Introduction
From a set of 2,900 platelet expressed proteins7, 47 candidate highly expressed transmembrane proteins were identified. To identify functional biomimetic peptides, we enriched for activity by choosing cytoplasmic decamers (<30 residues from membrane) conserved in orthologous sequences of other vertebrates and enriched for molecular diversity by adding peptides from the same region of related human proteins ("paralogous peptides"), in which at least one conserved residue differed markedly8 (Supplementary Figs. 1 and 2 online.). We synthesized decamer oligopeptides by standard Fmoc chemistry with N-terminal palmitylation (Pal) (to direct peptides toward the cell membrane1, 2, 3, 9) and desalted and purified by reverse phase HPLC3, 9, 10, 11. We experimentally screened 52 peptides (derived from 47 platelet proteins) and 26 paralogous peptides (at 10 
m and 50 m) in novel 96-well assays for platelet aggregation12 and ADP secretion13. These assays are sensitive to changes in diverse signaling pathways; in vitro assays of platelets from human donors have a functional correlate with bleeding and thrombotic disorders, and thus have close relevance to human biology. Moreover, the anucleate platelet favors peptide, rather than RNA, approaches. Agonists of platelet activation were defined as peptides that cause resting platelets to aggregate or secrete ADP, whereas antagonists were defined as peptides that inhibit thrombin-induced platelet aggregation or ADP release.
The screen revealed a large number of active peptides (22 with P < 0.01 in one of the four assays for the 50 M dose; null expectation of three findings), with different peptides typically showing assay-specific effects (Fig. 1, Table 1 and Supplementary Table 1 online). 13 of the 22 significant peptides showed the same significant effect (P < 0.05) at 10 M, which is consistent with specific, rather than general, nonspecific peptide action. One of the peptides, ITGA2B_Q (1022-Pal-GFFKRNRPPL-1031 of integrin 
IIb), contains the previously identified GFFKR3, 14 and RPP15 motifs, which confirms the ability of the screen to identify known motifs. Its paralogue, ITGA2B_P (1019-Pal-GFFKRVRPPQ-1028 of integrin V), shares both motifs and therefore it is not surprising that it also showed activity. However, for five of the evolutionarily related pairs of peptides (ABCA4, CLCN6, KIAA0792b, EDG2 and PTGIRa), only one of the pair had significant activity in any assay (50 M, P < 0.01, Fig. 1, Table 1 and Supplementary Table 1), which suggests that activity may be specific to the pathway of a particular protein subfamily member. For three other evolutionarily related pairs that showed activity (ADCY7b, CD226 and KCNC3), both peptides were active in at least one assay (50 M, P < 0.01), but the assays and magnitude of activity observed differed in each case between the two peptides, which is consistent with sequence-specific modes of action. The peptides of one pair actually showed substantially different activities relative to one another: CD226_Q potently inhibited platelet activation (showing a dose-dependent response in both standard light aggregometry and platelet spreading (Supplementary Fig. 3 online)), whereas the evolutionarily related peptide acted as an agonist (Figs. 1 and 2). Such diversity provides a natural library of compounds for investigating motif and domain function16.
Figure 1: Visualization of significant agonists and antagonists of platelet activation for 52 peptides from 47 platelet proteins.
Agonists are green and antagonists are red. Results are for eight assays: ADP release (ADP) and aggregation (Agg) for resting (R) and thrombin-activated (TA) platelets, with two doses of peptide (10 M and 50 M). Color intensity is proportional to statistical significance. Right panel: results for 26 paralogous peptides from related human proteins, shown alongside their related platelet protein peptide. Clustering of peptides (to the left) was based on neighbor-joining analysis of ten amino acid physicochemical properties over all residues of each peptide; this particular clustering was chosen from several alternative approaches as the one with the greatest tendency to cluster peptides of similar activity.
Figure 2: Effect of the CD226_Q peptide, derived from the CD226 protein, on platelet function.
(a) Results of high-throughput assays for CD226_Q peptide (Pal-RRERRDLFTE; black) and its nonplatelet paralogue CD226_P (Pal-RTFRGDYFAK; light gray). (b) Inhibition of standard platelet aggregation by CD226_Q peptide. Platelets were preincubated with buffer (black) or peptide (light gray; 50 M) for 3 min before the addition of 10 M thrombin receptor activation peptide (TRAP) or 5 mM DTT in the presence of 1 mg ml-
1 fibrinogen. (c) Inhibition of TRAP-induced platelet aggregation by CD226_Q peptide variants (50 M): TRAP, TRAP alone; Pal+, palmitylated peptide Pal-RRERRDLFTE; Scr, C-terminal scrambled peptide Pal-RRERRETFDL; Pal-
, nonpalmitylated peptide RRERRDLFTE. (d) Colocalization of CD226 protein with the platelet integrin IIb
3 in platelets activated by spreading on immobilized fibrinogen (20 g ml-
1) for 45 min; green, CD226; red, IIb3; yellow, colocalization. (e) Inhibition of platelet spreading on fibrinogen by CD226_Q peptide. Gel-filtered human platelets (3 
105 l-
1) were allowed to adhere and spread on fibrinogen-coated glass slides (20 g ml-
1) for the indicated times. Platelets were left untreated (Fg: fibrinogen alone) or preincubated for 3 min with 50 M peptide. Platelets are stained for polymerized actin with fluorescein isothiocyanate phalloidin. Error bars are s.e.m.
The immunoglobulin-like CD226 has been identified as a platelet17 and endothelial18 adhesion factor that binds platelets homophilically17. Antibodies to CD226 may inhibit or stimulate platelet activation19, 20. CD226 is associated with integrin clustering in T cells19. CD226 colocalizes with integrins in natural killer cells and T lymphocytes21, and we observed a strong colocalization in platelets (Fig. 2). CD226_Q also inhibited integrin-mediated platelet spreading on immobilized fibrinogen, whereas platelet spreading was not inhibited by the nonpalmitylated peptide (Fig. 2). Moreover, CD226_Q inhibited platelet responses induced by thrombin, but not direct integrin activation by DTT (P = 0.86, in contrast with thrombin, for which P = 0.000, Fig. 2). Thus, this peptide reveals a role for the CD226 protein in regulation of inside-out but not outside-in integrin activation. Identification of an inhibitor of inside-out signaling in an integrin-associated protein represents a new potential target mechanism for antiplatelet therapy. Three paralogous peptides (from EDG7, CLCN7 and TMEM63B) showed pronounced effects on platelet signaling when the queries did not (Table 1). All three are actually derived from proteins that are also expressed in platelets22.
To investigate whether bioactivity is caused by nonspecific features of the peptides, we property-clustered peptides; but we did not find very strong trends relating properties and activity (Fig. 1 and Supplementary Methods online). The most suggestive clustering, based on position-independent physicochemical properties, clustered four peptides (CD226_Q, EVER1_Q, LILRB3_Q and OCLN_Q) that all inhibited platelet aggregation (Fig. 1). However, the clustering was relatively weak, with no single common feature uniquely defining these four peptides (see charge distributions, Table 1). Two other aggregation-inhibiting peptides (CDH6_Q and KCNC3_P) did not group with this cluster, and the three other assays did not form obvious clusters (Fig. 1), which suggests that the weak clustering observed may have been a chance finding. Scrambling of peptide sequences can assess the influence of residue composition. We investigated scrambled peptides for two of the most active property-clustered aggregation antagonists, CD226_Q (280-Pal-RRERRDLFTE-289, derived from CD226) and OCLN_Q (266-Pal-KTRRKMDRYD-275, derived from OCLN), which also share a positively charged N terminus. Because short motifs easily arise by chance in randomized sequences from compositionally homogeneous peptides (such as the positively charged CD226_Q and OCLN_Q), scrambling the entire sequence is not an effective means of determining sequence specificity. Therefore, we scrambled the less charged C termini to test whether these more heterogeneous regions of the peptides determine specificity. The C-terminally scrambled Pal-RRERRETFDL peptide of CD226_Q lacked the strong aggregation-antagonist effect of the original peptide (Fig. 2c,e), whose action therefore seems to be dependent on the more unique C terminus. In contrast, the OCLN_Q peptide and its C-terminally scrambled Pal-KTRRKKDYRDM peptide showed a similarly strong aggregation antagonism (data not shown), which suggests that their activity may reflect their N-terminal charge distribution. Thus, the set of bioactive peptides includes motifs with varying degrees of specificity. This is to be expected, given the varying information content of motifs already known23. Finally, we tested nonpalmitylated forms of CD226_Q and OCLN_Q, and, as previously demonstrated for other peptides24, 25, neither was active in its nonpalmitylated form (Fig. 2).
Given the evidence that the activities observed are due to specific sequence features, we sought to identify possible shared motifs in the peptides that might account for similar activities in different peptides (correcting for evolutionary relationships between some peptide pairs)26 (Supplementary Methods). Shared motifs were not strongly enriched among the active peptides (Supplementary Table 2 online). Only 2 of the 20 top-ranked motifs identified among the 78 peptides, KXXYXSP and RPPQ, were shared by active peptides, which is consistent with a random distribution among active and inactive peptides. The RPPQ motif is found in peptides derived from KCNC3, KCNC1 and ITAV. Given that RPP is implicated in energy-dependent integrin signaling15, potassium channel and integrin peptides (Table 1) could potentially act through a common broad mechanism—but clearly not an identical mechanism, as ITGA2B_Q is an agonist and KCNC3_Q is an antagonist. The KXXYXSP motif is also found in peptides with opposite activities (such as the activator MMD_Q and the inhibitor ABCA4_Q; Table 1), which indicates that, even if they too act through a common partner, there is additional specificity in the rest of their sequences. Notably, the most similarity between any two unrelated peptides in the study was an RDLFT motif shared by CD226_Q and ADCY7a_Q, which includes four of five C-terminal residues of CD226_Q that (as indicated by scrambling) are strongly implicated in the inhibitory activity of this peptide (Fig. 2). ADCY7a_Q, however, showed no significant activity, which suggests that despite its obvious importance, the C terminus alone is not sufficient for activity.
Could these peptides have been identified by other computational approaches? We sought to identify (i) additional known motifs in the sequences, and (ii) novel motifs shared among proteins that share a protein-protein interaction partner. Only two known "ligand binding" motifs from the Eukaryotic Linear Motif database23 occurred (with expectation <1) in the 78 peptides (Supplementary Methods). Notably, one of these—the SH3_1 binding motif in the peptides from KCNC3 and KCNC1—recalls the SH3_1 binding of other potassium channels via similar proline-rich motifs27. Overall, though, it seems that most of the active peptides do not work through previously known and easily identified motifs. Computational discovery of motifs among related proteins that share an interaction partner is an alternative strategy for motif characterization28. A screen for motifs shared among proteins that interact with a common partner suggested four possible interaction motifs in active peptides (Supplementary Table 3 online) but did not identify motifs for most of the active peptides. Therefore, our approach complements the approach of discovering motifs shared by common interaction partners28 because it can define bioactive motifs that lack convergently evolved instances in proteins of known related function.
We conclude that systematic biomimetic oligopeptide screens are a highly efficient tool for discovering short signaling motifs in molecular signaling pathways; discovered peptides represent potential molecular templates for drug development. Hundreds of such short signaling motifs await discovery in the human proteome28. It is encouraging that our screen is enriched for adhesion protein–derived peptides (cadherins, occludin, integrins, LILRB3 and CD226); identifying and modeling the action of short interaction motifs represents a key strategy in designing small molecules that target adhesion in a wide range of therapeutic areas.
Methods
Peptide design.
The objective was to identify peptides that span residues that are strongly similar (conserved) to those in the corresponding proteins in other species (orthologs), but that differ from those in related human proteins (paralogues) (Supplementary Fig. 1). The peptides also had to lie within 30 amino acids of the membrane. Ancestral sequence predictions from HAQESAC (http://bioinformatics.ucd.ie/shields/software/haqesac/index.html) were used to calculate "burst after duplication" (BAD) statistics for each residue, which scores the number of amino acid property differences between the common ancestor of the paralogues and the common ancestor of the platelet query protein's orthologs29. BAD is built on the underlying assumption that sites critical to changes in gene function between paralogues are marked by a burst of radical amino acid substitutions directly after duplication, which are subsequently conserved within orthologous groups. Residues that are conserved within subfamilies but differ between them receive a high score and identify potential functional diversity in the corresponding regions of the parent proteins. (Supplementary Fig. 2 and Supplementary Methods). For query proteins with human paralogues, we designed decamer peptides that span membrane-proximal cytoplasmic residues in well-aligned regions with high BAD scores. For queries without paralogues, decamer peptides were selected on the basis of evolutionary conservation. We avoided unfavorable amino acid combinations and excessively hydrophobic peptides to reduce the risk of by-products during synthesis, postsynthetic degradation, and solubility problems. Further details on peptide design are available in Supplementary Methods.
Peptide synthesis.
Rationale for platelet assays.
High-throughput platelet aggregation assays.
We assayed each peptide at two doses (10 M and 50 M) on six donors (three male and three female)12 and expressed platelet activation as a proportion of the maximum activation observed for thrombin in the presence of the peptide vehicle on its own. Washed platelets were prepared as described previously3 and diluted in buffer A (130 mM NaCl, 10 mM trisodium citrate, 9 mM NaHCO3, 6 mM dextrose, 0.9 mM MgCl2, 0.81 mM KH2PO4, 1.8 mM CaCl2, 10 mM Tris-HCl, pH 7.4) to a concentration of 6 108 ml-
1. Peptide stock solutions (1 mM) were prepared in an appropriate vehicle depending on individual solubility (H2O, DMSO, 1% w/v; or methanol, 5% w/v final concentration; see Supplementary Table 1) and stored at -
80 °C. We performed dual agonist-antagonist assessment of platelet aggregation as follows: platelets (80 l) and peptides (10 M and 50 M) or the relevant vehicle were added to a 96-well plate to a final volume of 100 l and shaken at 37 °C. Absorbance readings (405 nm; 0.1 s; Wallac Victor2 1420 spectrophotometer) were taken before addition of peptide (T0) and at subsequent 3-min intervals. Thrombin (0.2 units ml-
1) was added after 6 min (T6), and an additional two absorbance readings were taken at T9 and T12. Changes in absorbance reflected agonist or antagonist properties of peptides to promote or inhibit platelet aggregation.
High-throughput ADP secretion assays.
We diluted platelets prepared as described above to 3 108 ml-
1 and aliquoted into black and white 96-well isoplates. We added peptides (at 10 M and 50 M) or the relevant vehicle and shook the plates at 37 °C for 3 min. Chronolume (10 l) was added after 3 min and luminescence recorded on a Wallac Victor2 1420 multilabel counter (Perkin Elmer) to measure the peptide-induced ADP release. Parallel plates were prepared with platelets activated by 0.2 units ml-
1 of thrombin. Changes in absorbance reflected agonist or antagonist properties of peptides to promote or inhibit ADP release.
CD226_Q inside-out versus outside-in signaling.
Confirmatory confocal microscopy.
Platelet spreading and phalloidin staining.
Statistical analysis of peptide activity.
To compensate for differences in assay readings for different donors and solvents, we converted raw aggregation and ADP secretion scores into a 'percentage (thrombin-induced) activation' (%Act) reading. Mean prethrombin controls were subtracted to give a resting baseline value of zero. Readings were then scaled by mean post-thrombin controls to give thrombin activation a value of 100%:
where R is the raw reading, Cn is the mean vehicle control reading before thrombin has been added, and Ct is the mean vehicle control after thrombin.
We compared %Act values for each peptide to all other peptides using a Mann-Whitney nonparametric test. For comparison, peptides were also compared with vehicle controls using a Wilcoxon nonparametric paired t-test of raw assay data. Each well for each peptide was individually paired with the mean value for the relevant vehicle controls on the same donor. Comparisons between peptide and paralogue were performed using the Mann-Whitney 'U' test. Statistical significance of follow-up P-selectin and aggregometry analyses was determined by one-tailed t-tests.
Additional methods.
Note: Supplementary information is available on the Nature Chemical Biology website.
Author contributions
D.C.S., D.K. and N.M. devised the experiment; R.J.E. and D.C.S. formulated the bioinformatics peptide design algorithm and performed the statistical analysis; R.J.E., N.M., D.K. and D.C.S. cowrote the paper; R.J.E. developed the bioinformatics tools and performed the peptide design and the bioinformatics searches for shared or known motifs in bioactive peptides; M.D. designed the selection rules for the synthesis of the peptides to ensure maximum integrity, stability and solubility of the corresponding palmitylated sequences and supervised the synthesis of these peptides; N.M., A.K. and E.D. developed and performed the high-throughput platelet activation assays; G.M. performed the colocalization microscopy; W.S., D.K. and N.M. designed and interpreted the platelet spreading assays; S.D.E.P. performed platelet expression analysis; M.F. provided proteomics data.
