize receptor activation in two possible ways. I, skelemin may disrupt the ␣ IIb ␤ 3 cytoplasmic complex but keep ␣ IIb and ␤ 3 CTs in close proximity by sitting in between and binding to both subunits. II, skelemin may bind to the individual ␣ IIb and ␤ 3 CTs from two different heterodimers, thus separating ␣␤ intraclasp while connecting two adjacent receptors. Tendency of SkIgC5 domain to oligomerize further stimulates and stabilizes integrins cluster formation by promoting both homo-and hetero-oligomerization. from two different heterodimers, thus separating ␣␤ intraclasp while connecting two adjacent receptors (Fig. 5, II). The first possibility is consistent with noticeable attenuation of trNOEs in ␣ IIb ␤ 3 CT complex upon addition of SkIgC45 (Fig. 3C). The latter option is consistent with the HSQC data on SkIgC45/␣ IIb interaction (Fig. 2D) where skelemin binding to ␣ IIb CT affects mostly residues of the membrane-proximal region, which are not directly involved in the ␣ IIb /␤ 3 CT interface. These residues are Phe 993 , Lys 994 , and Asn 996 , and they are located on the opposite side of ␣ IIb interface with ␤ 3 CT. Nevertheless, because skelemin is involved in the formation of the nascent focal adhesions regulating the cytoskeleton organization, both possibilities may enhance integrin clustering, which is essential for cell spreading, a major finding as to how skelemin regulates integrin signaling (11). If skelemin is recruited to the inactive integrin, it may contribute to the receptor activation in certain cell types because it does have some weak unclasping activity (possibility I). The unclasping event could be further enhanced by other integrin activators such as talin. In possibility II, skelemin may stabilize the activated state of integrin by promoting clustering via a hetero-oligomerization mechanism. Furthermore, both the ␣ IIb and ␤ 3 transmembrane plus cytoplasmic domains are shown to form homooligomers (31). Considering the strong tendency of SkIgC5 to oligomerize as well, IgC5 domains from two different skelemin molecules could form a dimer, bridging two adjacent integrin homo-subunits. Thus skelemin may play an important role in the stabilization of integrin clusters by promoting both homo-and hetero-oligomerization.
To conclude, we have (i) defined for the first time skelemin binding surface on platelet integrin ␣ IIb ␤ 3 , which involves both hetero-subunits; (ii) demonstrated that this interaction attenuates the inter-subunit clasp, found in inactivated state of the receptor; and (iii) structurally characterized skelemin tandem Ig-C2-like repeats 4 and 5 and showed that the two repeats contributed to integrin binding in a different but cooperative manner. Based upon these data, we propose that skelemin regulates integrin cluster formation at earlier stages of cell spreading. This model should help to design new experiments to better understand its mechanism of action.