Molecular basis for activation and biased signaling at the thrombin-activated GPCR proteinase activated receptor-4 (PAR4)

Proteinase-activated receptor (PAR)-4 is a member of the proteolytically-activated PAR family of G-protein–coupled receptors (GPCR) that represents an important target in the development of anti-platelet therapeutics. PARs are activated by proteolytic cleavage of their receptor N terminus by enzymes such as thrombin, trypsin, and cathepsin-G. This reveals the receptor-activating motif, termed the tethered ligand that binds intramolecularly to the receptor and triggers signaling. However, PARs are also activated by exogenous application of synthetic peptides derived from the tethered-ligand sequence. To better understand the molecular basis for PAR4-dependent signaling, we examined PAR4-signaling responses to a peptide library derived from the canonical PAR4-agonist peptide, AYPGKF-NH2, and we monitored activation of the Gαq/11-coupled calcium-signaling pathway, β-arrestin recruitment, and mitogen-activated protein kinase (MAPK) pathway activation. We identified peptides that are poor activators of PAR4-dependent calcium signaling but were fully competent in recruiting β-arrestin-1 and -2. Peptides that were unable to stimulate PAR4-dependent calcium signaling could not trigger MAPK activation. Using in silico docking and site-directed mutagenesis, we identified Asp230 in the extracellular loop-2 as being critical for PAR4 activation by both agonist peptide and the tethered ligand. Probing the consequence of biased signaling on platelet activation, we found that a peptide that cannot activate calcium signaling fails to cause platelet aggregation, whereas a peptide that is able to stimulate calcium signaling and is more potent for β-arrestin recruitment triggered greater levels of platelet aggregation compared with the canonical PAR4 agonist peptide. These findings uncover molecular determinants critical for agonist binding and biased signaling through PAR4.

G-protein-coupled receptors (GPCRs) 4 are the largest family of cell-surface receptors and regulate a host of important physiological responses (1,2). GPCRs respond to a variety of extracellular signals and regulate cellular behavior through engaging intracellular effector molecules to activate various cell-signaling pathways (3,4). In drug discovery, GPCRs are a valuable and highly-tractable class of drug targets with over 30% of all currently-approved drugs acting on GPCRs (5). GPCR activation occurs typically through binding of a soluble ligand and conformational changes that enable engagement of G-protein-dependent or ␤-arrestin-mediated signal transduction. The proteinase-activated receptors (PARs) are a fourmember family of GPCRs with a distinct mechanism of activation that involves limited proteolysis and unmasking of a receptor-activating motif called the tethered ligand. PARs were first discovered in an effort to identify the cellular receptors responsible for actions of thrombin that were independent from its role in the coagulation cascade (6). Various proteinases have now been described as activators of the different members of the PAR family, but canonically PAR1 and PAR4 are described as thrombin-activated receptors, whereas PAR2 is activated by trypsin and other trypsin-like serine proteinases. PAR4 can also be activated by trypsin. Whether PAR3 signals independently is unclear, although thrombin can cleave the PAR3 N terminus to reveal a tethered ligand (7,8). Following the discovery of PARs, it was quickly realized that two receptors, PAR1 and PAR4, serve as the thrombin receptors on human platelets (9,10). PAR1 and PAR4 both trigger platelet activation and This work was supported by a Canadian Institutes of Health Research (CIHR) grant 376560 (to R. R.) and by grants from the Natural Sciences and Engineering Research Council of Canada (to L. G. L. and P. B. S.). The authors declare that they have no conflicts of interest with the contents of this article. This article contains Figs. S1-S8 and Table S1. 1  aggregation. Given that anti-platelet agents are an important class of drugs for the treatment of various cardiovascular diseases, there has been a concerted effort aimed at understanding the role of these receptors in platelet activation and in identifying molecules that can act as PAR1 and PAR4 antagonists. A PAR1 antagonist was recently identified and approved for clinical use; however, significant bleeding side effects were noted in patients administered this drug (11)(12)(13). Although bleeding is a frequently-encountered side effect of therapeutically-targeting pathways involved in blood coagulation, it was hoped that targeting PAR1 would not present such liability. Much attention has now shifted to understanding and targeting the other thrombin receptor PAR4. Although both PAR1 and PAR4 are activated by thrombin, there are some key differences reported in the activation of these two receptors on platelets that argue for PAR4 being a better anti-platelet drug target than PAR1. PAR1 possesses a hirudin-like binding site that enables it to bind thrombin with relatively high affinity (14,15). In lacking this thrombin-binding site, PAR4 can only be activated at much higher concentrations of thrombin, such as may be encountered on a growing platelet thrombus. There are also reported differences in the kinetics of PAR1-and PAR4-dependent calcium signaling, with PAR4 causing a delayed but more sustained signal compared with PAR1 (16). This is thought to be due to dual proline residues in the PAR4 exodomain that provide low-affinity interactions with the thrombin-active site and an anionic cluster that slows dissociation from thrombin (17). Overall, PAR1 is thought to initiate platelet activation in response to low concentrations of thrombin, whereas PAR4 is engaged at higher concentrations to further consolidate and stabilize the clot. Given the liabilities seen with targeting PAR1 as a platelet thrombin receptor, there has been renewed interest in developing small molecule PAR4 antagonists (18 -20). Recently, small molecule antagonists of PAR4 have been described with efficacy in inhibiting injury-induced thrombosis in both nonhuman primates and humans (21)(22)(23). Despite these promising advances, there remains the question of whether a complete inhibition of PAR4 might also lead to bleeding liability. Our efforts have therefore focused on examining biased signaling through PAR4 in order to identify pathways that are responsible for platelet aggregation and that could be selectively manipulated for therapeutic efficacy. PAR4 can activate multiple signaling pathways through coupling to G␣ q/11 , G␣ 12/13 , and ␤-arrestin (24 -26). In previous work, we and others (26,27) showed that pepducins targeting different intracellular motifs in PAR4 could be employed to differentially inhibit platelet activation by blocking G-protein-coupled or ␤-arrestin-dependent signaling. The ability to selectively activate or inhibit GPCR signaling in a biased manner is of great interest and has been proposed as a strategy for obtaining therapeutically superior drugs for a number of conditions (28 -30). In the case of the finely-balanced signaling systems involved in coagulation and platelet activation, such subtle perturbations may be key to obtaining drugs that do not result in bleeding liability.
Here, we seek to gain an understanding of the molecular interactions responsible for agonist docking to PAR4 and to identify interactions that can be manipulated to obtain biased ligands for PAR4. By screening a peptide library derived from the most widely used PAR4 agonist peptide, AYPGKF-NH 2 , we have identified key residues that confer selective activation of G␣ q/11 -coupled calcium signaling and/or ␤-arrestin recruitment. Through in silico docking and confirmatory mutagenesis experiments, we further identify an extracellular loop 2 residue (Asp 230 ) that is critical for receptor activation and signaling. We also examine platelet activation by peptides that show differential activation of PAR4 signaling. Overall, these studies advance our understanding of PAR4 activation and signaling and will help guide future drug discovery efforts targeting this receptor.

Results
To understand the rules governing agonist peptide activation of PAR4, we synthesized 37 hexapeptides through modification to the synthetic PAR4 agonist peptide AYPGKF-NH 2 . A number of different modifications were examined at each of the six positions, including alanine, D-isomer, and N-methyl substitutions, as well as other modifications aimed at further probing each of the six agonist peptide positions. Henceforth, the parental PAR4 agonist peptide for this study, AYPGKF-NH 2 (Ala 1 -Tyr 2 -Pro 3 -Gly 4 -Lys 5 -Phe 6 -NH 2 ; superscript denotes position from the N terminus), will be referred to as PAR4-AP. G␣ q/11 -coupled calcium signaling, ␤-arrestin-1/2 recruitment (␤-arr-1-rluc and ␤-arr-2rluc, respectively), and p44/42 mitogen-activated protein kinase (MAPK) signaling by each of these peptides was monitored and compared with responses triggered by PAR4-AP (Table 1). Whereas PAR4-AP-triggered G␣ q/11 signaling reached a clear plateau over the range of concentrations tested (0.3-300 M), ␤-arrestin recruitment did not, and therefore, the effective concentration required to achieve half-maximal response (EC 50 ) values could not be accurately assessed. In contrast, with the substituted peptides, G␣ q/11 signaling concentration-effect curves in some cases did not clearly plateau, whereas ␤-arrestin recruitment did. Where possible, EC 50 values are compared between PAR4-AP and substituted peptides. However, because clear maxima were not always obtained, comparisons of signal magnitude throughout this study are presented as the response elicited by a given compound at 100 M, shown as a percentage of responses elicited by PAR4-AP at the same concentration. Because the concentrationeffect curve for PAR4-AP exhibits a saturating calcium signal, peptides that failed to saturate calcium signaling responses are depicted as having EC 50 values that were "not determined" (ND). Maximal calcium signal responses are shown as a percentage of the response obtained with the calcium ionophore A23187 ( Table 1, Maximum % of A23187 calcium ionophore) and are referred to throughout the "Results" as a change in maximal calcium signaling as a measure of the efficacy of the peptides. Throughout the "Results," ␤-arrestin recruitment values are presented as a percentage (%) of the PAR4-AP-induced recruitment at 100 M concentration. For ease of comparison, concentration-effect curves for PAR4-AP-mediated G␣ q/11 calcium signaling and ␤-arrestin recruitment are displayed as a repeated reference value in Figs. 1-7.

Effect of alanine substitution on PAR4-AP-mediated signaling responses
Our first set of changes involved alanine substitutions at positions 2-6 of the PAR4-AP (calcium EC 50 ϭ 12.1 Ϯ 3.1 M;  50 or not determined (ND) if EC 50 values could not be calculated. Maximum calcium response achieved with each peptide is shown as a percentage of the maximum calcium response achievable in a given experiment revealed by calcium ionophore (% of A23187 calcium ionophore). Statistically significant shifts in EC 50 values compared with PAR4-AP, AYPGKF-NH 2 , were found using the F-statistic (*, p Ͻ 0.05). Given that saturation did not always occur, all peptides are compared with the parental PAR4-AP, AYPGKF-NH 2 , and shown as a percent of the response obtained with PAR4-AP at a given concentration (100 M). Finally, fold increase over unstimulated baseline is shown for phosphorylation of p44/42 MAPK activation. Statistical significance for comparisons at 100 M and p44/42 phosphorylation are calculated using one-way ANOVA of response compared to PAR4-AP (*, p Ͻ 0.05).

Effect of D-isomer amino acid substitutions on PAR4-APmediated signaling responses
To assess the impact of stereochemical inversion at each residue position, we generated five peptides with single amino acid substitutions from L-to their respective D-isomers (Fig. 1, D-F, and Table 1). Glycine possesses only one isomeric form and thus is not substituted. Both L-alanine to D-alanine
We were excited to observe evidence of ␤-arrestinbiased peptides within this group, all of which showed decreased calcium signaling and/or increased ␤-arrestin recruitment. Moreover, the leftward shifts in their associated ␤-arrestin concentration-effect profiles imply that substitution of N-methylated Ala 1 , Tyr 2 , Lys 5 , or Phe 6 residues increases peptide potency with respect to Gprotein-independent signaling. Only N-methylation of Gly 4 (Sar) resulted in a reduced overall ability to activate PAR4, with relative calcium and ␤-arrestin signals being reduced by ϳ90 and 60%, respectively.

Effect of N-terminal and position 1 modifications on PAR4-APmediated signaling responses
To further investigate the impact of N-terminal alterations to the PAR4-AP, we examined the impact of N-terminal acetylation and substitutions of position 1 (Fig. 2, A-C, and Table 1). Previously, it was demonstrated that protonation of the N terminus of the PAR1-tethered ligand (SFLLRN) is critical for agonist function (31). Additionally, substitution of glycine for mercaptopropionic acid (Mpr), which lacks an N-terminal protonated amine found in the native tethered ligand-mimicking peptide GYPGQV, resulted in complete loss of agonist activity (32). Given the evidence for N-terminal protonation in PAR agonism, we sought to determine the impact of N-terminal acetylation on PAR4-AP activity. N-terminal acetylation of the PAR4-AP (Ac-AYPGKF-NH 2 ) negatively impacted both calcium signaling (EC 50 ϭ ND) as well as the magnitude of ␤-arrestin recruitment to PAR4 as expected (␤-arr-1, 47.3 Ϯ 3.1%; EC 50 ϭ 15.0 Ϯ 4.3 M; ␤-arr-2, 45.5 Ϯ 2.8; EC 50 ϭ 26.7 Ϯ 5.6 M). In contrast to the previously reported complete loss of activity with Mpr-YPGQV peptide, we observed partial agonism with the loss of protonation through acetylation of PAR4-AP. This could also be influenced by the additional methyl group of the Ala in Ac-AYPGKF-NH 2 , which is not present in the Mpr-YPGQV peptide.
In both the native human PAR4-tethered-ligand sequence and the PAR4-AP, position 1 is occupied by small (tethered ligand Ϫ glycine, PAR4-AP Ϫ alanine) and nonpolar (PAR4-AP Ϫ alanine) residues. To investigate the role of position 1 in the PAR4-AP, we generated seven peptides with position 1 substitutions.

Effect of additional substitutions in positions 2-4 on PAR4-AP-mediated signaling
Previously, structure-activity studies with PAR4-activating peptides revealed an apparent requirement for an aromatic residue in position 2 (tyrosine) and found that alterations to position 3 (proline) and position 4 (glycine) were not well-tolerated (32). In this study, we observed that substitution of either tyrosine, proline, or glycine with alanine reduced calcium-signaling potency and ␤-arrestin recruitment; however, none of these substitutions completely abolished the activity of the peptide (Fig. 1, A-C). Interestingly, substitution of either L-tyrosine or L-proline with the respective D-isomer significantly reduced the activity of the PAR4-AP (Fig. 1, D-F, and Table 1). Together, these data suggested that although no individual residue is wholly required for agonist activity, the side chains of these residues and their positioning are important for agonist function. To further probe the contribution of these residues to PAR4-AP activity at PAR4, we generated peptides with alterations in the aromatic side chain of position 2, the back-
Peptides with substitution of Tyr 2 with either 4-fluoro-L-phenylalanine (A-F(4-fluoro)-PGKF-NH 2 ) or O-methyltyrosine (A-Tyr(Me)-PGKF-NH 2 ) retained calcium signaling and ␤-arrestin recruitment with no significant changes in potency compared with the PAR4-AP. These data may suggest that the side chain of tyrosine is a site in which further modifications can be made to the peptide without affecting agonism; however, by comparison with AAPGKF-NH 2 , the presence of an aromatic ring in position 2 is of importance. Given the results obtained with alanine or D-proline substitution of Pro 3 (Fig. 1, A-C and D-F, respectively), we furthered probed conformational contributions of this residue to the parental PAR4-AP. Proline was substituted with pipecolic acid (2-piperidinecarboxylic acid; Pip), which is structurally similar to proline but has an increased steric bulk and backbone elongation provided by a six-membered ring compared with the five-membered ring of proline. We observed partial agonism of calcium signaling in response to AY-Pip-GKF-NH 2 stimulation with no significant difference in agonist potency (EC 50 ϭ 30.7 Ϯ 4.9 M) and with a correlating partial agonism of ␤-arrestin recruitment (␤-arr-1, 38.8 Ϯ 3.0%; ␤-arr-2, 44.0 Ϯ 2.7%). Thus, it appears that increasing steric bulk/backbone length may affect efficacy but still results in agonism. To further investigate the role of backbone conformation provided at position 3, proline was substituted with nipecotic acid (3-piperidinecarboxylic acid; Nip) which is similar to pipecolic acid; however, it results in change of the peptide backbone direction (AY-Nip-GKF-NH 2 ). With this modification, both calcium signaling (EC 50 ϭ ND) and ␤-arrestin recruitment (␤-arr-1, 32.9 Ϯ 4.9%; ␤-arr-2, 30.9 Ϯ 4.5%) were significantly decreased in response to PAR4 stimulation. Elongation of the backbone in position 4 (glycine) by substitution with ␤-alanine (AYP-␤Ala-KF-NH 2 ) was also significantly detrimental to both calcium signaling (EC 50 ϭ ND) and ␤-arrestin recruitment (␤-arr-1,42.5 Ϯ 1.8%; ␤-arr-2, 24.9 Ϯ 1.6%). Together, these data suggest an important role for backbone conformation contributed by Pro 3 and Gly 4 compactness in the PAR4-AP. As well, these data suggest peptide backbone length at Pro 3 and Gly 4 has an important role, consistent with what we have observed with backbone elongation modifications at position 1 (Inp-YPGKF-NH 2 and ␤Ala-YPGKF-NH 2 ; Fig. 2, A-C, and Table 1).

Effect of additional substitutions in position 5 on PAR4-APmediated signaling responses
Previously, investigations into glutamine substitutions in the native tethered ligand-mimicking peptide (GYPGQV) revealed that agonism is maintained when this position is substituted with a charged residue such as arginine or ornithine (32). In this study, we have identified that substitution of position 5 (lysine) of the PAR4-AP (AYPGKF-NH 2 ) with alanine does not significantly affect the agonist capacity of the peptide (Fig. 1, A-C). Furthermore, we identified that substitution of L-lysine with D-lysine significantly decreases agonist stimulation of calcium signaling and ␤-arrestin recruitment (Fig. 1, D-F). To further probe the effect of positively-charged residues in position 5, we generated several PAR4-AP analogues with Lys 5 substitutions by either arginine or ornithine (Fig. 2, G-I, and Table 1). Substitution of Lys 5 with arginine resulted in a partial agonist with respect to calcium signaling but increased ␤-arrestin recruitment compared with the parental peptide (AYPGRF-NH 2 calcium EC 50 ϭ 4.0 Ϯ 1.1 M, ␤-arr-1, 154.2 Ϯ 10.3%; ␤-arr-2, 140.1 Ϯ 9.4%). This substitution delocalizes the positive charge of the side chain compared with lysine and increases steric bulk. Interestingly, substitution with ornithine, which is similarly charged but has a shortened distance between the charge and the peptide backbone compared with lysine, also resulted in partial agonism of PAR4-mediated calcium signaling accompanied by a modest increase in the recruitment of ␤-arrestins (AYPGOF-NH 2 calcium EC 50 ϭ 4.2 Ϯ 1.1 M, ␤-arr-1, 121.7 Ϯ 8.5%; ␤-arr-2, 112.1 Ϯ 7.7%). To further probe whether steric bulk in the side chain of position 5 affects PAR4-AP potency, we increased steric bulk through methylation of the ⑀-amine of lysine (AYPG-(N ⑀ -Me-K)-F-NH 2 ). This substitution yielded a full agonist for calcium signaling (EC 50 ϭ 6.7 Ϯ 1.9 M) and a modestly more efficacious agonist for the recruitment of ␤-arrestins (␤-arr-1, 122.5 Ϯ 12.4%; ␤-arr-2 116.7 Ϯ 11.2%). All of these substitutions maintain a positive charge and are able to act as agonists for PAR4. To determine whether charge in position 5 is required, we substituted lysine with citrulline, which maintains a similar structure to arginine and lysine while removing a positive charge (AYPG-Cit-F-NH 2 ). Interestingly, we observed partial agonism with respect to calcium signaling (EC 50 ϭ 5.4 Ϯ 0.8 M) with equipotent ␤-arrestin recruitment compared with the PAR4-AP, consistent with the alanine scan data for this position. Thus, positively charged residues in position 5 seem to contribute to agonism. Interestingly, we observed similar increases in ␤-arrestin recruitment and equipotent calcium signaling when stericity was either increased (AYPGRF-NH 2 ) or decreased (AYPGOF-NH 2 ), suggesting that the positive charge in this position is more important than the size of the side chain. Finally, although both the charge and size of the size chain have an effect, all modifications to position 5, with the exception of D-lysine ( Fig. 1, D-F, and Table 1) and N ␣ -methyl-lysine ( Fig. 1, G-I, Table 1) modifications, were generally well-tolerated, and there appears to be modest freedom in the side chain of this position. This makes the side chain a good candidate for further investigations of structure-activity relationships as well as a good region to label with an imagine modality.

Effect of additional substitutions in position 6 on PAR4-APmediated signaling responses
Through our initial investigations into position 6 substitutions, we observed that replacement of Phe 6 with either alanine, D-phenylalanine, or N-methyl-phenylalanine had varied effects on PAR4-AP potency and efficacy. Previous investigations into position 6 substitutions in the native tethered ligand (GYPGQV) revealed the requirement for an aromatic residue, as substitution with nonaromatic residues such as lysine or ornithine significantly decreased or abolished the activity of the peptide (32). In addition to being an aromatic residue, phenylalanine is a hydrophobic residue, so we investigated the impact of altering the hydrophobicity/ hydrophilicity and stericity while maintaining an aromatic residue (Fig. 2, J-L, and Table 1).

Probing antagonism of PAR4-AP-stimulated calcium signaling by peptides unable to stimulate calcium signaling
The failure of some substituted peptides to stimulate calcium signaling could reflect either a failure to bind or a loss of efficacy. To determine whether such peptides can act as antagonists of PAR4-AP-stimulated calcium signaling, we preincubated HEK-293 cells stably expressing PAR4-YFP prior to stimulation with PAR4-AP. Pretreatment and competition with 100 M of peptides unable to stimulate PAR4-dependent calcium signaling were largely unable to affect the calcium response elicited by 30 M PAR4-AP (AYPGKF-NH 2 ) compared with vehicle-treated controls. Interestingly, two peptides with substitution in position 3 or 4, which were unable to stim-ulate calcium signaling, were able to decrease calcium signals elicited by PAR4-AP. Calcium signaling in response to AYPGKF-NH 2 (30 M) was significantly reduced following pretreatment with AY-Nip-GKF-NH 2 (10.6 Ϯ 0.9% of A23187; p Ͻ 0.05) or AYP-Sar-KF-NH 2 (13.0 Ϯ 1.5% of A23187; p Ͻ 0.05) compared with nonpretreated control (20.8 Ϯ 1.3% of A23187; p Ͻ 0.05) (Fig. S2).

PAR4-mediated MAPK signaling is G␣ q/11 -dependent and ␤-arrestin-independent
To further probe the contribution of different pathways downstream of PAR4 to p44/42 phosphorylation, we examined the effect of blocking G␣ q/11 signaling with the specific antagonist YM254890 and examined the role of ␤-arrestins using a ␤-arrestin-1/2 double knockout HEK-293 cell derived by CRISPR/Cas9-mediated targeting (Fig. S5)

Homology modeling of PAR4 and in silico docking of AYPGKF-NH 2
To gain a better understanding of the ligand-binding pocket in PAR4, we turned to homology modeling and in silico docking experiments. A homology model of PAR4 was generated using known coordinates of the thermostabilized, antagonist-bound (AZ8838) human PAR2 crystal structure (2.8 Å resolution, PDB code 5NDD (36)). Sequences of human PAR4 were aligned with the structure-resolved residues of PAR2 using Clustal Omega (37,38). We generated a homology model of WT PAR2 on the PAR2 structure, which contained modifications necessary for crystallization. Twenty human PAR4 homology models, using the WT human PAR2 model as a template, were generated in MODELLER (Webb and Sali (69)). The PAR4 model with the lowest discrete optimized protein energy (DOPE) score was utilized for in silico analyses.
Docking AYPGKF-NH 2 to the PAR4 homology model (Gal-axyPepDock; Lee et al. (39)) revealed a number of highlypredicted peptide-receptor interaction sites ( Fig. 5A; Ser 67 , Tyr 157 , His 229 , Asp 230 , Leu 232 , Leu 234 , Asp 235 , Ala 238 , Gln 242 , His 306 , Tyr 307 , Pro 310 , Ser 311 , Ala 314 , Gly 316 , Tyr 319 , and Tyr 322 ). Of these, the most frequently predicted interactions were receptor residues His 229 , Asp 230 , and Gln 242 . In peptidedocked models, Gln 242 made contact with alanine, tyrosine, and lysine, with the most frequent prediction being with phenylalanine. His 229 was predicted to interact with lysine and phenylalanine. Asp 230 was frequently predicted to interact with lysine and phenylalanine, and contacts with lysine appeared in 60% of predictions (Fig. 5B). Interestingly, when visualizing receptor residues predicted to interact with the peptide, we found that a large number of these residues were localized on extracellular loop 2 (ECL2).

Calcium signaling and ␤-arrestin recruitment to predicted peptide-binding site mutant PAR4
We experimentally examined in silico binding predictions through site-directed mutagenesis introducing alanine substitutions at the three most highly-predicted interacting residues, His 229 , Asp 230 , and Gln 242 . Additionally, we generated a double mutation, PAR4 H229A,D230A , to determine whether Dashed baseline shows mean of unstimulated baseline for comparison. Dotted baseline shows mean response achieved following AYPGKF-NH 2 stimulation. (#, p Ͼ 0.05 AYPGKF-NH 2 compared with untreated control; *, p Ͼ 0.05 peptide agonists compared with AYPGKF-NH 2 ; one-way ANOVA). Peptides that were unable to stimulate calcium signaling were also generally unable to stimulate phosphorylation of p44/42. Interestingly, several peptides were able to significantly increase phosphorylation of p44/42 compared with AYPGKF-NH 2 . (Representative blots are shown in Fig. S4; n ϭ 4 -5.)

PAR4 activation and biased signaling
combined loss of His 229 and Asp 230 would result in additive detriment to receptor activation. We evaluated the functional consequences of these mutations by recording calcium signaling in response to AYPGKF-NH 2 with all mutants. We observed no calcium response in cells expressing PAR4 D230A -YFP or PAR4 H229A,D230A -YFP mutant receptors up to 300 M AYPGKF-NH 2 (Fig. 5C). Similarly, peptide-induced ␤-arrestin-1/2 recruitment to PAR4 D230A -YFP and PAR4 H229A,D230A -YFP was significantly reduced compared with the WT PAR4. Both PAR4 H229A and PAR4 Q242A mutations were fully able to recruit ␤-arrestin-1/2 to the same levels observed with WT PAR4-YFP (Fig. 5, D and E).
To investigate whether these sites are important for tethered ligand activation of the receptor, we stimulated cells expressing WT and mutant constructs with thrombin. Thrombin cleavage of the receptor reveals the native tethered ligand, GYPGQV (40). Comparable with activation with AYPGKF-NH 2 , we observed significantly diminished calcium signaling in HEK-293 cells expressing PAR4 D230A -YFP and PAR4 H229A, 230A -YFP mutants, even up to 10 units/ml thrombin (Fig. 5F). Additionally, ␤-arrestin-1/2 recruitment was also impaired by PAR4 D230A and PAR4 H229A, D230A mutants (Fig. 5, G and H). We observed no appreciable difference in calcium signaling or ␤-arrestin recruitment when PAR4 H229A -YFP and PAR4 Q242A -YFP mutants were stimulated with thrombin, with the exception of a modest decrease in ␤-arrestin-1 recruitment to the PAR4 Q242A -YFP mutant receptor. Thus, Asp 230 interactions with either PAR4-AP or the tethered ligand are critical for receptor activation.
To examine whether differences in cell-surface expression of the mutant receptors could underlie the defect in signaling, we employed confocal microscopy to examine receptor expression and localization. We did not find any differences in expression levels and found appropriate cell-surface localization of the mutant receptors (Fig. S6).

Platelet aggregation response to differential activation of PAR4 signaling
To determine the effects of targeted signaling observed with several peptides, we investigated PAR4-mediated platelet aggregation in response to the ␤-arrestinbiased, calcium-signaling null peptide AyPGKF-NH 2, as well as a modestly more potent calcium agonist AYPGRF-NH 2 (compared with AYPGKF-NH 2 ). Thrombin (1 unit/ml) and 100 M AYPGKF-NH 2 were used as controls to ensure successfullywashed platelet preparations. In response to thrombin stimulation, we observed a characteristic aggregation trace resulting in 100% aggregation at 1 unit/ml. In agreement with previously published data, AYPGKF-NH 2 (100 M) stimulation resulted in ϳ50% platelet aggregation (26). We observed that platelet aggregation was not stimulated by 100 M AyPGKF-NH 2 . Interestingly, when platelet preparations were stimulated with AYPGRF-NH 2 , in which we observed a modest increase in potency in calcium-signaling assays,

Discussion
PAR4 is emerging as a novel anti-platelet drug target. Here, we examine structure-activity requirements for PAR4 activation of calcium signaling, ␤-arrestin recruitment, and MAPK activation. PAR cleavage by a number of different enzymes has been reported (41). Canonically, thrombin has been described as a PAR1 and PAR4 activator, and trypsin is described as a PAR2 and PAR4 activator. Although PAR3 can be cleaved by thrombin, its ability to act as a signaling recep-tor remains ambiguous. In each case, cleavage by serine proteases occurs at an arginine-serine site, which reveals a unique tethered ligand specific to each PAR (42)human PAR1-SFLLRNP… (6, 15), human PAR2-SLIGKV… (43,44), and human PAR4-GYPGQV… (40,45). Although these tethered-ligand sequences are diverse, they retain some highly-conserved properties, including a highly-conserved charged residue in position 5 of the revealed tethered ligands (arginine in PAR1; lysine in PAR2). Additionally, PAR2 and PAR4 have a conserved glycine in position 3 and valine in position 6 of their tethered ligands. The similarities of the proteolytically-revealed tethered ligands are also observed in the optimized tethered ligand-mimicking synthetic agonist peptides. Hydrophobic residues leucine, isoleucine, and proline are preferred in position 3 . Agonist-stimulated ␤-arrestin-1/2 recruitment: HEK-293 cells were transiently transfected with BRET pair of either ␤-arrestin-1-rluc or -2-rluc and PAR4-YFP or site-directed mutant PAR4-YFP (described above), and agonist-stimulated arrestin recruitment to PAR4 was monitored. Cells were incubated with increasing concentration of either AYPGKF-NH 2 (D and E) or thrombin (G and H) for 20 min, and BRET values were recorded. BRET values from cells treated with HBSS were also collected to normalize the data. Renilla luciferase substrate coelenterazine-h (5 M) was added 10 min prior to data collection. Data are presented as the normalized net BRET ratio (eYFP/rluc Ϫ baseline (HBSS control)). AYPGKF-NH 2 concentration is in mass, and thrombin concentration is shown as units (n ϭ 4).

PAR4 activation and biased signaling
of PAR1, PAR2, and PAR4 agonist peptides, respectively. Additionally, there is a basic residue in position 5 of the tethered ligand peptide in all three receptors (arginine in PAR1 and PAR2 and lysine in PAR4 activating peptides). These conserved residue properties highlight some of the key requirements that a PAR4 agonist peptide must meet. Previous structure-activity work starting with the proteolytically-revealed sequences of human (GYPGQV) and murine (GYPGKF) PAR4 identified the hexapeptide AYPGKF-NH 2 or SYPGKF-NH 2 to be potent and selective agonists for PAR4 (32). In this study, we started with the hexapeptide AYPGKF-NH 2 as our parental peptide agonist and incorporated various modifications at each of the six positions in this peptide to investigate determinants of agonist-induced PAR4-coupling to G␣ q/11 , MAPK, and ␤-arrestin-1/2 recruitment to guide development of biased agonists of PAR4.
Previously, it was demonstrated that substitution of alanine or serine in position 1 of the GYPGKF-NH 2 -activating peptide retained the ability to trigger tritiated inositol 1,4,5-triphosphate release to a level comparable with that observed with thrombin activation and much higher than the release caused by the GYPGKF-NH 2 (32). Interestingly, substitution of position 1 with threonine was reported to significantly reduce the potency of the activating peptide (32). In keeping with results from previous studies, we find a requirement for small aliphatic side chains in position 1 of PAR4-activating peptides, with substitutions that increased bulk in this position consistently resulting in poor agonists for both calcium signaling and ␤-arrestin recruitment (N-Me-A Fig. 1, G-I; N-Me-S, Aib, Inp, V, Nle in Fig. 2, A-C; and Table 1). Additionally, we observed that both stereochemical inversion of L-alanine to D-alanine (D-alanine, Fig. 1, D-F) and backbone elongation (␤Ala, Fig. 2, A-C) were detrimental to both calcium signaling and ␤-arrestin recruitment. Additionally, we observed that capping of the N-terminal charge via acetylation significantly decreases agonist potency. Together, these results indicate that a positive charge and the localization of that charge in addition to precise side-chain positioning at position 1 are critical for effective PAR4-agonism.
The aromatic side chain, and to a lesser extent the hydroxyl group, of the Tyr 2 residue of the parental peptide also proved to be important for peptide activity with alanine or D-isomer substitutions resulting in decreased activity of all pathways tested. These results are consistent with previous reports that replacement of tyrosine with other aromatic residues, such as phenylalanine, significantly reduced both the selectivity and the potency of the activating peptide (32). Interestingly, substituting L-Tyr 2 for either D-tyrosine (Figs. 1D and 3) or N-methyltyrosine (Figs. 1G and 3) resulted in significant decreases in calcium and p44/42 signaling, while having relatively modest effects on ␤-arrestin recruitment (Fig. 1, E-F, H, and I, respectively, and Table 1). Also, substitution of tyrosine with either 4-fluorophenylalanine or O-methyltyrosine did not significantly affect calcium transients, p44/42 phosphorylation, or ␤-arrestin recruitment compared with AYPGKF-NH 2 (Figs. 2, D-F and 3, and Table 1). Our findings, together with previously reported data (32), point to the necessity of an aromatic residue in position 2 to maintain agonist potency. Furthermore, our data reveal that backbone orientation in position 2 is a key determinant of agonist potency. Future experiments fine-tuning the aromatic substituents may be beneficial for peptide analogue development.
Our data reveal that alterations affecting proline backbone conformation or glycine residue compactness are not well-tolerated, as these decreased calcium signaling, p44/42 MAPK signaling, and ␤-arrestin recruitment to PAR4. Calcium signaling has previously been reported to be negatively impacted by alterations to either of these two residues proline and glycine (32). Both proline and glycine residues are frequently found in ␤-turns, in which proline provides a characteristic kink, and the available data support the notion that the conformation(s) (Pro 3 and Gly 4 ) and rigidity (Pro 3 ) of PAR4 agonists permitted by these residues is/are important for receptor activation. To test this hypothesis further, we made several substitutions in position 3 (proline) aimed at evaluating favored backbone conformations. We observed that a return to a predominantly trans-amide conformation by substitution of proline with alanine (Fig. 1, A-C), results in decreased potency in all pathways studied. Interestingly, this substitution did not abolish signaling but did significantly decrease agonist potency, which suggests that the backbone conformation of proline may be necessary for receptor binding but not receptor activation. Substitution of proline to pipecolic acid (Pip., Fig. 2, D-F), which has a similar peptide backbone orientation to proline with a six-membered aromatic ring, decreased ␤-arrestin recruitment with relatively modest effects for calcium signaling activation. Interestingly, substitution of proline with nipecotic acid (Nip., Fig. 2, D-F) abolished both calcium signaling and p44/42 phosphorylation while also significantly reducing ␤-arrestin recruitment. Nipecotic acid is similar to pipecolic acid with the exception that nipecotic acid changes the direction of the peptide backbone compared with that adopted by proline, whereas pipecolic acid favors a similar backbone directionality and conformation similar to proline. Interestingly, we observed a similar pharmacological profile observed with AY-Nip-GKF-NH 2 when we substituted D-proline in position 3, which also changes the backbone orientation from that of the parent peptide. These findings demonstrate that the backbone conformation provided by proline in position 3 of the parental peptide is neces- Figure 6. PAR4 agonist-mediated platelet aggregation. Both thrombin (1 unit/ml; shown in black) and AYPGKF-NH 2 (100 M; shown in gray) initiated PAR4-dependent rat platelet aggregation. We observed that stimulation of washed platelet samples with a ␤-arrestin biased peptide that is unable to stimulate calcium signaling, AyPGKF-NH 2 (100 M; shown in yellow), did not stimulate platelet aggregation. Furthermore, stimulation of washed platelets with the peptide agonist AYPGRF-NH 2 (100 M; shown in purple) caused a robust platelet aggregation comparable with the response with 1unit/ml thrombin response (n ϭ 2-4; representative trace shown).

PAR4 activation and biased signaling
sary for agonism of PAR4-signaling pathways. Furthermore, substitution of a Pro 3 with alanine (AYAGKF-NH 2 , Fig. 1, A-C) significantly and detrimentally altered the pharmacological profile compared with the parental peptide PAR4-AP, suggesting the possibility that agonist competency contributed by position 3 is largely conformational and not chemical. Substitution of sarcosine or ␤-alanine at position 4 also negatively affected all signaling pathways, whereas substitution of Gly 4 with alanine resulted in a decrease in potency with calcium signaling as well as a reduction of ␤-arrestin recruitment. Together, these substitutions reveal the requirement for a small flexible residue in position 4, consistent with previous reports (32).
It has been previously demonstrated that substitution of position 5 of glutamine in the native tethered ligand (GYPGQV) with arginine or ornithine results in decreased potency in the PAR4 agonist peptide for calcium-signaling pathways (32). However, there is a basic residue found in position 5 of PAR1-(arginine, TFLLR), PAR2-(arginine, SLIGRL), and PAR4 (lysine, AYPGKF)-tethered ligand-mimicking peptides. Given the conservation of charged residues in position 5 of PAR-tethered ligand-mimicking peptides, we generated PAR4-AP-derivative peptides with Lys 5 substitutions that maintained or altered charge in this position to determine whether this is a requirement for activity at PAR4. Substitution of Lys 5 with alanine had no appreciable effect on calcium or MAPK signaling, while resulting in decreased ␤-arrestin recruitment (Fig. 1, A-C). Substitution of L-Lys 5 with D-lysine, however, significantly reduced potency and efficacy in calcium signaling and ␤-arrestin recruitment assays and furthermore resulted in a loss of stimulation of MAPK signaling (Figs. 1,  D-F, and 3). We continued our exploration of this position with an N ␣ -methyl-lysine substitution, which increases steric bulk through the addition of a methyl group to the backbone while also removing a hydrogen bond donor and increasing the amount of peptide in the cis-transformation of the amide bond between positions 4 and 5. We found this substitution detrimental to calcium signaling but advantageous for ␤-arrestin recruitment, resulting in comparable maximal levels of recruitment to those observed with PAR4-AP but at lower concentrations (Fig. 1, G-I). Thus, backbone conformation and bulk at position 5 are important for agonism of calcium signaling and MAPK activation, while possibly being an important site for modification in the search for ␤-arrestinbiased ligands.
We next turned our attention to analogues that resemble the side chain of Lys 5 , with a specific interest on structure-activity in addition to ascertaining the importance of charge in this residue. In contrast to the observation with similar substitutions to the tethered ligand-mimicking peptide (GYPGQV) (32) wherein signaling was decreased compared with PAR4-AP, we found that both calcium signaling and ␤-arrestin recruitment are only modestly impacted by position 5 substitution to ornithine in the PAR4-AP, which decreases steric bulk and shortens the distance between the backbone and the positively charged amine compared with lysine (AYPGOF-NH 2 ; Fig. 2, G-I).
Interestingly, however, we observed significantly increased p44/42 phosphorylation in response to AYPGOF-NH 2 stimulation (Fig. 3). This is somewhat puzzling because we found p44/42 phosphorylation to be G␣ q/11 -dependent (Fig. 4A); however, it is important to note that MAPK and other cellular signaling pathways downstream of receptor activation are subject to amplification/deamplification processes within the cell (46), and these observations may point to additional differences in signaling stimulated by this peptide that needs to be examined. We further probed stericity in this position through increasing bulk with substitution of Lys 5 with arginine (AYPGRF-NH 2 ). Interestingly, this substitution also results in a delocalization of charge compared with the charge on the Lys 5 side chain. This modification provided a similar calcium-signaling profile compared with that observed with ornithine substitution; however, it also resulted in a significant increase of ␤-arrestin recruitment compared with the parental PAR4-AP. Interestingly, there was no significant change in p44/42 phosphorylation unlike what was observed with ornithine substitution. Another peptide with increased steric bulk, N ⑀ -methyllysine substitution in position 5, improved calcium-signaling potency, modestly improved ␤-arrestin recruitment, and significantly increased p44/42 phosphorylation (AYPG-(N ⑀ -Me-K)-F-NH 2 , Fig. 2, G-I). We further investigated whether charge was a significant determinant of agonism by substituting Lys 5 with citrulline (AYPG-Cit-F-NH 2 ), which lacks charge but retains a similar structure compared with arginine and, to a lesser extent, lysine. We observed decreased calcium, equivalent ␤-arrestin recruitment, and significantly increased p44/42 phosphorylation compared with the parental peptide (Figs. 2,  G-I, and 3). Together, these data reveal that changes to sidechain charge, bond donors/acceptors, and steric bulk result in varied pharmacological responses that may be exploited to bias signaling responses in favor of ␤-arrestin recruitment (AYPGRF-NH 2 ) or p44/42 phosphorylation (AYPGOF-NH 2 ; AYPG-(N ⑀ -Me-K)-F-NH 2 ) with modest improvements to potency in calcium signaling. Interestingly, we observed that charge is not wholly required for receptor activation as both alanine and citrulline substitutions resulted in agonism of PAR4.
Finally, our structure-activity investigation turned to position 6, phenylalanine, of the parental peptide. Given the significant impact that alterations to position 2 (tyrosine) had on agonist potency, we predicted that there may be astacking interaction between tyrosine and phenylalanine, which is supported by the backbone kink provided by proline in position 3. Substitution of position 6 with alanine or D-phenylalanine reduced calcium-signaling potency while reducing ␤-arrestin recruitment and p44/42 phosphorylation. These alterations suggest that phenylalanine in position 6 is not essential for PAR4 activation by agonist peptide but is important for agonist potency and full activation of all signaling pathways. Methylation of the backbone nitrogen in phenylalanine (AYPGK-(N-Me-F)-NH 2 ) resulted in improved activation of calcium signaling, ␤-arrestin recruitment, and p44/42 phosphorylation compared with AYPGKF-NH 2 (Fig. 3). Addition of either methyl (AYPGK-F(4-Me)-NH 2 ), fluorine (AYPGK-F(4fluoro)-NH 2 ), or hydroxyl (AYPGKY-NH 2 ) (Fig. 2, J-L) off of the para-position of the aromatic side chain of phenylalanine was deleterious to all signaling pathways studied. Substitution of unnatural residues that increase the size of steric bulk and -stacking potential yielded mixed results based on the orien-

PAR4 activation and biased signaling
tation of the naphthyl. Substitution of phenylalanine with (1-naphthyl)-L-alanine (1Nal) increased p44/42 phosphorylation but decreased both calcium signaling and ␤-arrestin recruitment (Fig. 2, J-L). Changing the alkyl chain connection from position 1 to position 2 of the naphthyl (AYPGK-2Nal-NH 2 ; Fig. 2, J-L) resulted in activity similar to AYPGKF-NH 2 . With (2-naphthyl)-L-alanine showing some flexibility in the steric bulk of position 6, it would be interesting for future work to investigate increasing the -stacking potential at position 2 while simultaneously increasing the -stacking potential at position 6, because we hypothesize that these two residues' aromatic regions of the PAR4-AP may interact.
To determine whether any of the peptides that failed to stimulate PAR4-dependent calcium signaling were able to act as antagonists, cells were pretreated with these peptides and subsequently stimulated with AYPGKF-NH 2 to see whether the parental peptide was able to still elicit a signal. We observed that both AY-Nip-GKF-NH 2 and AYP-Sar-KF-NH 2 were both able to significantly decrease the calcium signal elicited by PAR4-AP (Fig. S2). Interestingly, each of these substitutions resulted in ablation of both calcium and p44/42 signaling as well as significantly decreasing ␤-arrestin recruitment. These results may indicate that targeting PAR4 with small molecules mimicking positions 3 and 4 dipeptides may provide a starting place for the design of novel PAR4 antagonists. Additionally, peptides that further explore modifications to positions 3 and 4 could be investigated for potent PAR4 antagonists.
In platelets, it has previously been demonstrated that PAR4mediated calcium mobilization induces activation and aggregation (24,47). Both thrombin and AYPGKF-NH 2 initiated PAR4-dependent platelet aggregation to levels comparable with those previously reported in the literature in the same rat platelet system (26). As was hypothesized, we observed that stimulation of washed platelet samples with a ␤-arrestinbiased peptide (AyPGKF-NH 2 ; 100 M) that is unable to stimulate PAR4-dependent calcium signaling did not stimulate platelet aggregation (Fig. 6). Furthermore, stimulation of washed platelets with the agonist peptide, AYPGRF-NH 2 (Fig.  2, G-I), enhanced platelet aggregation to levels comparable with those observed with thrombin activation of PAR4 (Fig. 6). AYPGRF-NH 2 stimulates calcium signaling to levels comparable with those seen with PAR4-AP but triggers more ␤-arrestin-1 recruitment than the parental peptide. These data suggest that calcium signaling is essential for triggering platelet activation, but ␤-arrestin recruitment may also play a role.
Having ascertained some of the governing agonist peptide residue characteristics enabling PAR4 activation and validated their signaling consequences in an ex vivo system, we turned our attention to investigating the ligand-binding site of the receptor. The extracellular loops of many GPCRs have been demonstrated to be important facilitators of ligand entry into the orthosteric site, with ECL2 possessing the most diversity in sequence and structure (48,49). Analysis of common mechanisms of GPCR-ligand interaction derived from crystal structure and biochemical experiments reveals a possible role of ECL2 secondary loop in binding and facilitating entry, whereas the diversity of residue sequences across class A GPCRs is thought to confer their ligand specificity (48,49). In many class A GPCRs, a disulfide bond between ECL2 and a transmembrane 3 cysteine (Cys 3.25 , Ballesteros-Weinstein numbering) is thought to stabilize the extracellular topology such that ligand recognition at ECL2 and subsequent orthosteric binding can occur (50). All PAR receptors possess this canonical TM3 Cys 3.25 (PAR1 Cys-175 3.25 , PAR2 Cys-148 3.25 , PAR3 Cys-166 3.25 , and PAR4 Cys-149 3.25 ), and residues adjacent to this conserved cysteine are important for ligand binding and receptor activation. Investigations into the role of PAR ECL2 domains have previously demonstrated that all PAR receptors possess a homologous CHDXL motif (…CHDVL…-PAR1 C254-L258 and PAR2 C226-L230 ; …CHDVH… PAR3 C245-L249 …CHDAL; … PAR4 C226-L230 ) in their ECL2 (40,51). Given that the activating peptides of PAR1 and PAR2 (51,52) are similar (SFLLRN, PAR1; SLIGRL, PAR2), additional studies were conducted that revealed that ligand specificity is conferred by acidic residues distal to the CHDXL motif (PAR1, 259 NETL 262 ; PAR2 231 PEQL 234 (52)(53)(54). Interestingly, the equivalent human PAR4 ECL2 residues distal to the CHDAL domain, 233 PLDA 236 , share sequence homology with those distal to the human PAR2 CHDVL domain ( 231 PEQL 234 ), both of which contain an acidic residue. Al-Ani et al. (51) demonstrated that glutamic acid residues in the rat PAR2 ECL2 (PEEVL) conferred specificity to PAR2 ligands. In previous studies, a series of aspartic acids in the PAR4 extracellular loop 2 (ECL2) and N-terminal (N)-tail were proposed to be key for thrombin binding to PAR4 (55)(56)(57). Mutations of ECL2 aspartic acid residues D224A, D230A/D235A (double mutation), and D22A/D230A/D235A mutants were all found to decrease thrombin cleavage of PAR4 and activation in yeast models (57). Furthermore, the activity of the PAR4 agonist peptide AYPGKF-NH 2 was also abolished in PAR4 mutants containing the ECL2 single mutant D224A, double mutant D230A/ D235A, or triple mutant D224A/D230A/D235A aspartic acid mutations to alanine which indicate that these aspartic acid residues may be important for both agonist peptide and tethered ligand bindings and activation of PAR4. The D230A/ D235A mutation to alanine removed two acidic residues in the highly-conserved …CHD 230 AL domain (PAR1 Asp 256 ; PAR2 Asp 228 ) as well as the homologous PLD 235 A… specificity-conferring domain observed in PAR1 (Glu 260 ) and PAR2 (Glu 232 ) thus making it difficult to assess whether the loss of cleavage and activation is due to the CHD 230 AL or PLD 235 A aspartic acid.
In this study, in silico docking to a homology model of PAR4 revealed 17 predicted sites of interaction with AYPGKF-NH 2 (Ser 67 , Tyr 157 , His 229 , Asp 230 , Leu 232 , Leu 234 , Asp 235 , Ala 238 , Gln 242 , His 306 , Tyr 307 , Pro 310 , Ser 311 , Ala 314 , Gly 316 , Tyr 319 , Tyr 322 ; Fig. 5 A). Interestingly, these predicted sites reveal residues that have been shown or predicted to have a role in PAR4 activation by thrombin or agonist peptide. These sites include the aspartic acid residue Asp 230 found in the highly-conserved CHDXL domain in PARs as well as Asp 235 in the downstream region that is homologous to the specificity-conferring motif found in PAR1 and PAR2 (57). Additionally, previous in silico docking revealed a role for aspartic acid and histidine in this motif in binding of AYPGKF-NH 2 to PAR4 (58). We generated mutations of the most frequently-predicted interaction sites (His 229 , Asp 230 , Gln 242 , and His 229 /Asp 230 ) to alanine, which PAR4 activation and biased signaling revealed that Asp 230 is essential for productive interaction of the peptide with the receptor, as mutation to alanine abolished calcium signal potentiation as well as ␤-arrestin recruitment. Interestingly, this loss of activity was observed with both peptide agonist AYPGKF-NH 2 as well as with thrombin-stimulated activation of PAR4, which agrees with previous reports of the importance of these sites while also elucidating the importance of the conserved CHD 230 AL motif in PAR4 activation, consistent with the role of this domain in PAR1 and PAR2 (51-54, 57, 59). Another frequently predicted residue belonging to the CHDAL domain, His 229 , when mutated to alanine did not alter the ability of either AYPGKF-NH 2 or the tethered ligand as revealed by thrombin to activate calcium signaling or ␤-arrestin recruitment pathways. Furthermore, we observed that a double mutation of both of these sites to alanine (H229A/D230A) did not result in any additive detriment to PAR4 activation; therefore, we conclude that Asp 230 interaction with PAR4 peptide and the tethered ligand agonist performs an essential role in activation of PAR4.
Finally, in the context of platelet activation, we find that our modified peptides that are able to cause a greater level of calcium signaling and ␤-arrestin recruitment than PAR4-AP could also trigger greater platelet aggregation. In our hands, PAR4-AP was unable to elicit greater than 50% of the aggregation observed in response to 1 unit/ml thrombin, whereas the peptide AYPGRF-NH 2 was able to trigger aggregation comparable with that seen with thrombin. A role for PAR4-dependent calcium signaling is well-established in platelet aggregation (47,60,61), and consistent with this, AyPGKF-NH 2 , which is unable to stimulate calcium signaling, failed to induce aggregation. A role for ␤-arrestin-mediated PAR4 signaling in platelets has also been demonstrated (26,62). Together, these data suggest that calcium signaling is critical for initiating PAR4-mediated platelet aggregation, but the precise role and hierarchy of the different PAR4-signaling pathways in platelet responses remains to be fully established.
In summary, this study identifies key mechanisms involved in PAR4 activation and signaling. We also present and characterize a novel toolkit of PAR4 agonist peptides that could be used to study biased signaling through PAR4 in platelets as well as other physiological systems. We further show that platelet activation by PAR4 critically depends on the G␣ q/11 pathway, and selective targeting of this pathway might yield a useful anti-platelet agent. Overall, this study advances our understanding of agonist binding to PAR4 and will support future efforts aimed at defining signaling contribution in homeostatic signaling and the discovery of novel therapeutics targeting PAR.

Molecular cloning and constructs
The plasmid encoding PAR4-YFP was cloned and validated as described previously (26). The use of an in-frame, C-terminal eYFP fusion with PAR4 has been utilized in several published studies, and such modification does not significantly alter signaling downstream of PAR4 (26,(62)(63)(64). Plasmid DNA mutations in the predicted orthosteric binding site of PAR4 were generated using the QuikChange XL multisite-directed mutagenesis kit (Agilent Technologies, Mississauga, Ontario, Canada) to generate all mutants described in this study. All constructs were verified by Sanger sequencing (Robarts DNA Sequencing Facility, University of Western Ontario). Confocal microscopy of HEK-293 cells transfected with PAR4-YFP and PAR4-YFP mutants generated by site-directed mutagenesis was employed to ensure proper localization of receptors to the plasma membrane (Fig. S6). Confocal microscopy was conducted on an Olympus FV1000 (ϫ60 magnification) following fixation of samples with 4% methanol-free formaldehyde and nuclear staining with 4Ј,6-diamidino-2-phenylindole dihydrochloride (1:1000). Renilla luciferase-tagged ␤-arrestin-1/2 constructs were a kind gift from Michel Bouvier (Université de Montréal). PX458 vector for CRISPR/Cas9 targeting was a kind gift from Feng Zhang (Massachusetts Institute of Technology; Addgene catalogue no. 48138).

Site-directed mutagenesis of predicted PAR4 receptor-peptide interaction sites
Site-directed mutagenesis using Agilent QuikChange-XL was undertaken to generate mutations in highly-predicted receptor residues to validate the GalaxyPepDock model and to determine key binding-site residues for the native PAR4-tethered ligand and PAR4 agonist peptide AYPGKF-NH 2 . Residues were mutated by single-nucleotide substitutions replacing native residues with alanine. Successful mutations were confirmed by DNA sequencing (London Regional Genomics Centre, London, Ontario, Canada).

Cell lines and culture conditions
All media and cell culture reagents were purchased from Thermo Fisher Scientific. HEK-293 cells (ATCC; Manassas, VA) and CRISPR/Cas9 HEK-293 ␤-arrestin double knockout (␤-arr. knockout) cell lines were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 1% sodium pyruvate, and 1% streptomycin/penicillin. Cells stably transfected with PAR4-YFP vector were routinely cultured in the above media supplemented with 600 g/ml G418 sulfate (Geneticin; Thermo Fisher Scientific, Waltham, MA). Because trypsin activates PAR4, cells were routinely subcultured using enzyme-free isotonic phosphate-buffered saline (PBS), pH 7.4, containing 1 mM EDTA followed by centrifugation to remove PBS/EDTA, and plated in appropriate culture plates for further experimentation. Cells were transiently transfected with appropriate vectors using a modified calcium phosphate transfection (nuclease-free water, 2.5 M CaCl 2 , and 2ϫ HEPES-buffered saline). Transient transfections were rinsed PAR4 activation and biased signaling with PBS 24 h post-transfection and replaced with growth media. Transiently transfected cells were assayed or imaged at 48 h post-transfection.

General methods for peptide synthesis
All reagents were purchased from Sigma, ChemImpex, or Thermo Fisher Scientific and used without further purification. Peptides were synthesized using standard Fmoc-solid-phase peptide synthesis (SPPS) on Rink-amide 4-Methylbenzhydrylamine (MBHA) resin (256 mg, 0.1 mmol, and 0.39 mmol/g) with a Biotage Syrowave TM automated peptide synthesizer (0.4 mmol of O-(1H-6-chlorobenzotriazole-1-yl)-1,1,3,3-tetramethyluronium hexafluorophosphate (HCTU), 0.4 mmol of Fmoc-amino acids, and 0.6 mmol of DIPEA, 1 h coupling). Some peptides also involved additional manual synthesis (see procedures below). Following automated and manual synthesis, peptides were cleaved off the resin by being shaken in a cleavage mixture solution of 95% TFA, 2.5% TIPS, and 2.5% H 2 O (4 ml, 5 h), precipitated in ice-cold tert-butyl methyl ether, lyophilized, purified by preparative RP-HPLC, and further lyophilized to obtain a dry powder. Purity was assessed by analytical RP-HPLC and characterized by mass spectrometry (MS). The analytical RP-HPLC was performed on a system consisting of an analytical Agilent Zorbax SB-C8 column (4.6 ϫ 150 mm, 5 m), Waters 600 controller, Waters in-line degasser, and Waters Masslynx software (version 4.1). Two mobile phases were used: eluent A (0.1% TFA in acetonitrile) and eluent B (0.1% TFA in MilliQ water). The flow rate was set at 1.5 ml min Ϫ1 over 10 min with an additional 5-min wash (95% solvent A in solvent B). A Waters 2998 Photodiode array detector (200 -800 nm) and an ESI-MS (Waters Quattro Micro API mass spectrometer) were used to monitor the column eluate. The preparative RP-HPLC used the same system, eluents, and detection method as mentioned above for the analytical RP-HPLC, except that a preparative Agilent Zorbax SB-C8 column (21.2 ϫ 150 mm, 5 m) at a flow rate of 20 ml min Ϫ1 was used. The mass spectra for all peptides were determined in positive mode using an ESI ion source on either a Bruker micrOToF II mass spectrometer or a Xevo QToF mass spectrometer (Table S1). All peptides had purity Ͼ95% as determined by analytical HPLC (see Fig. S1).

Synthesis of Ac-AYPGKF-NH 2
The hexapeptide (AYPGKF) was synthesized using the standard automated Fmoc-SPPS discussed under "General methods for peptide synthesis." While still on resin, with all of the side chains protected and with the N-terminal Fmoc-deprotection completed, the N terminus of this peptide was acetylated by being shaken in a solution of acetic anhydride and DMF (v/v 1:4, 5 ml, 30 min). Following this, the peptide was cleaved, purified, and analyzed as mentioned under "General methods for peptide synthesis."

Synthesis of AYPG-(N ⑀ -Me-K)-F-NH 2
The hexapeptide (AYPGKF) was synthesized using the standard automated Fmoc-SPPS discussed under "General methods for peptide synthesis," except that Fmoc-Lys(Alloc)-OH was used in place of standard Fmoc-Lys(Boc)-OH. While still on resin, with all of the side chains protected, and with the N-terminal Fmoc-deprotection completed, the N terminus of this peptide was Boc-protected by adding Boc-ON (98.5 mg, 0.4 mmol) and DIPEA (70 l, 0.4 mmol) in DMF (4 ml) and shaken (4 h). The solution was removed, and this Boc protection was repeated overnight (20 h) to ensure reaction completion. The resin was washed three times with DMF (three times with 5 ml) and dry DCM (three times for 5 ml) and was placed under N 2 for site-selective Alloc deprotection of the on-resin lysine side chain. Phenylsilane (296 l, 2.4 mmol) in dry DCM (2 ml) was added to the resin. Tetrakis(triphenylphosphine) palladium(0) (23.1 mg, 20 mol) was dissolved in dry DCM (1 ml) and added to the resin. The peptide column was flushed with N 2 (2 min) followed by shaking (5 min). The resin was washed three times with dry DCM (three times with 5 ml). The procedure was repeated and shaken again (30 min). The resin was washed four times with DCM (four times with 5 ml) and DMF (four times with 5 ml). Following this N-terminal Boc protection and subsequent side chain Alloc deprotection, N-methylation of the on-resin lysine side chain was completed through the site-selective N-methylation procedure mentioned above. Following completion of the synthesis, the peptide was cleaved, purified, and analyzed as discussed under "General methods for peptide synthesis."

SPPS reaction monitoring
Three methods were used to monitor SPPS reactions. Two methods (the Kaiser test and the small-scale resin cleavage) were used to monitor general peptide synthesis. Three methods (the Kaiser test, the small-scale resin cleavage, and the Chloranil test) were used to monitor synthesis of peptides involving N-methylation. For the Kaiser test, several resin beads were placed in a test tube followed by the addition of 42.5 mM phenol in ethanol (50 l), 20 M potassium cyanide in pyridine (50 l), and 280.7 mM ninhydrin in ethanol (50 l). The mixture was then heated (100°C, 5 min), and a positive test indicates the presence of a free primary amine. For the small-scale resin cleavage method, several beads and cleavage mixture (500 l) were shaken and worked up as per the full cleavage of the resin procedure described above, and the desired peptide was confirmed through HPLC-MS. For the Chloranil test, several resin beads were placed in a test tube followed by the addition of 357.8 mM acetaldehyde in DMF (50 l) and 81.3 mM p-chloranil in DMF (50 l) and allowed to stand at room temperature (5 min), where a positive test indicates the presence of a free secondary amine.

Calcium-signaling assay
HEK-293 cells or HEK-293 cells stably expressing PAR4-YFP were loaded with the calcium-sensitive dye (Fluo-4 NW, F36206, Life Technologies, Inc., and Thermo Fisher Scientific) and aliquoted into cuvettes with Hanks' buffered salt solution (HBSS containing Ca 2ϩ and Mg 2ϩ ). Samples were excited at 480 nm, and emission was recorded at 530 nm. Baseline fluorescence from PAR4-YFP excitation remains constant throughout the experiment. Following baseline recording, agonist-induced changes in fluorescence were recorded. Change in fluorescence from baseline to maximum response at a given agonist concentration (⌬) were expressed as a percentage of maximum cellular response elicited by the calcium ionophore (A23187, Sigma).

Calcium-signaling assay with PAR4 orthosteric site-directed mutants
HEK-293 cells were transiently transfected with 2 g of WT YFP tagged PAR4 or PAR4 mutant receptors (described above) using the calcium phosphate transfection protocol as described above. Calcium signaling was investigated as described above for the WT receptor.

Testing peptide library specificity for PAR4
Previous studies have pointed to the possibility of PAR agonist peptides cross-activating other members of this family. For example, the PAR1-tethered ligand peptide SFLLRN is able to activate PAR2, and substitution of a phenylalanine residue in certain PAR4 agonist peptides confers PAR1 activity (32,52,67). Activating peptides typically requires an aromatic side chain in position 2 (Phe in PAR1), and we wondered whether substitution of tyrosine for phenylalanine may decrease selectivity at PAR4 and increase cross-talk with PAR1. We examined whether substitutions made to AYPGKF-NH 2 affected their selectivity using a calcium-signaling assay in HEK-293 cells that endogenously express PAR1 and PAR2. None of the peptides tested elicited a signaling greater than 5% of the calcium ionophore A23187 showing that they retain selectivity toward PAR4 (Fig. S7). Additionally, we investigated specificity using BRET for ␤-arrestin-2 recruitment to cells expressing PAR4-YFP or PAR2-YFP. None of the peptides were able to induce ␤-arrestin PAR4 activation and biased signaling recruitment in PAR2-YFP-expressing HEK-293 cells (net BRET of 0.05 and over) with the exception of AYPGK-1Nal-NH 2 (Fig. S8).

PAR4-stimulated MAPK signaling
HEK-293 and CRISPR/Cas9 ␤-arrestin-1/2 double knockout HEK-293-Agonist-stimulated MAPK signaling in CRISPR/ Cas9 HEK-293 ␤-arr. knockout and WT HEK-293 cells transfected with PAR4-YFP were analyzed by Western blotting. Cells were placed in serum-free media (serum-free DMEM, Gibco) for 3 h before beginning the experiment. Cells were then stimulated with 30 M AYPGKF-NH 2 for varying time periods (0, 2, 5, 10, 20, 30, 60, and 90 min) at 37°C. Treated cells were placed on ice following the stimulation period. Total protein was extracted by adding kinase lysis buffer with phosphatase and protease inhibitors (20 mM Tris-HCl, pH 7.5, 100 mM NaCl, 10 mM MgCl 2 , 1 mM EDTA, 1 mM EGTA, 0.5% Nonidet P-40, 2.5 mM sodium pyrophosphate, 1 mM ␤-glycerophosphate, 1 mM Na 3 VO 4 , 25 mM NaF, 1 mg/ml leupeptin, 1 mg/m aprotinin, 1 mM phenylmethylsulfonyl fluoride, and 1 mM DTT) for 20 min, and cell membranes were cleared by centrifugation (13,300 ϫ g for 3 min). Protein concentrations were measured using the Pierce BCA protein assay (Thermo Fisher Scientific). Protein samples were heat-denatured at 98°C for 8 min in denaturing Laemmli buffer (containing 2-mercaptoethanol) and resolved on 4 -12% gradient Invitrogen Bolt BisTris Plus gels (Thermo Fisher Scientific). The resolved proteins were transferred via wet transfer in buffer (1ϫ Tris-glycine, 20% methanol) to a polyvinylidene difluoride membrane and blocked in TBST buffer (TBS with 0.1% (v/v) Tween 20) supplemented with 1% ECL Advance Blocking Agent (GE Healthcare) for 40 min at room temperature. p44/42 (Thr-202/Tyr-204) phosphorylation was detected with specific antibodies (Cell Signaling Technology, Danvers, MA; diluted 1:1000 in TBST) overnight at 4°C. Phospho-p44/42 immunoreactivity was detected using the horseradish peroxidase-conjugated anti-rabbit secondary antibody (Cell Signaling Technology; 1:10,000 in TBST for 1 h). After washing the membrane with TBST (three times with 10-min washes), the peroxidase activity was detected with the chemiluminescence reagent ECL Advance (GE Healthcare) on iBright image Station (Invitrogen). Membranes were then stripped with blot-stripping buffer (Thermo Fisher Scientific) at room temperature and blocked in TBST with 1% ECL Advance Blocking Agent before incubation with the p44/42 (1:1000 in TBST with 1% ECL Advance Blocking Agent) overnight at 4°C. Following incubation with primary antibody, blots were rinsed in TBST and incubated with appropriate horseradish peroxidase-conjugated anti-rabbit or anti-mouse secondary antibody (1:10,000 in TBST for 1 h). Membranes were washed and imaged as described above. Band intensities representing activated MAPK were quantified using FIJI (68). Phospho-kinase levels were normalized for differences in protein loading by expressing the data as a ratio of the corresponding total p44/42 signal. Data are expressed as a fold increase over unstimulated baseline control.
MAPK signaling in PAR4-YFP stable cells with peptide library-Agonist-stimulated MAPK signaling in HEK-293 cells stably expressing PAR4-YFP were analyzed by Western blotting following stimulation with PAR4 peptide agonists. Cells were placed in serum-free media (serum-free DMEM, Gibco) for 3 h before beginning the experiment. Cells were stimulated with either control (no agonist), AYPGKF-NH 2 , or agonist peptides described at 100 M concentration. Cells were stimulated for 10 min with agonist and lysed, quantified, and blotted as described above.

Preparation of washed platelets and light-transmission aggregometry
Animal care and experimental procedures were carried out under a protocol (AUP no. 2018-047) that was reviewed and approved by the University of Western Ontario Animal Care Committee in accordance with the Canadian Council on Animal Care (CCAC) standards and guidelines. Blood was drawn from the abdominal aorta or heart of Sprague-Dawley rats into trisodium citrate (3.4% w/v). The blood was centrifuged at 200 ϫ g for 15 min at 22°C to isolate platelet-rich plasma (PRP). PRP was transferred to a 14-ml polypropylene round-bottom tube (Falcon) and centrifuged at 1800 rotations per min (594 ϫ g) for 10 min at 22°C to pellet platelets and obtain platelet-poor plasma (PPP). PPP was decanted and used for 100% light transmission control. To obtain the washed platelet sample, platelets were washed twice in Tyrode's buffer (12 mM NaHCO 3 , 127 mM NaCl, 5 mM KCl, 0.5 mM NaH 2 PO 4 , 1 mM MgCl 2 , 5 mM glucose, and 10 mM HEPES; Ramachandran et al. (26)) and resuspended to a volume equivalent to that of the initial blood draw. Washed platelet sample was equilibrated for 1 h at ambient temperature following the addition of 1 M CaCl 2 (1 l/ml). 400 l of this washed platelet suspension was used for light transmission aggregometry on a platelet aggregometer (Model 700; Chrono-Log Corp, Havertown, PA).

Structural homology modeling of PAR4
Residues resolved in the PAR2 crystal (Protein Data Bank (PDB) code 5NDD) were used to generate a homology model of WT human PAR2. Homology model of WT human PAR2 was then aligned with the WT human PAR4 amino acid sequence (National Center for Biotechnology Information (NCBI) accession number NM_003950) using Clustal Omega (37,38). Only residues resolved in the crystal structure of PAR2 were used as template coordinates for PAR4 homology modeling. Twenty human PAR4 homology models were generated using the human PAR2 crystal structure coordinates (PDB code 5NDD) (36) as a template in MODELLER (version 9.16) (69). The model with the lowest DOPE score was taken as the best structure and visualized using PyMOL (version 1.7.4, Schrödinger, PAR4 activation and biased signaling LLC). Electrostatic surface potential was calculated after assignment of partial charges by using PDB2PQR and by using the Adaptive Poisson-Boltzmann Solver plugin in PyMOL (70).

In silico docking of AYPGKF-NH 2 to the PAR4 homology model
In silico docking of the PAR4 agonist peptide, AYPGKF-NH 2 , to the PAR4 homology model was undertaken using Galaxy-PepDock (39). The lowest energy PAR4 homology model PDB file was loaded into GalaxyPepDock along with a text file of the PAR4 agonist peptide sequence, which GalaxyPepDock uses to predict a peptide structure for docking analyses (Ala-Tyr-Pro-Gly-Lys-Phe; amidation is not included in the structure prediction). Ten models of PAR4-AP docking were generated. Frequency of receptor-peptide residue interactions, taken as atom-atom distances of less than 3 Å, was quantified manually to determine highly recurrent interactions.

Data analysis and statistical tests
Data are shown as mean Ϯ S.E. Calcium signaling and ␤-arrestin-1/2 recruitment concentration-effect curves were calculated using nonlinear regression curve fitting (three parameters; Prism7). EC 50 values are reported in Table 1 where the signal has saturated. Where signaling is not saturated, the EC 50 is reported for calcium as not determined (ND). The F-statistic calculated to compare EC 50 values was calculated using the extra sum-of-squares analysis (71). For Table 1, values comparing percentage of signaling or recruitment at 100 M to values obtained from the parental PAR4-AP, AYPGKF-NH 2 , are shown. Significance was determined for comparisons with 100 M AYPGKF-NH 2 and p44/42 phosphorylation by one-way ANOVA. Western blotting protein densitometry was quantified using FIJI. Transformation and analysis were conducted on raw data, and significance is indicated by * (p Ͻ 0.05).