Rapid identification of phosphopeptide ligands for SH2 domains. Screening of peptide libraries by fluorescence-activated bead sorting.

A method for the identification of high-affinity ligands to SH2 domains by fluorescence-activated bead sorting (FABS) was established. Recombinant SH2 domains, expressed as glutathione S-transferase (GST) fusion proteins, were incubated with a phosphotyrosine (Y*)-containing peptide library. 6.4 × 105 individual peptides of nine amino acids in length (EPX6Y*X19X7X19X7X6) were each displayed on beads. Phosphopeptide interaction of a given SH2 domain was monitored by binding of fluorescein isothiocyanate-labeled antibodies directed against GST. High-fluorescence beads were isolated by flow cytometric sorting. Subsequent pool sequencing of the selected beads revealed a distinct pattern of phosphotyrosine-containing motifs for each individual SH2 domain: the SH2 domain of the adapter protein Grb2 predominantly selected beads with the sequence Y*ENDP, whereas the C-terminal SH2 domain of the tyrosine kinase Syk selected Y*EELD, each motif representing the most frequently found residues C-terminal to the phosphotyrosine. For deconvolution studies, soluble phosphopeptides comprising variations of the Grb2 motifs were resynthesized and analyzed by surface plasmon resonance.

A method for the identification of high-affinity ligands to SH2 domains by fluorescence-activated bead sorting (FABS) was established. Recombinant SH2 domains, expressed as glutathione S-transferase (GST) fusion proteins, were incubated with a phosphotyrosine (Y*)-containing peptide library. 6.4 ؋ 10 5 individual peptides of nine amino acids in length (EPX 6 Y*X 19 X 7 X 19 X 7 X 6 ) were each displayed on beads. Phosphopeptide interaction of a given SH2 domain was monitored by binding of fluorescein isothiocyanate-labeled antibodies directed against GST. High-fluorescence beads were isolated by flow cytometric sorting. Subsequent pool sequencing of the selected beads revealed a distinct pattern of phosphotyrosine-containing motifs for each individual SH2 domain: the SH2 domain of the adapter protein Grb2 predominantly selected beads with the sequence Y*ENDP, whereas the C-terminal SH2 domain of the tyrosine kinase Syk selected Y*EELD, each motif representing the most frequently found residues C-terminal to the phosphotyrosine. For deconvolution studies, soluble phosphopeptides comprising variations of the Grb2 motifs were resynthesized and analyzed by surface plasmon resonance.
Key to organization and control of intracellular signal transduction pathways are conserved protein modules that regulate signal transduction through their ability to mediate proteinprotein interaction. One conserved protein module consists of 100 amino acids, occurs with variations in a wide number of different signaling proteins, and is referred to as Src-homology-2 (SH2) 1 domain (1). The SH2 domain can fold into a compact and functional module independently of surrounding sequences (2,3). In an interacting protein it recognizes short sequence motifs bearing phosphotyrosine. This conserved feature phosphotyrosine is embedded in variable sequences that can increase the affinity by three orders of magnitude (K D ϭ 10 -100 nM) and direct the specificity to the various SH2 domains (4,5). The peptide binding site of an SH2 domain is bipartite (6). A conserved pocket lined by basic residues binds the phosphotyrosine: this pocket contains the only invariant SH2 residue, an arginine, which also forms two hydrogen bonds with phosphate oxygens of the phosphotyrosine (7). The second binding surface is more variable and allows specific recognition of the amino acids immediately C-terminal to the phosphotyrosine (for reviews, see Refs. 8 and 9). Examples for ligand binding specificity of SH2 domain-mediated protein interaction are given by Fantl et al. (10), who have shown by mutational studies that the SH2 domains of phospholipase C-␥, phosphatidylinositol 3-kinase, and Ras-GAP (GTPase-activating protein) recognize distinct phosphopeptide motifs in the platelet-derived growth factor receptor.
As a first systematic approach toward defining the linear SH2 domain binding motifs of a series of recombinant SH2 domains and to provide information about the relative importance of amino acid residues at position ϩ1 to ϩ3 C-terminal to phosphotyrosine, a synthetic phosphopeptide library with a total degeneracy of 5832 different peptides was applied (11,12). Obviously, this complexity was not sufficient to cover a large range of potential SH2 domain binding motifs. From the crystal structure of peptides bound to the N-terminal SH2 domain of the Syp tyrosine phosphatase, for example, it became clear that sequence specificity can extend across the five residues following the phosphotyrosine (13).
We have now established a method to identify specific binding motifs for SH2 domains from a more diversified phosphopeptide library. This library was of a diversity of 6.4 ϫ 10 5 individual motifs that were degenerated at six positions around the phosphotyrosine. The library was synthesized on beads, using the "split and pool synthesis" approach (14,15) by which each bead carries a unique peptide sequence. So far, combinatorial peptide libraries on large beads (bead diameter of 100 -200 m) were incubated with acceptor molecules coupled to alkaline phosphatase and subsequently screened under a low-power dissecting microscope (14). Intensely stained beads were manually picked and individually sequenced. More advanced developments used encoded beads of smaller size with oligonucleotides or polyhalobenzenes as tags (16,17). These tags could be read from single beads by PCR or by electron capture gas chromatography, providing the amino acid sequence information of the peptide linked to the bead. In our approach, we used beads of 10-m diameter, a size optimally suited for flow cytometry. We applied pool sequencing on the isolated beads to obtain consensus information.
Here we demonstrate the feasibility of this approach by identifying high-affinity binding peptides to the SH2 domains of Grb2 (18) and the tyrosine kinase Syk (19).

Construction of pGEX2T-SH2 Expression Plasmids and Purification of Recombinant Fusion Proteins-
The cDNA fragment encoding the single C-terminal SH2 domain of human Syk (19) was amplified from reverse-transcribed RNA of Epstein-Barr virus-transformed B cells using the polymerase chain reaction (PCR (20)). The cDNA fragment encoding the SH2 domain of human Grb2 (18) was amplified by PCR from reverse-transcribed RNA of the same source using the primer pair 5Ј-atgcggatccCCCAAGAACTACATAGAAATG-3Ј and 5Ј-atgcggatc-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
cCCCAAGAACTACATAGAAATG-3Ј and subsequently cloned into pGEX2T (21). Nucleotide sequences of the constructs and expression of proteins in Escherichia coli (BL21(DE3) pLysS) with the predicted molecular mass were confirmed in each case. Purification of glutathione S-transferase (GST) fusion proteins was performed as described (21).
Synthesis of the Phosphotyrosine Peptide Library PL-6E5-1.5 g of amino-functionalized Dynosphere monosized polymer particles (Dyno Particles AS, Lillestrøm, Norway) with a diameter of 9.7 m (CV 1.5%), approximately 2.1 billion polymer particles, were used in solid phase peptide synthesis by the Fmoc/t-butyl protection strategy. Fmoc amino acids including Fmoc-Tyr(PO(OH) 2 ) were obtained from Novabiochem (Lä ufelfingen, CH) and Propeptide (Vert-Le-Petit, France). 4 volume eq of Fmoc-protected amino acid (0.4 M solution in dimethylformamide (DMF) were coupled by in situ activation with 4 eq of benzotriazol-1yl-oxytripyrrolidinophosphonium hexafluorophosphate (0.4 M in DMF) and 8 eq of N-methylmorpholine (0.8 M in DMF) for 45 min at room temperature. Fmoc-⑀-aminohexamic acid was attached as spacer to the amino groups of the polymer particles. The complete synthesis of the library was carried out according to the split and pool synthesis method (14,15), where at each coupling stage each amino acid is coupled individually in separate reactors. After completion of the coupling, the contents of all reactors are combined, washed with DMF 6 times, N-deprotected with 20% piperidine/DMF 2 times for 5 min, washed with DMF 6 times, and divided into the reactors of the next coupling step. Side chain deprotection was carried out with 90% trifluoroacetic acid, 5% water, and 5% triisopropylsilane as scavengers at room temperature for 2 h. The beads were washed with trifluoroacetic acid, DMF, and methanol. By this process we generated the complete set of possible peptide sequences (in this case EPX 6 Y*X 19 X 7 X 19 X 7 X 6 ϭ 636,804 different peptides) with each bead carrying a peptide of unique sequence covalently linked to the polymer particle (SDZ226-953).
Incubation of Library PL-6E5 Beads with SH2-GST and FITC-labeled Antibodies to GST-50 or 120 mg of the library beads (1 mg ϭ 1.2 ϫ 10 6 beads) were incubated in the presence of 0.03 nM SH2-Grb2-GST or 2.0 nM SH2-Syk-C-GST, respectively in 200 ml of PBS buffer (PBS, 1% bovine serum albumin, 0.01% NaN 3 , 0.05% Tween 20) for 23 h at 4°C under slow vertical rotation. The beads were centrifuged at 1500 ϫ g for 5 min, washed once, and incubated for 90 min in 0.3 ml of PBS buffer with 43 g of anti-GST-IgG-FITC, filtered through cell strainer (Falcon 2350) with 30 ml of PBS buffer, and washed once. For analytical tests, the incubation conditions were the same using various concentrations of SH2-GST, 50 g of library beads in a total of 1 ml of PBS buffer and, after washing, staining with 5 g anti-rabbit-IgG-FITC in 50 l of PBS buffer. The anti-rabbit-IgG-FITC was obtained from the unlabeled antibody (Molecular Probes A-5800) by following the labeling method as described in Ref. 22. Dialysis steps were replaced by chromatography on PD-10 columns, Sephadex G-25 (Pharmacia, Sweden).
Isolation of Beads with High Fluorescence-Beads with high fluorescence were isolated by two sorting applications on an Elite (Coulter) cell sorter. Passing of the beads through a 70-m mesh size filter was mandatory in order to avoid fluid problems in the sorter. An average of about 6% of the bead mass was retained on the filter. Analysis of these beads after forceful partial dispersion yielded an identical affinity distribution histogram as the beads in the filtrate, indicating no preferential retainment. The presort was triggered at a fluorescence threshold which included the highest fluorescent beads amounting to 0.7% of the total population. The sorting mode was "coincidence abort off" and "one droplet sort." The actual throughput was 10,000 beads per s, and the recovery was 65%. The isolated beads were subjected to a second round of sorting in the mode ''one droplet sort'' and "anti-coincidence on." Triggering was set on forward scatter. Only beads in the monomeric state were collected by applying a corresponding gate in the forward and side scatter diagram. The FITC-fluorescence gate was set including the region of the upper 0.13% high fluorescent beads of the original starting batch. The beads were collected directly on a trifluoroacetic acid-treated cartridge filter (Applied Biosystems) for sequencing, the loaded filter was rinsed on blotting paper with 0.2 ml of water and dried. The final purification sort yielded 8,000 interacting beads out of 60 million beads originally (0.013%). Using the C-terminal SH2 domain of Syk in an analogous experiment, 15,000 beads were isolated from 140 million beads employed in the incubation.
Pool Sequencing of Isolated Beads-Sequencing was performed on an ABI 470A gas phase Sequencer using STD program; for phenylthiohydantoin-derivative analysis, an on-line ABI 120A PTH analyzer was used. The signals of the phenylthiohydantoin-derivatives of Thr, Trp, and Ser were corrected for losses by determining yields from sequence analysis of protein standards ␤-lactoglobulin A (for Thr and Trp) and human serum albumin (for Ser). The setup was not suitable for reliable detection of phosphotyrosine; it was therefore not determined.
Peptide Synthesis and Purification-Eighteen phosphotyrosine peptides were synthesized on Fmoc-Rinkamid-Mega crowns, 5.3 mol/ crown (Chiron Mimotopes, Clayton, Australia) fitted to a pin holder in a 1-ml format polypropylene microtiter plate (Beckman). To 17 eq of Fmoc amino acid solution (0.4 M in a 0.4 M benzotriazol-1-yl-oxytripyrrolidinophosphonium hexafluorophosphate/DMF) were added 34 eq of N-methylmorpholine (0.4 M in DMF). The final reaction volume was 500 l. After coupling of Fmoc-⑀-Ahx and (ϩ)-biotin, the cleavage of the peptide from the crowns and the side chain deprotection was carried out as mentioned above with 500 l of cleavage solution. The peptides were precipitated three times from the trifluoroacetic acid solution with diethyl ether, dissolved in water, and lyophilized.
Optical Biosensor Measurements-The Pharmacia BIAcore was employed. Streptavidin was coupled covalently to the sensor chip CM5 surface (corresponding to a signal of about 1000 resonance units) by derivatizing the carboxymethylated dextran hydrogel on the chip surface with N-ethyl-NЈ-(dimethylaminopropyl)carbodiimide and N-hydroxysuccinimide; unreacted groups were quenched during exposure to ethanolamine. Biotinylated phosphotyrosine peptides were then immobilized (reaching a signal of about 30 resonance units) to yield the "biospecific" chip. All runs were performed at 25°C in buffer A (20 mM HEPES, pH 7.4, 150 mM NaCl, 0.5 mM dithiothreitol, 0.005% (v/v) Tween 20). For competition experiments, the peptide Biot-cap-DPSY*VNVQ (SDZ117-432) was coated, and the binding of 50 nM Grb2-SH2-GST was monitored during 4 min of association (5 l/min flow rate) and 3 min of dissociation (20 l/min flow rate) in free buffer flow. Competing peptides were added to the SH2 domain sample at concentrations of 1 M. Results were expressed as "percent of maximum binding," setting as 100% the signal obtained after 7 s into the dissociation of a sample without competitor (in order to eliminate contributions of solvent effects to the signal). In affinity measurements, biotinylated phosphopeptides of interest were coated as above, and kinetic data were collected during 3 min of association and 4 min of dissociation (flow rates as above).

RESULTS
Peptide Library on Beads-A biased nonapeptide library was synthesized on Dynasphere beads of 10 m in size. Each bead carried about 8 fmol of peptide of a unique sequence. The nonapeptide contained an exclusive phosphotyrosine at the fourth position from the N terminus (defined as position "0" in our terminology). The amino acids presented at the other positions are listed in Table I. The diversity of the library theoretically amounted to 1 ϫ 1 ϫ 6 ϫ 1 ϫ 19 ϫ 7 ϫ 19 ϫ 7 ϫ 6 ϭ 636,804 different sequences. In order to keep diversity low, the amino acids represented in position Ϫ1, ϩ2, ϩ4, and ϩ5 were selected with the help of descriptors from a principle component analysis as described by Hellberg and colleagues (23). Sequencing of the library demonstrated that each amino acid was equally well represented as intended (see Fig. 2C), except for lysine. This was likely due to partial acylation during side chain deprotection of lysine by trifluoroacetic acid and, consequently, reduced detection by Edman sequencing.
Binding of SH2 Domains to Beads and Isolation of Highly Fluorescent Beads by FABS-The binding of SH2 domains to the phosphopeptides displayed on beads was monitored upon incubation with FITC-labeled antibodies recognizing the SH2 fusion protein partner GST. Binding of the fluorescent antibody to the beads in the absence of GST-SH2 was negligible at the concentrations used in the assays (Fig. 1, top panel). The binding of Grb2-SH2 to the beads revealed a continuous interaction spectrum from zero to high affinity (Fig. 1). About 5% of the beads showed detectable binding already at 0.01 nM Grb2-SH2-GST; at 0.1 nM, the binding reached saturation with 1% of the beads. At higher ligand concentrations, more beads were saturated forming a peak visible in the histogram at 10 3 arbitrary fluorescence units (Fig. 1, lowest panel). The incubation concentration of 0.03 nM Grb2-SH2-GST was considered optimal for subsequent FABS. A histogram region for high fluorescence was set which included 0.13% of the library. Beads within this region were collected on a FACS instrument and subjected to amino acid pool sequencing.
Motifs from Pool Sequencing-The sequencing results indicated that the binding of the Grb2-SH2 domain to the phosphopeptides was dependent on a small set of specific amino acids in every variable position of the library ( Fig. 2A). Specifically, these were the positions Ϫ1 and ϩ1 to ϩ5. In position ϩ2, asparagine was selected exclusively. This was in excellent agreement with reports that had pointed out the importance of this Asn for the specificity of the Grb2-SH2/ligand interaction (24,25). In the other positions, the selectivity was less stringent. From the results obtained with our biased library, we derived that peptides of sequences EPFY*(E,Q,V,I)N(D,E,V,Q)-(P,D)D would bind with high affinity to Grb2-SH2 domains. This was in good agreement with previously identified motifs of natural high affinity ligands as listed in Table II The C-terminal SH2 domain of Syk favored a motif very different from the one chosen by the Grb2-SH2 domain (compare Fig. 2, A and B). We suggest that peptides of the sequences EPEY*(E,Q,A,M)(E,Q,S,N)L(D,R,T)D bind with high affinity to the C-terminal SH2 domain of Syk (Fig. 2B). Again, an exclusive selectivity was found, but at a different position and for a different amino acid than in the case of Grb2-SH2. Leucine at position ϩ3 was preferentially selected, which is also found in the conserved immunoreceptor tyrosine-based activation motif (ITAM) with the consensus sequence YXX(L/I)XXXXXXXYXX-(L/I) and can associate with Syk (33). Interestingly, no isoleucine was selected although it is part of the consensus sequence defined by functional studies (34). In positions ϩ1 and ϩ2, high selectivity was found for glutamic acid agreeing perfectly with the YEEL motif in the ITAM present in the high affinity IgE receptor (Table II) (35).
Screening of Selected Peptides by BIAcore-The result of the pool sequencing is the summation of a large number of sequences of high-affinity peptides. In this process of averaging, a part of the information on individual sequences is lost. In particular, a certain combination of two amino acids might be adverse for binding although both were of high selectivity. A deconvolution step is theoretically required to regain the sequence representing the high affinity binding motif. To this end, 12 peptides were designed accounting for the preferences demonstrated by SH2-Grb2 and individually synthesized in soluble form. The sequences were selections of EPFY*(E,Q,V)-N(D,E,V)(P,D)D. As negative controls, we included six analogous sequences with proline at position ϩ3, an amino acid that was strongly disfavored by Grb2-SH2 at this position (see Fig.  2A). We used the BIAcore biosensor to determine in a first screen the degree of competition of the resynthesized peptides for the binding of Grb2-SH2 to DPSY*VNVQ, the target sequence for the SH2 domain of Grb2 in the natural ligand SHC. At a concentration of 1 M, eight of the twelve peptides competed as well as, or even better than, the SHC peptide (Table  III). The best inhibition (74%) was obtained with peptide 1 (EPFY*VNVPD). All negative control peptides were inactive and allowed binding to a degree of 96% or above. Thus, our deconvolution of amino acid preferences and prediction of SH2 target sequences proved feasible and allowed the identification of high-affinity binding peptides for this SH2 domain.
Affinity Measurements of Grb2-SH2-GST Peptide Interactions-The affinities of SH2 domains and their targets have been assessed by different techniques, including the BIAcore, and were reported to be in the nanomolar range (24,25). To obtain more quantitative information on the affinities of the deduced amino acid sequences, we selected two peptides for a kinetic study: peptide 1 as the best competitor and peptide 8 with an intermediate inhibition strength, comparable with the binding motif from SHC. Peptide 8 represented the sequence composed of the amino acids with highest selectivity at each position. To our surprise, both of these peptides bound Grb2-SH2 better than either of the natural peptide sequences from SHC (see above) or the Bcr/Abl protein (Table IV)  The amino acids presented at each position of the nonapeptide library listed in single-letter code. The phosphotyrosine Y* was placed in a fixed position, defined as position 0.
Amino acid pool sequencing of library beads. Displayed are the amino acids detected at each position from Ϫ3 to ϩ5; phosphotyrosine is at position 0 (positions are separated by shaded vertical bars). Selectivity on the Y axis represents the relative amount of an amino acid at a certain position multiplied by the number of different possible amino acids at this position. The sequencing results are from isolated beads binding with high affinity to Grb2-SH2 (A) or Syk-C-SH2 (B) and from the library beads directly without selection (C). 1 was only slightly lower than the one of peptide 8 with 25 nM. The on-rate k on of peptide 8 was about twice as high as the values of the three other peptides. The reason might be quite strongly electrostatic interactions of this highly negatively charged peptide with the positively charged SH2 domain.
In conclusion, the deconvolution step revealed a new sequence for a peptide with only a small gain in affinity compared to the theoretically optimal sequence of peptide 8 obtained from the pool sequencing. The new sequence shares the consensus Y*VNV with the natural ligands SHC and Bcr-Abl (Table IV). DISCUSSION Fluorescence-activated bead sorting applied to a biased, untagged phosphopeptide library and subsequent amino acid pool sequencing was shown to be a rapid procedure for identification of a small set of ligands that bind to the SH2 domains of Grb2 with high affinity. In our BIAcore measurements, two of the resynthesized peptides showed equal or even higher affinity than the reference peptides derived from the natural ligands SHC and Bcr-Abl. The most active peptide shared with these two ligands the sequence Y*VNV. In a competition assay, we observed that the single exchange of Val to Pro at position ϩ3 in six peptides dramatically reduced the inhibitory activity in each case (Table III). Grb2-SH2 selected few or no peptides that presented a proline at this position, as none was detectable in the pool sequencing ( Fig. 2A). This result stresses the precision and specificity with which SH2 domain and ligand interact. Proline, which was preferentially selected at position ϩ4 had a detrimental influence on binding if presented at the neighboring position ϩ3. In this respect, a high K D value of 11.6 M was also reported for a peptide EPQY*ENPPIYLK which is very similar to the peptide 14 (consensus in italics, Table III) in binding Grb2 (36).
Among other binding partners (see Table II), Grb2 also associates with the protein-tyrosine phosphatase Syp (PTP-1D, PTP2C, SHPTP2) (37). From several potential binding sites for GRB2 with the consensus YXNX motif in Syp, the tyrosines in the motifs Y 542 TNIK and Y 580 ENVG actually get phosphorylated in response to platelet-derived growth factor (38,39). Comparison of the two sequences with the ones probed in this study clearly suggests that the Tyr 580 sequence is most likely the relevant high affinity Grb2 binding site in Syp.
The amount of approximately 8 fmol of peptide that was covalently bound per bead was too low for single bead amino acid analysis by Edman degradation. Therefore, pool sequencing was chosen for analysis as an alternative to encoding of the beads (40). Alignment of the pool sequences for consensus comparisons was simplified by the presence of the phosphotyrosine which inserts into a specific deep cleft on the surface of the SH2 domain. The advantage of pool sequencing is the availability of preference or consensus motifs from a statistically significant number of sequences, in our case at least 8,000. This is in contrast to the rather low numbers of sequences usually collected in phage library experiments. However, our type of data needed to be further deconvoluted to identify the sequences with highest binding affinities. The sequence featuring the amino acid with highest selectivity at each position was not necessarily expected to be the motif with the highest affinity. The deconvolution step resulted in a final motif with only small improvement of affinity: 25 nM for peptide 8 and 20 nM for peptide 1. Therefore, the sequence composed of amino acids with highest selectivity at each position is likely to represent a ligand with sufficiently high binding affinity.
High sample throughput is the major advantage of flow cytometric processing of the library on beads. A sorting rate of more than 1000 beads per s is easily achieved with conventional instruments. This rate was even 1 order of magnitude higher in the low purity presort with triggering the detection at a high fluorescence threshold. The FABS method for screening a library does not necessarily require a sophisticated high The amino acids displayed in bold correspond to the amino acids detected with selectivities higher than 2 in the same position in the pool sequencing results (Fig. 2, A and B). In the motif of the human epidermal growth factor (hu-EGF) receptor the I and Q put in italics had the fourth highest preferences in our analysis in these positions, but their selectivities were below 2.   speed sorter, as later experiments with batches of 150 -400 million beads were processed with or without magnetic presorting (MACS; Miltenyi Biotech GmbH, Germany) on a smaller flow cytometer (FACSort; Becton Dickinson). 2 Short analytical runs of the library beads incubated with the binding partner at different concentrations ( Fig. 1) immediately yielded information on the range of affinity interactions present on the beads. These runs also indicated adequacy or any modification required in the structure of the library and finally provided information on the optimal incubation conditions for each SH2 domain in the screening runs.
A discrepancy between the K D of 20 nM for the peptide 1/Grb2-SH2 interaction and the incubation concentration of 0.03 nM SH2-Grb2-GST is striking. This, although 600 times lower, is the concentration just below saturation binding for library beads expressing highest affinity. The reason for this difference might be that GST can form dimers in solution (41). Thus, the (SH2-Grb2-GST) 2 molecule can bind in a mono-or bivalent manner to a surface coated with peptides. On the BIAcore chip, monovalent binding predominated since the peptide density was kept low with the intention of favoring such binding (under such conditions, kinetic constants were very close for another SH2 domain in the absence and presence of the GST fusion partner; data not shown). On the other hand, on the library beads, mono-and divalent binding was about equally involved, as judged from dissociation studies including Src-SH2-GST and beads carrying peptides with the sequence of the polyoma middle T antigen (data not shown). Our screening method was also applied to SH2 domains, e.g. ZAP-C-SH2, that bind with a reported K D of 3.5 M (42). The bivalent binding mode due to the fusion protein and subsequent additional cross-linking by the FITC-labeled anti-GST antibodies may have contributed substantially to stabilizing the binding of the SH2 domains on the beads, allowing the screening also for low micromolar interactions.
Songyang et al. (11,12) published similar selectivity data for various SH2 domains using a soluble peptide library with variations at positions ϩ1, ϩ2, and ϩ3 and a diversity of 5Ј832 (11,12). The library was screened by affinity chromatography on immobilized SH2 domains. Their results correlate well with ours with respect to the pivotal importance of Asn at position ϩ2 for Grb2 and Leu at position ϩ3 for Syk-C. Some differences in the results for Grb2 may be due to species differences in SH2 domains employed in both studies. Overall, the selectivities obtained by the FABS method are strikingly more pronounced than those found by the affinity-elution method.
We have recently improved our screening by utilizing a library with a higher diversity. 2 Nineteen amino acids were presented at each position from Ϫ3 to ϩ5, except for position 0 with the unique phosphotyrosine. The handling in screening of this library of a theoretical diversity of 1.7 ϫ 10 10 needed only minor adjustments compared to the previous one of 6.4 ϫ 10 5 diversity. Our results for the Grb2-SH2 domain selectivities could be confirmed with the new library. The exclusive selectivity for Asn at position ϩ2 was also found with now 19 instead of 7 amino acids presented at this position.
In the future, this method will facilitate the process of drug discovery when employed in screening of high diversity, nonpeptidic libraries against a large number of interesting target proteins involved in disease.