Four proline-rich sequences of the guanine-nucleotide exchange factor C3G bind with unique specificity to the first Src homology 3 domain of Crk.

The widely expressed cellular Crk protein has the domain structure SH2-SH3-SH3. We have previously demonstrated that the more N-terminal SH3 domain of Crk (CrkSH3(N)) specifically binds several cytoplasmic proteins. A cDNA encoding one of these proteins was isolated and found to have two different splice forms. The sequence is virtually identical to C3G, a guanine-nucleotide exchange factor. The center region of the 145-155-kDa protein contains four similar proline-rich sequences which are capable of binding individually to the SH3(N) domains of c-Crk and v-Crk. Comparison of these sequences in C3G to proline-rich sequences in other Crk-binding proteins suggests that positively charged amino acids following the prolines play an important role in the binding to the CrkSH3(N) domain. The endogenous C3G could be coprecipitated with Crk from cell lysates of cells expressing high levels of c-Crk or v-Crk, suggesting high binding affinity and a possible interaction in vivo. Unlike many other SH3-binding proteins which interact with multiple SH3 domains, C3G from cell lysates binds preferentially to the CrkSH3(N) domain. This unique binding specificity supports the idea that C3G plays an important role in Crk signaling pathways.

The presence of SH3 domains in proteins of lower eukaryotes such as Saccharomyces cervisiae is indicative of the functional importance of SH3 domains throughout evolution. Despite significant sequence diversity, structural comparisons of many SH3 domains show similar folds (11). The first SH3 binding protein was isolated through its ability to bind to the AblSH3 domain, and the sequence responsible for this interaction was localized to a 10-amino acid proline-rich fragment (12,13). It has been suggested that proline-rich SH3 binding motifs have a structure similar to that of a polyproline I1 helix (14). In such a helix each turn consists of three amino acids. Amino acids which are three positions apart are therefore oriented coplanar in space. Ultrastructural analysis of the interaction of a high affinity, proline-rich binding peptide with the phosphatidylinositol 3-kinase-pS5-a-SH3 domain demonstrated that two coplanar proline residues form contacts with two grooves on the surface of the SH3 domain (15). These grooves contain highly conserved aromatic amino acids and are spaced approximately a t a distance of one turn of a polyproline I1 helix.
The Crk proteins belong to a family of proteins that consist almost entirely of SH2 and SH3 domains, with little intervening sequence. This family presently includes v-Crk (161, two forms of c-Crk proteins, c-Crk-I and c-Crk-I1 (17,181,CRKL (191, and its homologs Sem5 (23) and Drk (24,251,, and Nck (27). Expression of v-Crk or elevated expression of c-Crk-I leads to cell transformation and increased cellular phosphotyrosine levels (4,16,18,28,29). The biological role of c-Crk proteins is currently unknown. Since these proteins lack apparent catalytic domains, their function probably lies in their ability to bind specific proteins via their SH2 and SH3 domains. Proteins that interact with the CrkSH2 domain via phosphorylated tyrosine residues have been first identified in cells transformed by v-Crk or v-Src (4,29,30). The CrkSH2 domain was shown to preferentially bind phosphoty-rosy1 peptides that contain a %(P)-X-X-Pro motif (7,301. Such a high affinity binding motif is generated upon phosphorylation of c-Crk-I1 by c-Ab1 in the spacer region between the two CrkSH3 domains. Binding of the CrkSH2 to this phosphotyrosine residue may regulate c-Crk functions (31). The CrkSH3(N) domain has previously been shown to interact with several proteins, including Sos, Abl, Arg, and unidentified proteins between 145 and 155 kDa as well as an abundant protein of 185 kDa (31-33). We are currently identifying these SH3-binding proteins to determine their functional properties and to better understand the binding specificity of the Crk SH3 domains. In this study we report the cloning of one of these proteins, C3G, that was also identified recently by Tanaka et al. (34). While other CrkSH3-binding proteins like Sos' and Ab1 bind to multiple SH3 domains (12,13,35), we S. M. Feller, B. S. Knudsen, and H. Hanafusa, manuscript in preparation.

32781
found that C3G is very specific for Crk. Analysis of the CrkSH3(N) binding sequences within C3G identified prolinerich sequences that bind to the CrkSH3(N) and no other SH3 domain tested.
Library Screening for CrkSH3 Binding Proteins-2 x lo5 plaques of a HeLa expression library (Clontech) were screened with %-labeled gst-CrkSH3SH3 for the first screen. Four positive plaques were identified. Two of them gave a strong signal and the corresponding phages were plaque purified in subsequent screens using unlabeled gst-CrkSH3SH3 as a probe. Bound gst-CrkSH3SH3 was detected with an affinity-purified gst antiserum and 'T-labeled protein A (Amersham Corp.). The first screen with the 35S-labeled probe gave a very low background since it did not result in isolation of gst binding phages. The two isolated phages contained inserts of 1.2 kb with an identical restriction pattern. The inserts were excised from the phage with EcoRI, subcloned into pBluescript I1 SK (+/-) (Stratagene) and pGEX-1N (AMRAD) or pGEX-3X (Pharmacia) and their identity confirmed by sequencing. One of these clones was used as a probe to obtain a 5.5-kb clone from a ApCEV15-M426 human embryonic fibroblast library (a gift of Toru Miki and Stuart Aaronson) (39) which encoded a 5"truncated C3G cDNA that extended to the poly(A) tail. The original C3G clone of 1.2 kb was also used to probe a blot of HeLa poly(A)+ selected RNA. The missing region of C3G was cloned by polymerase chain reaction. The clones were sequenced using a commercial sequencing kit (UBI).
C3Ggst-fusion Constructs-pGEX constructs of the C3G proline-rich sequences were constructed by subcloning into pGEX vectors. The correct reading frame was confirmed by sequencing through the junctions. CB-2, CB-3, and CB-4 were generated from polymerase chain reaction products and sequenced entirely. Proteins generated from these constructs were expressed and purified essentially as previously described (40). During all steps of purification protease inhibitors (0.3 PM aprotinin, 50 pg/ml antipain-dihydrochloride, 100 pg/ml phenylmethylsulfonyl fluoride, 1 PM leupeptin, and 1 p pepstatin A) were added.
SH3 Blot Assay-1 pg of gst or equimolar amounts of SH3-fusion proteins were subjected to SDS-PAGE, electroblotted, and incubated with 2 pg/ml (70,000 d p d m l ) 36S-labeled gst-C3G CB region as described elsewhere (32). Filters were exposed overnight.
Precipitation of C3G and Crk Proteins from Cell Lysates"CrW3Yl cells (rat fibroblasts expressing v-Crk (43)), c-CrWCEF (chick embryo fibroblasts overexpressing c-Crk (44)) and HeLa cells were harvested in RIPA (20 mM Tris-HC1 pH 7.5, 150 mM NaCl, 1% Triton X-100, 0.5% deoxycholate, 0.1% SDS, 1 mM Na,EDTA) with protease inhibitors as described elsewhere (32). Particulate material was removed by centrifugation for 30 min a t 10,000 x g. Cytosolic HeLa proteins (S100) for precipitation of C3G were obtained as described elsewhere (32). Protein complexes with gst-fusion proteins were precipitated with glutathione-Sepharose beads (Pharmacia Biotech Inc.). For precipitation with the C3G antibody, 1 mg of total cell lysate of CrW3Y1 or c-CrWCEF was precipitated with 10 p1 of anti-C3G. Protein complexes were precipitated with protein G-Sepharose beads (Pharmacia Biotech Inc.). In all precipitation experiments, beads were washed three times with RIPA. Proteins were then separated by SDS-PAGE, electroblotted and probed with the antibodies described in the figure legends.
ELISA-gst-CB-1 to -4 proteins were biotinylated with sulfosuccinimidyl 6-(biotinamid0)hexanoate (Pierce) according to the manufacturer's instructions. ELISA plates (Nunc) were coated with 100 nglwell gst-Grb2, gst-CrkSHJSH3, or gst in 0.1 M NaHCO,, pH 9.6, overnight a t room temperature (45). Wells were then blocked with 2% bovine serum albumin. After washing three times with blocking buffer (20 m~ Tris-HC1, pH 7.5, 100 mM NaCl, 0.05% Tween-20, 1 mM EDTA, 0.2% bovine serum albumin and 0.1% ovalbumin), biotinylated gst-CB fusion proteins were added at the concentrations indicated in Fig. 6. To measure bound gst-CB fusion proteins, wells were incubated with horserad-ish peroxidase-coupled streptavidin (Life Technologies, Inc.). A chromogenic substrate, 2,2'-azinobis(3-ethylbenzothiazoline-6-sulfonic acid) diammonium salt (ABTS, Pierce) was then added and the OD was measured at 405 nm after 5 and 15 min and the 5 min OD value subtracted from the 15 min OD value to yield AOD,,,. In each experiment, the AOD,,, at saturation of SH3 domain binding sites was considered 100%. AOD,,, values for all other points within the same experiment were expressed as a percentage of this 100% value. Binding to gst alone was subtracted a t each data point.

RESULTS
Cloning of a Protein that Binds to the CrkSH3 Domain-The N-terminal CrkSH3 domain (CrkSH3(N)) binds specifically to a limited number of proteins from HeLa cell lysates (32). A HeLa expression library was screened to clone proteins encoding CrkSH3 binding proteins. A phage containing a 1.2-kb insert with a n open reading frame was isolated. In a filter binding assay, the protein expressed from the phage showed strong binding to gst-CrkSH3SH3, weak binding to gst-Grb2 and no binding to gst-Nck or gst alone.3 The sequence of the 1.2-kb phage insert is almost identical to the middle region (amino acids 280-653) of C3G, a CrkSH3-binding protein that was recently published by Tanaka et al. (34). However, the two sequences differ at nucleotide 861 which changes threonine 287 in the previously reported sequence to a proline in our sequence  1B). In addition, two differentially spliced C3G mRNAs were identified by polymerase chain reaction. The shorter one has a 114-nucleotide deletion between nucleotides 151 and 265 (amino acids 50-88) of the coding sequence. Even though this difference was not detectable in Northern blots, multiple bands between 145 and 155 kDa appeared in Western blots that may represent the two splice variants. The C-terminal domain of C3G is homologous to the catalytic domain of the guaninenucleotide exchange factor CDC25 (46).
The amino acid sequence of the phage insert contains four stretches of proline-rich sequences (Fig. 1, A and C). They were subsequently shown to bind the CrkSH3 domain and were therefore named Crk Binding sequence 1 to 4 (CB-1 to -4). Alignment of the sequences showed remarkable similarity spanning nine amino acids (Fig. X ) . CB-2 and CB-4 have 819 amino acids identical with CB-1, whereas CB-3 has 619 identical amino acids with CB-1. (CB-1 to -4) are good candidates for SH3 domain binding sequences since they contain a Pro-X-X-Pro motif (14). One additional proline-rich sequence is localized N-terminal t o the first CB motif (amino acids 267-276). It differs significantly from the other four CB sequences and from sequences in other known Crk binding proteins (Table I) and was not analyzed in detail. C3G Preferentially Binds to the CrkSH3(N) Domain-Since the library screen involved a gst-fusion protein that contained both c-Crk SH3 domains, the domain in Crk sufficient for the binding to C3G was not yet clearly identified. The C3G CBregion containing all four CB-sequences was expressed as an 35S-labeled gst-fusion protein and tested for binding t o full length c-Crk, the CrkSH2 domain and the individual Crk SH3 domains which were immobilized on a filter ( Fig. 2A). The C3G CB-region bound only to the isolated CrkSH3(N) domain and full length c-Crk. To assess whether C3G CB-sequences can also bind to other SH3 domains, the 35S-labeled C3G CB-fragment was tested against a panel of SH3 domains (Fig. 2B). B     Only gst-fusion constructs that contained the CrkSHXN) domain bound strongly to the '"S-labeled C3G CB-region. Grb2, the individual Grb2SH3 domains and the phosphatidylinositol 3-kinase-p85-a-SH3 domain only bound weakly. Binding to the Grb2 and phosphatidylinositol 3-kinase-p85-n SH.7 domains was only ohserved in this filter binding assay but did not occur in precipitations from solution or in the ELISA (Figs. 3 and 6). This may result from the partial denaturation of the blotted SH3 domains.

K D S R D C S E R A P K S P D A L E S A Q S E E E V D E L S L I D H N E I M S R L T L K Q E C D D C P D V R G C S C N I L T T Y R T F I S P E E L I K K L Q Y R L L V H A T E T D R K D L V L Y C E A F Y E K F S P F A D T F K K R V S K N T F F V L V R V V D E L C L V E L T E E I L K L L M E L V F R L V C N G E L S L
Subsequently the binding the C3G CB region to cellular proteins other than Crk was analyzed. Blotted HeLa proteins, initially precipitated with unlabeled gst-CR-1, were probed with "'S-labeled gst-CB-1. Apart from a faint signal of about 38 kDa, no additional binding proteins were detected beyond nonspecific binding to gst'. This hand was also detected when '"Slabeled cell lysates were precipitated with gst-CR-I:' and could be Crk or the Crk-related protein CRKL. It is not possible to distinguish between Crk and CRKL (19) in this assay, since the AhVCR-1 Ahl'CH-2

L Q A P E L P T K T R T A V S P L L P R K E R G P R L P I L P S K T R T S G S P A L P R K P R D
P V P P P V P P R R R P L I P P P L P P R K Y F X X P P X L P X K X R X .P .s proteins have almost identical molecular weights. The unambiguous identification of this CB-1-binding protein awaits further analysis with antisera specific for Crk and CRKL. The specificity of C3G-SH3 domain interactions in solution was further analyzed by incuhating proteins of the c-ytosolic fraction of HeLa cells (SIOO) with a panel of SH3-gst-fusion proteins. Protein complexes were precipitated with glutathione-beads, washed with RIPA and prohed with an antiserum raised against C3G. C3G was detectable as a broad band between 145 and 155 kDa in precipitates with gst-fusion proteins of full length c-Crk, the CrkSHXN) domain and v-Crk (Fig.  3). All these constructs contain the CrkSHXN, domain. N o binding was observed to any other SH3 domain including Grb2, the individual Grh2 SH3 domains and the phosphatidylinositol 3-kinase-pf35-n-SH3 domain. When the antiserum was preincubated with immunogen in a control experiment the 145/155 kDa band was the only signal that was no longer visualized.'! The proteins precipitated with Grh2 ( Fig. 3 1 were not competcd by the antigen. These data demonstrate that the unique hinding specificity of the C3G-CR region toward the CrkSH3NI domain is retained in the full length C3G protein. Binding of thc Indir*idrral Prolinr-rich C.?G-CR Srqrrcnccs to u-Crh nnd c-Crh-The four proline-rich motifs in C3G are remarkably similar, hut the differences in the amino acid sequences could influence the binding to the CrkSHS domain. To investigate the binding of each individual proline-rich sequence, gst-fusion proteins containing only a single proline-rich sequence were expressed (Fig. 4). Since the v-CrkSH3 domain slightly differs from the c-CrkSH3 domain in its sequence, the binding of both c-Crk and v-Crk proteins was tested. Cell Iysates of v-Crk expressing rat fihrohlasts (v-CrW3Y1 I or c-Crk overexpressing chick emhryo fihrohlasts (c-CrWCEFl were incubated with gst-CR-1 to -4 and analyzed hy Western blot for the binding to Crk proteins. Both c-Crk and v-Crk hound well to all CR constructs tested (Fig. 5 ) . The binding was stahle in RIPA, suggesting strong binding of each proline-rich CR-sequence to the Crk proteins. These experiments show that there are four potential Crk hinding sites in the C3G protein.
Annlysis of CrkSH.?C?G Binding hy EI,ISA-To quantifv the affinity of the C3G-CrkSH3 interaction we utilized an ELISA. The hinding of each proline-rich sequence was analyzed. Gst-CB fusion proteins were hiotinylated and prohed for binding to immohilized gst-CrkSH3SH.7, gst-Grh2, or gst alone. Binding to the CrkSH3SH3 domain was saturahle in all cases (Fig. 6). However, Grb2 showed little binding and the hinding did not saturate even at the highest concentrations of gst-CI3 fusion proteins used in these experiments. Binding to gst alone was minimal and was suhtracted from the hinding to gst-CrkSH3SH3 and gst-Grh2. The high hinding affinity hetween gst-CrkSH3SH3 and the four proline-rich sequences is reflected in half-saturation points of 10.4 nsr (gst-CR-11, 27 nsr (gst-CR-2), 11.3 n u (Est-CB-3,. and 2.85 nsr rgst-CR-41. The ELISA binding data are consistent with the solution phase binding results shown in Fig. 3 and suggest that the affinity of Grb2 to C3G is not high enough for stahle hinding. Copreripitation of Crk a n d C.?G from ( M I 1,ysntc.T-Since the endogenous cellular Crk-I1 does not form detectahle complexes with cytoplasmic proteins in different unstimulated cells ,321. we used cells that overexpress Crk proteins to examine the formation of complexes between C3G and Crk. The endogenous C3G was precipitated with the C3G antiserum from c-Crk/CEF or v-CrW3Y1 cell lysates and the immune complexes prohed for Crk proteins (Fig. 7, A and R ) . v-Crk is a fusion protein containing viral Gag sequences and was detected with a monoclonal antibody directed against Gag. Crk proteins were only detected in precipitates with the C3G antiserum hut not with the preimmune serum. In summary. our data strongly suggest that the interaction hetween C3G and Crk is highly specific and therefore likely of biological relevance. DISCUSSIOS We have cloned a Crk hinding protein. C3G. that contains a domain homologous to guanine-nucleotide exchange factors for Ras GTPases and hinds with remarkahle sprcificity to the SHXN) domain of Crk. The stahle association hetween Crk and C3G in cell lysates suggests a biologically important interaction of these two proteins in a yet undefined c-Crk signaling pathway, as well a s a possihle role for C X ; in v-Crk transfnrmation. While this manuscript was in preparation, a virtually identical protein sequence was reported hy Tanaka rt nl. (841. However, our results differ sipificantly from the previous report in the elucidation of C3G sequences that mediate hinding to the CrkSH3Nl domain and the hinding specificity of v-CrW3Yl crlls wrre incuhated with 20 pgnfgst-fusion proteins as ~ndicatrd ( s r r also Fig. 41. Complrscs wrrr prrcipitatrtl with glutathione-heads. The prrcipitates wrrr washed with RII'A, srparnted hy SI)S-PAGE i 11"; I and blotted. The hlot was prohrd with a monoclonnl nntihody rnisrd against the Gag portion of thc v-Crk protrin. 13. e-Crk protein was precipitatrd from total cell lysates (500 pg) of c-CrWCEF's with gst-fusion protrins as indicntd Thr precipitatrd protrins wrre separatrd and immohilized as in A and prohrd with the Crk antiserum. 50 pg of total crll lysate was Western hlottrd to show t h r mip-ation o f t h r e-Crk protein.
C3G for SH3 domains. In our study, four homologous, prolinerich sequences, CR-1 to -4, were identified. Each sequence can bind individually and with similar affnity to the SH.7 domains of c-Crk and v-Crk. In contrast, Tanaka

30-
CrkSH3N) binding sequences identified in C3G rind t h r Ah1 kinase (3.7, with proline-rich sequences of the CrkSI-13 X I binding proteins Sos and Arg (32). also points to thc importance of basic amino acids in CrkSH3(N) binding sequences. The consensus sequence for high affinity binding t o the CrkSH3(N) domain obtained from this alignment (Table I) includes two basic amino acids following the prolines and differs therefore significantly from the consensus sequence previously proposed (34).
Factors that determine the specificity of the various SH3 domains for certain proline-rich sequences are currently not well understood. Only one example of a highly specific SH3 interaction has been previously reported. It occurs between a proline-rich sequence of p47pk0x and the p67PkoX SH3 domain (47,48). Other proline-rich sequences which have been tested often interact with multiple SH3 domains. The results vary to some extent, depending on the experimental system. Interactions of different SH3 domains with short proline-rich peptides or protein fragments have been analyzed in vitro, using purified proteins (13,35). However, binding of the corresponding full length proteins is not always detectable i n vivo. For example, the GTPase dynamin binds in vitro to the SH3 domain of the p85-a subunit of P13-kinase (351, but no interaction of these proteins was detected when p85-a and dynamin were overexpressed together in cells (49). This lack of correlation between in vitro and in vivo binding results has also been observed for the AblSH3 binding proteins 3BP-1 and 3BP-2 and some partial clones obtained by screening of expression libraries with Crk. The binding of the C3G proline-rich sequences to SH3 domains of Grb2 and phosphatidylinositol 3-kinase-p85-a which are immobilized on a membrane (Fig. 2B) is likely a similar artifact of the assay system and was not observed in solution phase binding assays. In agreement with the specific, high affinity interaction between Crk and C3G, indicated by the coprecipitation of these proteins from cell lysates, apparent Kd values (halfsaturation points) in the nanomolar range were obtained for the binding the CrkSH3SH3 protein in the ELISA, while Grb2 showed very poor, nonsaturable binding. The ELISA dissociation constants are about 1000-fold lower than the values obtained with small proline-rich peptides in solution binding studies (15). This difference results most likely from the surface adsorption of one of the binding partners. We propose that C3G is not important for intracellular signaling through Grb2.
The consequence of Crk binding to C3G could be the activation of Ras or a Ras-related protein. C3G contains a C-terminal domain with homology to the Ras guanine-nucleotide releasing factor (GRF) CDC25 of S. cervisiae (34,501. It belongs therefore into a growing family of mammalian GRFs with CDC25 homology. This family includes currently a brain specific Ras-GRF (CDC25"'') and the widely expressed Sos proteins (50, 51). In addition to their catalytic exchange factor domain, CDC25"'' and the Sos proteins contain functionally equivalent, non homologous sequences that control their association with growth factor receptors (52, 53). Sos has four proline-rich stretches of amino acids C-terminal of the catalytic domain which mediate receptor binding via the Grb2 adapter protein, and CDC2CiMrn has sequences at the N terminus which are crucial in ligandinduced activation of the exchange factor activity (52). While the biological relevance of having two C3G splice variants (Fig.  lA) is currently not understood, divergent N-terminal sequences seem to be common in the family of Ras GRFs. Four different RNAs, which are likely a consequence of differential splicing, were detected for CDC25Mm (52). They encode a fulllength protein and three proteins with truncated N termini. Only the longest protein can function in a ligand-dependent manner. It may therefore be interesting to functionally compare the two forms of C3G.
The specific complex formation between SH3 domains and cellular proteins can have at least three functional conse-quences. Recent reports indicate that the binding of SH3-containing proteins to enzymes can regulate their catalytic activity. Binding of Grb2 or the SrcSH3 domain t o dynamin (35), as well as the binding of the SrcSH3 domain t o phosphatidylinosi-to1 3-kinase-p85-a (53), increases the enzymatic activities of dynamin and PIS-kinase. Besides the regulation of enzymatic activities (35,53,54), SH3 domains can mediate the substrate recognition of enzymes (31,33) and the targeting of proteins to specific subcellular locations (55). Future studies will aim to determine whether the binding of C3G to Crk influences C3G activity or whether the binding to Crk merely relocalizes the C3G protein. The best studied SH3 domain-dependent interaction of functionally similar proteins is the binding of the adapter protein Grb2 to the mammalian guanine-nucleotide exchange factor Sos. Upon exposure of cells to some mitogenic stimuli, Grb2 is recruited from the cytoplasm to the membrane where it forms a link between tyrosine kinases and the Ras pathway. Several activated receptor tyrosine kinases generate high afinity binding sites for the Grb2SH2 domain (24,56-59). The Grb2 SH3 domains in turn bind to proline-rich sequences at the C terminus of Sos (21, 60), thereby bringing Sos in proximity to Ras. The expression of v-Crk or c-Crk-I in PC12 cells potentiates neurite outgrowth. Based on these results, it has been postulated that Crk can activate the Ras pathway (61, 62). This activation could result from the binding of Sos and C3G to the CrkSH3(N) domain. While the function of Sos as a Ras exchange factor has been established (571, the biological activity of C3G in higher eukaryotes is not yet characterized. However, C3G is able to complement a temperature sensitive CDC25 mutant yeast strain (341, suggesting that C3G acts as a n exchange factor for a member of the Ras or Ral family of GTPases. The yeast complementation assay cannot distinguish these activities. A more precise functional analysis of C3G signaling pathways should greatly advance our understanding of the biological functions of c-Crk as well as mechanisms of v-Crk transformation.