Discovery and Validation of Novel Peptide Agonists for G-protein-coupled Receptors

Abstract

G-protein-coupled receptors (GPCRs) represent an important group of targets for pharmaceutical therapeutics. The completion of the human genome revealed a large number of putative GPCRs. However, the identification of their natural ligands, and especially peptides, suffers from low discovery rates, thus impeding development of therapeutics based on these potential drug targets. We describe the discovery of novel GPCR ligands encrypted in the human proteome. Hundreds of potential peptide ligands were predicted by machine learning algorithms. In vitro screening of selected 33 peptides on a set of 152 GPCRs, including a group of designated orphan receptors, was conducted by intracellular calcium measurements and cAMP assays. The screening revealed eight novel peptides as potential agonists that specifically activated six different receptors in a dose-dependent manner. Most of the peptides showed distinct stimulatory patterns targeted at designated and orphan GPCRs. Further analysis demonstrated a significant in vivo effect for one of the peptides in a mouse inflammation model. G-protein-coupled receptors (GPCRs) represent an important group of targets for pharmaceutical therapeutics. The completion of the human genome revealed a large number of putative GPCRs. However, the identification of their natural ligands, and especially peptides, suffers from low discovery rates, thus impeding development of therapeutics based on these potential drug targets. We describe the discovery of novel GPCR ligands encrypted in the human proteome. Hundreds of potential peptide ligands were predicted by machine learning algorithms. In vitro screening of selected 33 peptides on a set of 152 GPCRs, including a group of designated orphan receptors, was conducted by intracellular calcium measurements and cAMP assays. The screening revealed eight novel peptides as potential agonists that specifically activated six different receptors in a dose-dependent manner. Most of the peptides showed distinct stimulatory patterns targeted at designated and orphan GPCRs. Further analysis demonstrated a significant in vivo effect for one of the peptides in a mouse inflammation model. G-protein-coupled receptors (GPCRs) 2The abbreviations used are: GPCR, G-protein-coupled receptor; PBS, phosphate-buffered saline; PMN, polymorphonuclear leukocyte; Ang, angiotensin. represent a large group of receptors that are directly involved in cellular signaling networks and are considered to be an important family of targets for pharmacological intervention (reviewed in Ref. 1.Imming P. Sinning C. Meyer A. Nat. Rev. Drug Discov. 2006; 5: 821-834Crossref PubMed Scopus (656) Google Scholar). It is estimated that approximately one-third of therapeutic development in the pharmaceutical industry is targeted toward drug discovery in the area of GPCRs (2.Hill S.J. Br. J. Pharmacol. 2006; 147: 27-37Crossref PubMed Scopus (141) Google Scholar). Recent analysis of the human genome identified over 800 putative members of the GPCR superfamily, over half of which are thought to be involved in sensory mechanisms and are not considered to be drug targets. Among the known GPCRs, ∼300 are considered potential pharmaceutical targets (3.Lin S.H. Civelli O. Ann. Med. 2004; 36: 204-214Crossref PubMed Scopus (42) Google Scholar). Out of this group around 150 GPCRs have no designated natural/endogenous ligand or activator. These receptors are termed “orphan GPCRs” and have recently become very attractive to many industrial and academic researchers as they hold great potential as targets for novel drug discovery (4.Leifert W.R. Aloia A.L. Bucco O. Glatz R.V. McMurchie E.J. J. Biomol. Screen. 2005; 10: 765-779Abstract Full Text PDF PubMed Scopus (36) Google Scholar). Lately, a major effort was made by many academic institutes as well as biotech and pharmaceutical companies to identify both natural and surrogate ligands for both known and orphan GPCRs (5.Kutzleb C. Busmann A. Wendland M. Maronde E. Curr. Protein Pept. Sci. 2005; 6: 265-278Crossref PubMed Scopus (19) Google Scholar, 6.Jiang Z. Zhou Y. Curr. Protein Pept. Sci. 2006; 7: 459-464Crossref PubMed Scopus (6) Google Scholar). The most common method to screen for novel GPCR activators is through large scale screening of hundreds and thousands of compounds on many GPCRs, testing for in vitro activation of any given receptor by assessing its signaling response of an expressing cell to binding of a specific compound or tissue extract. This method, called “reverse pharmacology” (7.Howard A.D. McAllister G. Feighner S.D. Liu Q. Nargund R.P. Van der Ploeg L.H. Patchett A.A. Trends Pharmacol. Sci. 2001; 22: 132-140Abstract Full Text Full Text PDF PubMed Scopus (240) Google Scholar, 8.Civelli O. Trends Pharmacol. Sci. 2005; 26: 15-19Abstract Full Text Full Text PDF PubMed Scopus (106) Google Scholar), gave rise in the past decade to over 30 new receptor/ligand pairs and many previously known or novel ligands, most of them peptides (5.Kutzleb C. Busmann A. Wendland M. Maronde E. Curr. Protein Pept. Sci. 2005; 6: 265-278Crossref PubMed Scopus (19) Google Scholar). However, this method is very expensive and time-consuming and might lead to many potential ligands either being missed or falsely annotated (6.Jiang Z. Zhou Y. Curr. Protein Pept. Sci. 2006; 7: 459-464Crossref PubMed Scopus (6) Google Scholar, 9.Eglen R.M. Comb. Chem. High Throughput Screen. 2005; 8: 311-318Crossref PubMed Scopus (50) Google Scholar). To increase the feasibility of this type of testing, it is necessary to pre-screen any potential group of ligands and focus on high chance, low risk candidates. This might allow more thorough screening and validation of true novel activators with better chances of becoming therapies. Computational screening can supplement experimental efforts as a preliminary screening tool for efficient identification of natural ligands, especially peptides (10.Bock J.R. Gough D.A. J. Chem. Inf. Model. 2005; 45: 1402-1414Crossref PubMed Scopus (90) Google Scholar). Bioinformatics can serve as a powerful tool that provides reliable predictive measures to select for the high potential candidates and provide a spotlight pointed at potential new candidates for experimental discovery, thereby enabling higher success rates in identification of novel GPCR ligands (11.Huang E.S. Drug Discov. Today. 2005; 10: 69-73Crossref PubMed Scopus (12) Google Scholar). In this study, we describe the experimental validation of computationally discovered novel GPCR-activating peptides. A subgroup of 33 potential candidate peptides was experimentally screened. Here we report the in vitro validation of eight novel GPCR peptide activators (hit rate of ∼25%) as potential candidates for related clinical utilities. One of the examined peptides was tested in vivo and was found effective in a mouse inflammation model. Data Set Preparation for the Proteolytic Site Predictor—All mammalian proteins (28,780) were downloaded from the Swiss Protein Database (release 43) (12.Bairoch A. Apweiler R. Wu C.H. Barker W.C. Boeckmann B. Ferro S. Gasteiger E. Huang H. Lopez R. Magrane M. Martin M.J. Natale D.A. O'Donovan C. Redaschi N. Yeh L.S. Nucleic Acids Res. 2005; 33: D154-D159Crossref PubMed Scopus (1446) Google Scholar). Of these, 11,705 proteins were classified as secreted or membrane proteins according to Swiss-Prot annotation. From these 11,705 proteins, 553 experimentally validated 18-mer peptide convertase proteolytic sites were extracted based on Swiss-Prot FT annotation lines. This subset of sequences was used as a learning set for Breiman's random forest classifier algorithm (13.Breiman L. Machine Learning. 2001; 45: 5-32Crossref Scopus (74749) Google Scholar), whereas the negative set (non proteolytic sites) consisted of all extracellular 18-mers, excluding the boundaries of pro-peptides, chains, and peptides, according to Swiss-Prot annotations. Creation of a Predicted Human Peptidome Dataset—A predicted human secretome was generated by analyzing human proteins from the following two sources: (i) the Uniprot/Swiss Protein Database (version 47) and (ii) Refseq sequences from the NCBI nr protein data base (GenBank™ version 149). There were 29,530 proteins from NCBI nr, including proteins with various validation annotations, reviewed, predicted, and modeled. The annotation for Uniprot proteins was used to filter out proteins that do not have a signal peptide, validated or predicted. For NCBI nr proteins, the SignalP 3.0 software (14.Bendtsen J.D. Nielsen H. von Heijne G. Brunk S. J. Mol. Biol. 2004; 340: 783-795Crossref PubMed Scopus (5683) Google Scholar) was used to predict the existence and location of a signal peptide. For each protein precursor sequence, the following procedure was used to create the potential peptide products of the precursor. (The full cleavage prediction concept is detailed in Kliger et al. (15.Kliger Y. Gofer E. Wool A. Toporik A. Apatoff A. Olshansky M. Bioinformatics (Oxf.). 2008; 24: 1049-1055Crossref PubMed Scopus (13) Google Scholar).) The predicted signal peptide was removed, and potential cleavage sites were scored based on the outcome of the machine learning algorithm used for cleavage site prediction. Based on precision/recall analysis, each putative cleavage site was assigned a score value between 0 and 10, where 0 represents a known cleavage site (as annotated on Uniprot) and 10 corresponds to unlikely cleavage sites. Sites with a score of 4 or less, and sites that contained two consecutive basic amino acids (e.g. Arg or Lys), regardless of their score, were selected. Peptides whose end points were either a selected cleavage site or an end point of the precursor (signal peptide cleavage site at the N terminus or the C terminus of the precursor) were created. Peptides longer than 100 amino acids were discarded. It should be noted that different peptides from the same precursor may have an overlapping sequence, and one can be a sub-sequence of another. The peptide set was extended by sequentially removing C-terminal basic amino acids and removing an additional C-terminal glycine, leaving an amide attached to the C terminus of the peptide (this mimics a known natural peptide maturation process called “glycine-directed peptide amidation”). All the intermediate peptides resulting from this extension were considered as separate peptide products. The complete list of predicted peptides (the “peptidome”) contained hundreds of thousands of peptides, which were then analyzed as a source data by the GPCR-activating ligand predictor. GPCR Ligand Predictor—The set of known GPCR peptide ligands was extracted from Uniprot, using the Uniprot annotation (SRS from Swiss-Prot and by relevant keywords from GPCRDB (16.Horn F. Bettler E. Oliveira L. Campagne F. Cohn F.E. Vriend G. Nucleic Acids Res. 2003; 31: 294-297Crossref PubMed Scopus (330) Google Scholar)) and was manually curated based on literature search. The sequence lengths and positions of the peptides were extracted from the Swiss-Prot feature table. The learning data set contained 64 precursors and 123 peptides. Of these, 87 did not overlap with another peptide of the same precursor. Several known GPCR ligands were removed from the learning set (for example, the chemokine family and the hormone proteins that are larger than 100 amino acids) giving an average peptide size of 26 amino acids. The learning process also required a negative set of peptides that are not known as GPCR peptide ligands. Two negative sets of peptides were constructed. The first set was extracted based on Uniprot using annotations to filter peptides that are not GPCR peptide ligands. The second set was a collection of 3000 randomly chosen peptides from the novel peptide data base (peptidome) described above. Using the generated learning set, a classifier based on the “Random Forest” algorithm (13.Breiman L. Machine Learning. 2001; 45: 5-32Crossref Scopus (74749) Google Scholar) was implemented to select peptides that are likely to be GPCR ligands. Following a thorough optimization process, the classifier was programmed to analyze the following index parameters. The length of the peptide (in amino acids). The frequency of each amino acid in the peptide relative to the peptide length (21 parameters, one for each amino acid and an extra parameter for those that are not conventional amino acids). The amino acids in proximity to the N-terminal cleavage site from both sides as follows: the first four amino acids in the N terminus of the peptide and the last four amino acids of the precursor preceding the cleavage site (eight parameters, and each can be one of 21 values, one for each amino acid and an extra value for other letters or for amino acids beyond the signal peptide cleavage site). This parameter (four amino acids) was found most effective in the optimization process. The amino acids in proximity to the C-terminal cleavage site from both sides are as follows: the last three amino acids in the C terminus of the peptide and the first three amino acids of the precursor following the cleavage site (six parameters, each can be one of 21 values, one for each amino acid and an extra value for other letters or for amino acids beyond the precursor end). This parameter (3 amino acids) was found most effective in the optimization process. The last is the number of in the for of the precursor. Based on these the was to between the learning set of known ligands and the negative This was then used to each predicted peptide a score the of being a GPCR-activating the end of the process, a of 100 peptides with the classifier was created. These peptides a analysis process and literature by a different The of GPCR-activating peptides was based on the following and tissue of the with to the receptor; of the cleavage and GPCR classifier of the candidate peptides and their mouse of the sequence and cleavage sites of the peptide in the precursor proteins of all of the peptide the precursor with to known and (as a negative and the number of and annotations. The peptides were also for their based the with sequences in data with very parameters. This process in a subset of novel peptides most likely to be GPCR ligands. each peptide, from the subset of peptides, a number of were made to and In peptides a the was by either a or a on the and amino acid of the peptide. For peptides a at the C the was by an Peptides were by (The and by high to by from one peptide that and one peptide, which was all 33 peptides were in or in G-protein-coupled list of receptors was generated by all human (16.Horn F. Bettler E. Oliveira L. Campagne F. Cohn F.E. Vriend G. Nucleic Acids Res. 2003; 31: 294-297Crossref PubMed Scopus (330) Google Scholar)) receptors according to their clinical All known and receptors were GPCRs known to be activated by ligands and as were removed from the list with receptors for proteins over amino acids) as The list of receptors contained GPCRs with receptors annotated as with a predicted high of being activated by a peptide sequence and to GPCRs activated by peptides and literature receptor was also for its most known ligand as a for the screening GPCR for by the Predicted experimental screening was by on all the selected GPCRs by the to signaling to the thus enabling of GPCR activation by testing for as described J. Biomol. Screen. 2003; 8: Full Text PDF PubMed Scopus Google Scholar). Peptides were in All were at of the GPCRs were in one of the following or were using as were on the preceding were with a GPCR and using a according to the were for then and the of the were with according to the were a and was following of were with the at of contained each of the examined peptides in We a as a peptide that a and distinct increase in intracellular calcium that is and significant a the of calcium and peptide with value than of the calcium for at two GPCR for with different GPCRs the list of receptors as described were of a were with for 10 at then with either a or a candidate peptide (for by or a with 10 (for Following a with of either or the tested peptides, either with or intracellular cAMP was using the cAMP according to the Data were to of cAMP by a cAMP of GPCR and were as described above. peptide was in according to selected or in of 10, and and and All peptides were examined and with of a selected (as described in were by of using version 4 The for the was as peptides that were found to of was by for the validation was by an was by an was by an protein was by for the validation was by an protein was by for the validation was by an protein was by for the validation was by for the validation protein was by an protein was by an was by for the validation C-terminal was by an was by for the validation in a new In of and Swiss were from All were for to to allow to on the of the were by of to the of the and the peptides were with number Peptides or were at a of at the of of A were with of and of and on was by an of the with as were analyzed by For of in in were analyzed using and were by analysis of A value was as of in Predicted Human GPCR-activating list of the 100 peptides as potential GPCR ligands were manually examined and according to additional peptide 33 peptides were and 152 GPCRs. was out on all of GPCRs by the to signaling to the thus enabling of and for the GPCR activation by testing for J. Biomol. Screen. 2003; 8: Full Text PDF PubMed Scopus Google Scholar). The of this method in either cAMP signaling was estimated at for Of the 152 receptors were orphan receptors to the and GPCRs known ligands Of the receptors that were showed a response to a known Out of the 152 receptors by the activation four receptors and showed distinct activation with at one of the novel peptides at a of peptides and that showed either a or a for either receptor were examined by response on the specific GPCRs they analysis and of the were with the most for the receptor as described in the and the are in and In all we were to a dose-dependent activation of with to the used activation and by were in the activation by peptides was by is from the activated by same and were examined by cAMP response was were examined by cAMP response was were in the activation by peptides was by is from the Data were examined by cAMP response was in a new We activation of with peptide and to the we conducted in a dose-dependent with estimated the than that of the used E.S. M. S. S. Br. J. Pharmacol. 2006; 147: PubMed Scopus Google Scholar). In experimental that and in their activation of receptors M. S. Sci. 2004; PubMed Scopus Google Scholar). the is specific to and and it did not any of the known receptors and that were in the For we identified potential peptides to the of two of the novel peptides, and was to that of the E.S. M. S. S. Br. J. Pharmacol. 2006; 147: PubMed Scopus Google Scholar) However, of the peptides any effect the receptors of the family that are known to be activated by The receptor is involved in of inflammation by as and P. Zhou Y. J. Res. 2005; 31: PubMed Scopus Google Scholar, M. F. Drug 2006; 5: PubMed Scopus Google Scholar). We identified peptides that activated this and and The peptide in higher and higher than which was used as a it is not the most for this receptor J. 2003; PubMed Scopus Google Scholar) as well as of the other peptides did not another family However, a J. A. L. F. C. C. R. A. M. L. H. G. A. G. M. J. and Y. in a dose-dependent activation of a family not We also identified two peptides, and that activated the receptor and In with we were to calcium by the a known ligand for C. R.P. S. M. M.J. R. Sci. S. A. 2003; PubMed Scopus Google Scholar). However, peptides did the receptor at to that found for A of peptides were specific to and did not or receptors which are known to be activated by the not a signal to a subset of GPCRs with very low activation by either the or any of the novel examined peptides was selected to be tested for activation by of the 33 examined peptides. These receptors were for cAMP or as described Two of the peptides and showed distinct activation of two related GPCRs and and of these receptors were recently identified as being activated by and Mol. Pharmacol. 2006; PubMed Scopus Google Scholar). A on an orphan family not was also was examined by a receptor for which was not activated by peptide In of the by activation of an and process, of which to a effect M. F. Drug 2006; 5: PubMed Scopus Google Scholar). To the in vivo of novel peptide and two of its and we used a of the The novel peptide as well as and were tested at of and mouse and were given at the of A These to and of and were used for the peptides. the was at the using both (for and cell analysis for (for of A significant of was in the of the with at Out of the two a of cell at the tested a effect at the that was not were on not In this we describe the of a screen for novel peptide GPCR A subset of the predicted candidate peptides was on a large group of GPCRs, and eight peptide agonists were identified for six different GPCRs. The peptides was validated experimentally using as dose-dependent of calcium and by assessing the in vivo in one and machine both related to all potential secreted proteins (the and all potential cleavage we were to create a that hundreds of thousands of potential peptides, and by of the predicted were analyzed for and assigned a score of being GPCR ligands by a machine learning This collection contained hundreds of candidate peptides that were scored and by distinct and The candidates were for and for screening for activation on a group of GPCRs. The screening 33 peptides that were for activation of 152 GPCRs selected according to and Out of the 33 peptides showed a distinct activation of at one The of candidates is with the average of novel GPCR peptide ligand discovery which is on average around two to three new candidates in the past decade S. A. Trends Pharmacol. Sci. 2003; 24: Full Text Full Text PDF PubMed Scopus Google Scholar). All of these peptides are and most of them are from protein Most of the peptides that showed also one of the peptides, designated peptide which showed a dose-dependent activation of the was examined for in vivo activation of the receptor using a of The activation of as by of was found to be out of two peptides from and at one was found to an in vivo response by the to the peptide. of the activation of the receptors by the novel peptides of the same peptides on different GPCRs. This not of the receptors, as the two GPCRs, and to the same whereas the and to the same to Ref. P. A. Biol. Google Scholar). no sequence was found between the discovered peptides and the known GPCR ligands, we predict that other as might a between the novel and known peptides. A preliminary prediction of the eight novel peptides by using the Bioinformatics (Oxf.). PubMed Scopus Google Scholar) revealed that all of them at one a feature common in many known peptide GPCR ligands not We that we were to and a of concept for discovery of GPCR peptide We that we be to analysis and additional novel peptide GPCR ligands by this We predict that more of the discovered and validated activators found for these GPCRs be for their potential as therapeutic compounds targeted at receptors involved in distinct and