| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
,1,1
* Department of Genitourinary Medical Oncology,
Department of Cancer Biology, and
Department of Biostatistics, The University of Texas M. D. Anderson Cancer, Houston, Texas, USA
1Correspondence: Wadih Arap and Renata Pasqualini, 1515 Holcombe, Blvd, Houston, TX 77030-4095, USA. E-mail: warap@mdanderson.org; rpasqual{at}mdanderson.org
ABSTRACT
In vivo phage display is a technology used to reveal organ-specific vascular ligand-receptor systems in animal models and, recently, in patients, and to validate them as potential therapy targets. Here, we devised an efficient approach to simultaneously screen phage display libraries for peptides homing to any number of tissues without the need for an individual subject for each target tissue. We tested this approach in mice by selecting homing peptides for six different organs in a single screen and prioritizing them by using software compiled for statistical validation of peptide biodistribution specificity. We identified a number of motif-containing biological candidates for ligands binding to organ-selective receptors based on similarity of the selected peptide motifs to mouse proteins. To demonstrate that this methodology can lead to targetable ligand-receptor systems, we validated one of the pancreas-homing peptides as a mimic peptide of natural prolactin receptor ligands. This new comprehensive strategy for screening phage libraries in vivo provides an advantage over the conventional approach because multiple organs internally control for organ selectivity of each other in the successive rounds of selection. It may prove particularly relevant for patient studies, allowing efficient high-throughput selection of targeting ligands for multiple organs in a single screen.—Kolonin, M. G., Sun, J., Do, K.-A., Vidal, C. I., Ji, Y., Baggerly, K. A., Pasqualini, R., Arap, W. Synchronous selection of homing peptides for multiple tissues by in vivo phage display.
Key Words: vascular • receptor • ligand • combinatorial peptide library • phage display
SELECTIVITY OF TISSUE expression for surface molecules of cells lining the vasculature has been uncovered in disease and during normal development (1
2
3)
. Systematic profiling of selectively expressed "vascular addresses" has revealed prospective molecular targets that may be used to direct therapies to specific tissues (1
, 4)
. Vascular mapping by in vivo phage display is based on the preferential ability of short peptide ligands from combinatorial libraries (displayed on the pIII protein of an M13-derived phage vector) (5)
to home to a specific organ after systemic administration (1
, 2
, 4
, 6)
. Peptides targeting several tissues and disease states isolated in mouse models (1
, 2
, 7
8
9)
and patients (10
, 11)
have led to the identification of the corresponding vascular receptors (7
8
9
10
11
12)
.
Because a single biopanning round of a large (
109 unique combinatorial sequences) phage library may not always sufficiently enrich for organ-homing peptides, until now, phage display screens have encompassed recovery of phage from the organ of interest in three to four rounds of library selection, utilizing one subject per each round (2
, 4
, 6)
. Thus, systematic characterization of the vasculature by using this "conventional" in vivo phage display screen setup has been rate-limited by the need to separately perform individual multiround screens for motifs homing to each tissue studied. To screen phage libraries in patients more efficiently, we have initiated attempts to isolate tissue-homing peptides for several tissues in a single screen (10)
. The proof-of-principle screen involved only one round of selection and, therefore, required statistical analysis of very large numbers of phage clones (thousands, as opposed to hundreds in a conventional screen) in order to identify homing peptide motifs. Although this effort yielded a prostate vasculature-selective ligand-receptor pair (10
, 12
, 13)
, for future systematic vascular mapping, there are practical limitations to the number of phage clones that can be routinely processed, thus requiring further optimization of the methodology.
To meet such challenge of the setting where maximal information recovery is critical, here we integrated a comprehensive strategy to use phage display for synchronous identification of organ-homing peptides for multiple tissues in a single screen. This strategy allows us to analyze an order of magnitude fewer peptide sequences than those surveyed for the initial single-round multi-organ screen (10)
. We evaluated our approach in the mouse model by performing a screen to isolate peptides independently segregating to six different organs and developed a statistical analysis platform for identification of motifs selectively enriched in targeted tissues. Finally, to demonstrate the efficiency of this approach, we biochemically validated one of the identified pancreas-homing peptides as a mimic of peptide hormones that bind to prolactin receptor (PRLR).
MATERIALS AND METHODS
Synchronous screening of phage libraries in vivo
C57Bl/6 female mice were injected intravenously (i.v.) with 1010 transducing units (TU) of previously described (6)
library CX7C (round 1) or a mixture (109 TU per organ) of amplified phage recovered from each of the organs studied (rounds 2 and 3). For each round, phages were allowed to circulate for 15 min prior to organ recovery (without heart perfusion). After each round of selection, phage peptide-coding inserts were sequenced as described (6)
, amplified for each organ individually, and subsequently pooled for the next round of in vivo selection.
Statistical analysis of selected peptide motifs
Calculation of tripeptide motif frequencies in CX7C peptides encountered in each target tissue (in both directions) was done by using a character pattern recognition program based on SAS (version 8.1.2, SAS Institute) and Perl (version 5.8.1), as described (10)
. To identify tripeptides progressively enriched from round 1 to round 3 of panning, a Bayesian Beta/Binomial model was implemented by estimation of the posterior probability distribution for each tripeptide (14
, 15)
; posterior distributions for the proportion of each tripeptide in rounds 1 through 3 were calculated by using Splus (version 6). To determine the selectivity of tripeptide motif distribution in tissues, we used Fisher’s exact test (one-tailed) and calculated the P values for the count of each tripeptide in a target tissue, as compared with its count within the 2210 tripeptides of the unselected library (Table 1
, middle column) or the combined tripeptides from the other five tissues (Table 1
, right column). Statistical uncertainty was further assessed by Monte Carlo simulations based on an established algorithm (16
, 17)
. Using MATLAB, all tripeptide sequences (unselected library and selected for each organ) were pooled, and the combined tripeptide pool was distributed in 1000 simulations into permutated groups corresponding in size to those analyzed in Table 1
for each organ and the pre-selection library. For each round, Fisher’s exact test was performed on the 1000 scrambled datasets, and distributions of the 50 lowest P values generated in each simulated test were compared with the distribution of 50 lowest P values from the actual experimental data (the most significant of which are shown in Table 1
, middle column).
|
High-throughput identification of peptide-mimicked proteins
To facilitate large-scale peptide sequence analysis, we constructed an interactive peptide sequence management database based on MySQL. We developed a web-based peptide sequence retrieval and management software based on Common Gateway Interface (CGI) and Perl, and integrated it with the statistical analysis software. To identify candidate cellular proteins mimicked by selected peptides, we consolidated our database with on-line ClustlW software (http://www.ebi.ac.uk/clustalw/) to identify extended (4–7 residue long) motifs shared among multiple peptides homing to a specific tissue. Basic local alignment search tool (BLAST) (http://www.ncbi.nlm.nih.gov/basic local alignment search tool/) was used to identify proteins mimicked by the extended homing motifs by screening batches of ClustlW-identified peptide motifs against sequences contained in on-line nonredundant databases of mouse proteins. To identify PRLR ligand-matching motifs among phage-displayed pancreas-homing peptides binding to PRLR, a software was codified in Perl 5.8.1 and run against ClustlW-aligned protein sequences for sheep PL (oPL), and mouse mPL-I and mPRL. Each seven-mer peptide sequence was aligned in each orientation against the protein sequences from N- to C-terminus in one-residue shifts. The peptide-protein similarity scores for each residue were calculated based on a BLOSUM62 matrix modified to identify peptide matches of four or more residues in any position being identical to the corresponding amino acid positions in any of the three PRL homologues aligned.
Phage-peptide binding to PRLR
To identify peptides binding to PRLR, phage clones isolated from the pancreas in rounds 2 and 3 of the screening were individually amplified and pooled. For panning on immobilized PRLR, 109 TU of the mixed phage clones were incubated overnight at 4°C with 10 µg of purified recombinant rabbit PRLR (Protein Laboratories Rehovot, Israel), or BSA control, immobilized on plastic. Unbound phage were extensively washed off with PBS, and then the bound phage were recovered by infecting host K91 E. coli directly on the plate. For panning on cell surface-expressed PRLR, COS-1 cells from American Type Culture Collection (ATCC, Manassas, VA) were transiently transfected with pECE-PRLR as described (18)
and subjected to biopanning with the amplified pancreas-isolated phage using the BRASIL protocol (19)
. In each biopanning, bound phage were selected for tetracycline resistance, quantified by infecting host K91 E. coli and sequenced. Single amino acid substitutions and reversal of the PRL-matching motif displayed on phage was performed by polymerase chain reaction (PCR)-directed mutagenesis of the peptide-coding insert (TGTCGCGTGGCGAGCGTGCTGCCGTGT) and cloning it into the fUSE5 vector (5)
. Phage displaying forward (CRVASVLPC), reverse CPLVSAVRC, and the alanine point mutants were titered inparallel with insertless fd-tet phage and tested for binding to COS-1 cells transfected with PRLR by using the BRASIL method (19)
. COS-1 cells not transfected with PRLR served as a negative control.
Immunolocalization
Immunofluorescent detection of PRLR expression in COS-1 cells was performed by using anti-PRLR antibody (Ab) MA1–610 (Affinity Bioreagents) diluted to 20 µg/ml and a secondary FITC-conjugated goat anti-mouse Ab F-0257 (Sigma) at 1:100 dilution. For validating phage binding and internalization into COS-1 cells transfected with PRLR, 109 TU of phage displaying PRLR-binding or mutant peptides were subjected to cell binding and internalization as described (12)
. Immunodetection of cell-associated phage was performed with anti-fd Ab B-7786 (Sigma) at 1:500 dilution and a secondary Cy3-conjugated donkey anti-rabbit Ab 711–165-152 (Jackson)at 3 µg/ml. For phage-peptide immunolocalization in situ, 1010 TU of iv-injected phage were let circulate for 5 min. Immunohistochemistry on formalin-fixed, paraffin-embedded mouse tissue sections was performed as described (6
, 10)
by using anti-fd Ab B-7786 at 1:1,000 dilution and the LSAB+ peroxidase kit (DAKO).
RESULTS
Synchronous phage library screening in vivo
We reasoned that peptide-displaying phage clones of systemically administered library would segregate in the bloodstream irrespective of each other and, thus, target individual organs independently. If this hypothesis is correct, independent enrichment of phage-peptides targeting any number of organs should take place on successive rounds of selection. To establish the experimental framework for synchronous selection of peptides independently homing to different organs, we screened a phage-displayed cyclic random peptide library CX7C (C, cysteine; X, any residue). Six organs were targeted in mice: muscle, bowel, uterus, kidney, pancreas, and brain (Fig. 1
). Three rounds of library selection were performed based on previously described methodology (6)
, but without the step of whole-body perfusion of the vasculature, which was skipped in order to simulate screening conditions used in patients (10)
. In each round, peptide-displaying phage were isolated from target organs, amplified, and pooled for the next round of selection. To identify organ-homing motifs, we sequenced the DNA corresponding to peptide inserts from 96 recovered phage clones after each of the three rounds of selection for each of the six organs and analyzed the sequences. Preferential cell binding of specifically homing peptides to differentially-expressed receptors results in enrichment, defined by the increased frequency of the peptide recovery in each subsequent round of the screen (4
, 6)
. Thus, we set out to profile the differential distribution of library-encoded peptides among the six organs, which could be used for the subsequent identification of their targets.
|
To analyze the spectrum of the peptides resulting from the screening and to compare those among different organs, we adopted a combinatorial statistical approach based on the premise that three residue motifs (tripeptides) provide a sufficient structure for peptide-protein interactions in the context of phage display (10)
. Since a tripeptide within a CX7C sequence may be responsible for receptor targeting in either orientation, we performed a computer-assisted analysis of the 7489 tripeptides contained in each direction within the CX7C inserts (2620 from the first round; 2554 from the second round; and 2315 from the third round) successfully sequenced for the organ-recovered clones. First, we evaluated the increase in recovery frequency for individual tripeptides in the three consecutive rounds of selection by using the Bayesian Beta/Binomial model (14
, 15)
. For each organ, we found a number of tripeptides progressively enriched (Table 1
, left column), thus, suggesting their superior affinity and/or specificity. Next, we surveyed the 2315 motifs recovered from the third round of selection to identify motifs with terminal frequencies higher than those present in the phage library prior to selection. The significance of motif representation increase on selection was assessed by the Fisher exact test (Table 1
, middle column). To show that the P value of 0.05 for establishment of selected tripeptides was chosen appropriately, we adapted the Monte Carlo algorithm (16)
, and confirmed in the third selection round that the P values of the actual data were smaller than at least 95% of the simulated P values (Fig. 2
). Of note, Monte Carlo simulations showed a progressive accumulation of tripeptides isolated with lower P values from the first to the third round (Fig. 2)
, consistent with enrichment of the corresponding motifs identified by the Bayesian Beta/Binomial model (Table 1
, left column). Finally, we used the Fisher exact test to analyze the motifs recovered from the third selection round for specificity of tissue homing by identifying tripeptides that were enriched in one of the six organs, but not in the rest of the organs studied (Table 1
, right column).
|
Identification of candidate biological ligands mimicked by homing peptides
The majority of peptide motifs identified by statistical analysis enriched during the screen also showed specificity of association with the organ from which they were recovered (Table 1)
. For validation of potentially specific organ-homing motifs, we chose to focus on tripeptides that fulfilled the criteria of the statistical tests applied (Table 1)
. Since peptide motifs binding to cell surface receptors have been previously found to mimic native ligands for these receptors (7
, 10
, 19)
, the ClustalW software was used to determine whether the tripeptides represented parts of longer motifs responsible for organ homing, which would facilitate peptide/protein similarity search. For some of the tripeptides, this analysis identified extended (four to seven residue) motifs shared among multiple CX7C peptides isolated from the given organ (Fig. 3
). Each of these extended motifs was screened by using the BLAST algorithm against a nonredundant database of mouse proteins to identify regions of similarity within proteins potentially mimicked by the motifs (Table 2
).
|
|
We systematically analyzed the BLAST output for selected motif similarities to extracellular signaling factors that had been reported to regulate organ-dependent vascular growth or homeostasis. This revealed 19 motifs as segments of such proteins and, in some cases, identified several motifs that homed to the same organ and that matched different domains within the same protein (Table 2)
. For example, muscle-homing motifs GRSG+R and SGASAV, matched to two different domains of the Jagged2-like protein (Table 2)
that belongs to a family of ligands for Notch receptors known to regulate vascular development and function (20
, 21)
. Similarly, pancreas-homing motifs ASVL (in the reverse orientation) and WSGL showed close similarity to different domains of a placental lactogen (22)
. Interestingly, for muscle and pancreas, we also matched homing tripeptides to different ligands that share a receptor with a functional role in vascular biology in the target organ (Table 2)
. Among skeletal muscle-homing motifs, tripeptides FSG and SGI were partially overlapping in the extended DFSGIA+ region of similarity to disintegrin family metalloproteinases ADAM and Spi 12, respectively, which cleave Notch receptors (23)
. In the pancreas, motif SWSG matched to prolactin (PRL)-like protein M (PLP-M), which belongs to the same family as the placental lactogen PL-I (containing the reverse ASVL and WSGL) and also binds to the PRL receptor (22
,24)
.
Biochemical validation of PRLR as the target for pancreas-homing peptides
To demonstrate the possibility of efficient characterization of circulation-accessible receptors by synchronous biopanning, as a proof-of-principle, we chose to validate the PRL receptor (PRLR) as a peptide target in the pancreas. PL-I and PLP-M identified by BLAST analysis belong to the conserved family of PRL-like peptidic hormones that have been shown to function in the pancreas during pregnancy (22
, 25)
. Because PRLR is the only known receptor for these proteins (22
, 24)
, we proposed that the ASVL, WSGL, and SWSG motifs target PRLR in vivo by mimicking PRL family hormones.
To test this hypothesis, we attempted to reveal a biochemical interaction of pancreas-homing motifs with PRLR. We used the BRASIL (biopanning and rapid analysis of selective interactive ligands) method (19)
to screen a pancreas-homing phage sublibrary (pooled clones recovered in rounds 2 and 3) against PRLR expressed on the surface of COS-1 cells (18)
. In parallel, we screened the same sublibrary on purified recombinant PRLR (26)
. A single round of each selection for PRLR-binding phage-peptides resulted in over 90% of the clones sequenced comprised by seven different peptides, five of which were enriched on both immobilized and cell surface-expressed PRLR (Fig. 4
A). Phage displaying these peptides had specific affinity to PRLR, as determined by subjecting the same sublibrary to binding of an immobilized BSA control (Fig. 4B
). Remarkably, computer-assisted analysis of sequences revealed that all of the selected peptides contained amino acid motifs similar to those present in proteins of PRL family (Fig. 4A
). Furthermore, a clear cluster of matches was identified around one of the hormone domains that had been shown (27)
to mediate receptor interaction (Fig. 4A
). As a negative control, the same similarity search algorithm did not reveal such matches for the selected sequences to unrelated proteins such as insulin, IL-11 and bZIP (data not shown).
|
Peptide motif CRVASVLPC recovered as a PRLR-binding prolactin mimic (Fig. 4)
contained a tripeptide, SVL, also identified as one of those enriched in the pancreas during the screen (Table 1)
. The CRVASVLPC motif matched one of the PL-I sites involved in PRLR interaction (27)
, as it had amino acids identical to those found in one or more of the three aligned PRL homologues in four out of seven positions (Fig. 4A
). Similarity of this peptide in reverse orientation to a part of PL-I (Fig. 4A
), initially identified by the BLAST analysis (Table 2)
, was found by RasMol-assisted analysis of 3D protein structure to reside within the domain exposed on the surface of PRL family proteins (data not shown). To demonstrate direct physical interaction between CRVASVLPC and PRLR, we tested binding of CRVASVLPC-phage to COS-1 cells transfected with PRLR and found it to be 9-fold higher than its nonspecific binding to nontransfected COS-1 cells that served as a negative control (Fig. 5
A). To address the issue of a possible importance of the motif orientation for PRLR binding, we constructed phage displaying CPLVSAVRC, the PRL-mimicking motif in the opposite orientation. Reversal of the motif did not significantly decrease its binding to PRLR on the expressing cells (Fig. 5B
). However, disruption of the motif by alanine-scanning mutagenesis of any amino acid significantly decreased binding to PRLR-expressing cells (Fig. 5A
), thus indicating cooperation of the RVASVLP residues comprising the motif in the receptor recognition. To further demonstrate the specific affinity of the CRVASVLPC motif for its receptor, we showed that the PRL mimic specifically bound to cells expressing PRLR by using immunofluorescence (Fig. 5C
). Phage displaying either CRVASVLPC or CPLVSAVRC were found bound and internalized specifically by cells expressing PRLR, but not by nonexpressing control cells, whereas none of the CRVASVLPC mutants displayed detectable PRLR-expressing cell binding and internalization (Figs. 5C-D
and data not shown).
|
Finally, since the SVL tripeptide found within the PRL mimetope CRVASVLPC was isolated from the pancreas, we evaluated weather the motif homes to PRLR in the pancreatic blood vessels. We showed that the previously reported pancreatic expression of mouse PRLR protein in the vasculature and in the islets (28)
closely resembles the in vivo distribution of phage displaying the CRVASVLPC motif (Fig. 5E-H
). Immunohistochemistry of mouse tissues on i.v. CRVASVLPC-phage administration revealed localization of the CRVASVLPC motif to pancreatic blood vessels and the pancreatic islet cells (Fig. 5E
), but not to skeletal muscle (Fig. 5F
). In contrast, a control in vivo–administered phage clone displaying CYAIGSFDC sequence homing to the skeletal muscle was found predominantly in the vasculature of the skeletal muscle, but not in the pancreas (Fig. 5H
). Taken together, these data indicate that the peptide CRVASVLPC binds to PRLR and suggests that it targets vasculature-exposed PRLR in the pancreas.
DISCUSSION
Here, we present an advance in the in vivo phage display methodology: simultaneous screening of peptide libraries on multiple organs. Parallel analysis of fewer sequences (hundreds) in multiple biopanning rounds performed here, as opposed to thousands of sequences required for the identification of homing motifs in a single round, as reported preciously (10)
, provides control over the progressive organ-specific enrichment of individual motifs throughout the consecutive screening rounds. Our approach enables highly efficient isolation of organ-homing ligands from combinatorial libraries and may streamline vascular mapping efforts, thus leading to a better understanding of the functional protein–protein interactions in the vasculature.
Both Bayesian and frequentist statistics have led to the identification of largely overlapping populations of peptide motifs enriched in the target organs, thus reinforcing the validity of the identified homing motifs. Organ-homing peptides were prioritized for further validation based on their sequence similarity to domains of mouse proteins. Protein-peptide similarity-based survey failed to identify prototype proteins for some of the homing tripeptides. This is not surprising, as the approach does not reveal peptide motifs that mimic conformationally defined interacting sites within proteins. Moreover, it is certainly not a given that every similarity-based lead will ultimately yield a biologically significant interaction. Nevertheless, this strategy has proven effective in the identification of receptor-ligand leads that might subsequently be validated directly by using binding assays (12
, 19)
.
As an example, we validated a pancreas-homing motif CRVASVLPC as a peptide mimic of natural ligands of PRLR, which include PL, PRL, and a number of other hormones from the same family. A concept implemented here is the ability to systematically "retro-BLAST" the receptor-binding peptides onto prototype ligands in order to map sites of ligands involved in receptor interaction. By such retro-targeting of the receptor ligands with peptides derived from in vivo selections, we were able to identify multiple apparent PRL mimics, many of which had been missed in the statistical survey of selected tripeptides. We show that one of the PRL mimics, CRVASVLPC, specifically binds to PRLR in the context of phage coat pIII protein, and that the precise amino acid sequence of the motif appears to be essential for binding. We demonstrate that the CRVASVLPC motifs binds to PRLR in both forward and reverse orientation, however it should be noted that such orientation independence may not be a general feature that is applicable to every ligand-receptor pair interaction. We also show that the CRVASVLPC peptide homes to pancreatic vasculature in vivo, consistent with vascular PRLR expression and with the established function of placental lactogens in the pancreas during pregnancy (22
, 24
, 25
, 28
29
30)
. These findings are also consistent with reported PRLR expression in other organs, as neither VSL, nor other motifs apparently mimicking biological PRLR ligands, qualified the statistical test of organ-specific motif distribution (Table 1
, right column).
In summary, synchronous selection of homing peptides for multiple tissues accelerates the process of phage display-based vascular mapping in vivo. Direct combinatorial screenings in patients (10
, 11
, 13)
have opened the possibility of systematic isolation of peptide ligands for therapeutic targeting. In the future, the use of peptide libraries for probing the vasculature in individual patients for diagnostic and therapeutic purposes may become a reality. The high-throughput targeting strategy bestowed by the synchronous biopanning reported here will open new possibilities for rapid and efficient identification and validation of vascular receptors and their ligand-directed targeting.
ACKNOWLEDGMENTS
We thank Dr. Li-yuanYu-Lee for the plasmid pECE-PRLR. We also thank Xuemei Wang and Carlotta L. Cavazos for technical assistance. This work was supported by grants from the NIH (CA103030, DK67683, and CA90810 to W.A.; CA90270 to R.P.; and CA111853 to M.G.K.) and by awards from the Gillson-Longenbaugh Foundation, the AngelWorks Foundation, and the V Foundation (to W.A. and R.P).
Received for publication September 26, 2005. Accepted for publication November 1, 2005.
REFERENCES
This article has been cited by other articles:
|
|
 |
|
  A. Aartsma-Rus and G.-J. B. van Ommen Antisense-mediated exon skipping: A versatile tool with therapeutic and research applications RNA, October 1, 2007; 13(10): 1609 - 1624. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. A.H. Jarvinen and E. Ruoslahti Molecular Changes in the Vasculature of Injured Tissues Am. J. Pathol., August 1, 2007; 171(2): 702 - 711. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 V. L. T. Ballard, J. M. Holm, and J. M. Edelberg Quantitative PCR-based approach for rapid phage display analysis: a foundation for high throughput vascular proteomic profiling Physiol Genomics, September 14, 2006; 26(3): 202 - 208. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
