Steps toward mapping the human vasculature by phage display

Abstract

The molecular diversity of receptors in human blood vessels remains largely unexplored. We developed a selection method in which peptides that home to specific vascular beds are identified after administration of a peptide library. Here we report the first in vivo screening of a peptide library in a patient. We surveyed 47,160 motifs that localized to different organs. This large-scale screening indicates that the tissue distribution of circulating peptides is nonrandom. High-throughput analysis of the motifs revealed similarities to ligands for differentially expressed cell-surface proteins, and a candidate ligand-receptor pair was validated. These data represent a step toward the construction of a molecular map of human vasculature and may have broad implications for the development of targeted therapies.

Full text

NATURE MEDICINE • VOLUME 8 • NUMBER 2 • FEBRUARY 2002 121
ARTICLES
Despite major progress brought about by the Human Genome
Project
1,2
, the molecular diversity of blood vessels has just
begun to be uncovered. Many therapeutic targets may be ex-
pressed in very restricted but highly specific and accessible lo-
cations in the vascular endothelium. Thus potential targets for
intervention may be overlooked in high-throughput DNA se-
quencing or in gene arrays because these approaches do not
generally take into account cellular location and anatomical
and functional context.
We developed an in vivo selection method in which peptides
that home to specific vascular beds are selected after intra-
venous administration of a phage-display random peptide li-
brary
3
. This strategy revealed a vascular address system that
allows tissue-specific targeting of normal blood vessels
4,5
and
angiogenesis-related targeting of tumor blood vessels
6–10
.
Although the biological basis for such vascular heterogeneity
remains unknown, several peptides selected by homing to
blood vessels in mouse models have been used by several
groups as carriers to guide the delivery of cytotoxic drugs
9
, pro-
apoptotic peptides
6
, metalloprotease inhibitors
7
, cytokines
11
,
fluorophores
12
and genes
13
. Generally, coupling to homing pep-
tides yields targeted compounds that are more effective and
less toxic than the parental compound
6,9,11
. Moreover, vascular
receptors corresponding to the selected peptides have been
identified in blood vessels of normal organs
14
and in tumor
blood vessels
15,16
. Together, these results show that it is possible
to develop therapeutic strategies based on selective expression
of vascular receptors
17
.
Although certain ligands and receptors isolated in mouse
models have been useful to identify putative human ho-
mologs
10,15
, it is unlikely that targeted delivery will always be
achieved in humans using mouse-derived probes. Data from
the Human Genome Project indicate that the higher complex-
ity of the human species relative to other mammalian species
derives from expression patterns of proteins at different tissue
sites, levels or times rather from a greater number of genes
1,2
. In
fact, several examples of species-specific differences in gene ex-
pression within the human vascular network have recently sur-
faced. The divergence in the expression patterns of the
prostate-specific membrane antigen (PSMA) between human
and mouse illustrates such species specificity. Selective expres-
sion of PSMA occurs in the human prostate, but not in the
mouse prostate; instead, the mouse homolog of PSMA is ex-
pressed in the brain and kidney
18
. Additionally, PSMA is a
marker of endothelial cells of tumor blood vessels in humans
19
,
whereas the mouse homolog of PSMA is undetectable in
tumor-associated neovasculature in the mouse (W.D. Heston,
pers. comm.). Another example of such divergence is the TEM7
gene, which is highly expressed in a selective manner in the
endothelium of human colorectal adenomas
20
. By contrast,
mouse Tem7 is expressed not in tumor blood vessels but in
Purkinje cells instead
21
. Thus, striking species-specific differ-
Steps toward mapping the human vasculature
by phage display
WADIH ARAP
1,2
, MIKHAIL G. KOLONIN
1
, MARTIN TREPEL
1
, JOHANNA LAHDENRANTA
1
,
M
ARINA C
ARDÓ-VILA
1
, RICARDO J. G
IORDANO
1
, PAUL J. M
INTZ
1
, PETER U. A
RDELT
1
,
V
IRGINIA
J. YAO
1
, CLAUDIA
I. VIDAL
1
, LIMOR
CHEN
1
, ANNE
FLAMM
3
,
H
ELI VALTANEN
9
, L
ISA M. WEAVIND
5
, M
ARSHALL E. HICKS
6
, R
APHAEL E. POLLOCK
7
,
G
REGORY H. B
OTZ
5
, CORAZON D. B
UCANA
2
, ERKKI K
OIVUNEN
9
, DOLORES C
AHILL
10
,
P
ATRICIA
TRONCOSO
8
, KEITH
A. BAGGERLY
4
, REBECCA
D. PENTZ
3
, KIM
-ANH DO
4
,
C
HRISTOPHER J. LOGOTHETIS
1
& R
ENATA PASQUALINI
1,2
Departments of
1
Genito-Urinary Medical Oncology,
2
Cancer Biology,
3
Clinical Ethics,
4
Biostatistics,
5
Critical Care,
6
Diagnostic Radiology,
7
Surgical Oncology and
8
Pathology,
The University of Texas M.D. Anderson Cancer Center, Houston, Texas, USA
9
Department of Biosciences, Division of Biochemistry, University of Helsinki, Helsinki, Finland
10
Max Planck Institute of Molecular Genetics, Berlin, Germany
M.G.K. and M.T. contributed equally to this study.
Correspondence should be addressed to W.A.; email: warap@notes.mdacc.tmc.edu,
or R.P.; email: rpasqual@notes.mdacc.tmc.edu
The molecular diversity of receptors in human blood vessels remains largely unexplored. We de-
veloped a selection method in which peptides that home to specific vascular beds are identified
after administration of a peptide library. Here we report the first in vivo screening of a peptide
library in a patient. We surveyed 47,160 motifs that localized to different organs. This large-
scale screening indicates that the tissue distribution of circulating peptides is nonrandom. High-
throughput analysis of the motifs revealed similarities to ligands for differentially expressed
cell-surface proteins, and a candidate ligand–receptor pair was validated. These data represent a
step toward the construction of a molecular map of human vasculature and may have broad im-
plications for the development of targeted therapies.
© 2002 Nature Publishing Group http://medicine.nature.com

122 NATURE MEDICINE • VOLUME 8 • NUMBER 2 • FEBRUARY 2002
ARTICLES
ences in protein expression and ligand–receptor accessibility
dictates that vascular targeting data obtained in animal models
must be carefully evaluated before extrapolating to human
studies.
Screening phage-display libraries in humans
We reasoned that in vivo selection of phage-display random
peptide libraries in humans would advance the identification
of human vascular targeting probes and facilitate development
of targeted delivery of therapeutic and imaging agents to the
vasculature. This study reports the initial step toward develop-
ing an in vivo phage display–based, ligand–receptor map of
human blood vessels.
A large-scale preparation of a phage random peptide library
containing the insert CX
7
C (C, cysteine; X, any amino-acid
residue) and designed to display a constrained cyclic loop
within the pIII capsid protein was optimized to create the high-
est possible insert diversity
3
. The diversity of the library was ap-
proximately 2 × 10
8
and its final titer was approximately 1 ×
10
12
transducing units (TU) per ml.
A patient (see Methods) received an intravenous infusion of
the unselected random phage library, and 15 min after infu-
sion tissue biopsies were obtained to provide histopathological
diagnosis and to recover phage from various organs (Fig. 1a).
Here we demonstrate the feasibility of producing phage-dis-
play random peptide libraries on a very large scale and of se-
lecting phage clones that home to different human organs in
vivo through the systemic circulation (Fig. 1b).
High-throughput analysis of selected peptides
To analyze the distribution of inserts from the random peptide
library, we designed a high-throughput pattern recognition
software to analyze short amino-acid residue sequences. This
automated program allowed surveillance of peptide inserts re-
covered from the phage library screening.
Based on SAS (version 8; SAS Institute) and Perl (version 5.0),
the program conducts an exhaustive amino-acid residue se-
quence count and keeps track of the relative frequencies of n
distinct tripeptide motifs representing all
possible n
3
overlapping tripeptide motifs in
both directions (n << n
3
). This analysis was
applied for phage recovered from each tar-
get tissue and for the unselected CX
7
C ran-
dom phage-display peptide library. Counts
were recorded for all overlapping interior
tripeptide motifs, subject only to reflection
and single-voting restrictions. No peptide
was allowed to contribute more than once
for a single tripeptide motif (or a reversed
tripeptide motif). Tripeptide motifs in both
directions are chemically nonsymmetrical
and not necessarily equivalent. However,
because we often recovered forward and
reverse tripeptides recognizing the same
receptor by in vivo phage display, we chose
to take reflection into account, with the
understanding that this is not a general
feature that is applicable to every
ligand–receptor pair interaction. Each
peptide contributed five tripeptide motifs.
Tripeptide motif counts were conditioned
on the total number for all motifs being
held fixed within a tissue. The significance of association of a
given allocation of counts was assessed by the Fisher’s exact test
(one-tailed). Results were considered statistically significant at P
< 0.05. In summary, to test for randomness of distribution, we
compared the relative frequencies of a particular tripeptide
motif from each target to those of the motifs from the unse-
lected library; such an approach is intrinsically a large-scale
contingency table association test.
Distribution of tripeptide motifs in vivo
To determine the distribution of the peptide inserts homing
to specific sites after intravenous administration, we com-
pared the relative frequencies of every tripeptide motif from
each target tissue to those from the unselected library. We an-
alyzed 4,716 phage inserts recovered from representative
samples of five tissues (bone marrow, fat, skeletal muscle,
prostate and skin) and from the unselected library. Tripeptide
motifs were chosen for the phage insert analysis because
three amino-acid residues seem to provide the minimal
framework for structural formation and protein–protein in-
teraction
22
. Examples of such biochemical recognition units
and binding of ligand motifs to their receptors include RGD,
LDV and LLG to integrins
23,24
, NGR to aminopeptidase
N/CD13 (refs 11,15) and GFE to membrane dipeptidase
4,14
.
Each phage insert analyzed contained seven amino-acid
residues and contributed five potential tripeptide motifs;
thus, counting both peptide orientations, a total of 47,160
tripeptide motifs were surveyed.
Comparisons of the motif frequencies in a given organ rela-
tive to those frequencies in the unselected library demon-
strate the nonrandom nature of the peptide distribution
(Table 1); such a bias is particularly noteworthy given that
only a single round of in vivo screening was performed. Of the
tripeptide motifs selected from tissues, some were preferen-
tially recovered in a single site whereas others were recovered
from multiple sites. These data are consistent with some pep-
tides homing in a tissue-specific manner and others targeting
several tissues. We next adapted the ClustalW software from
Fig. 1 In vivo phage-display screening for pep-
tides that home to human tissues through the
systemic circulation. a, Schematic flowchart of
the study. b, Phage recovery from various human
tissues in vivo. Tissue samples were processed
and phage recovered as described in Methods.
Shown are means ± s.e.m. of phage transducing
units (TU) per gram of tissue obtained from each
biopsy site.
a b
© 2002 Nature Publishing Group http://medicine.nature.com

NATURE MEDICINE • VOLUME 8 • NUMBER 2 • FEBRUARY 2002 123
ARTICLES
the European Molecular Biology Laboratory
25
to analyze the
original cyclic phage peptide inserts of seven amino-acid
residues containing the tripeptide motifs. This analysis re-
vealed four to six amino-acid residue motifs that were shared
among multiple peptides isolated from a given organ (Fig. 2).
We searched for each of these motifs in online databases
(through the National Center for Biotechnology Information
(NCBI; http://www.ncbi.nlm.nih.gov/BLAST/) and found
that some appeared within known human proteins. Phage-
display technology is suitable for targeting several classes of
molecules (adhesion receptors, proteases, proteoglycans or
growth factor receptors). Based on extensive work performed
in murine models, the in vivo selection system seems to favor
the isolation of peptides that recognize receptors that are se-
lectively expressed in specific organs or tissues
17
. As our mo-
tifs are likely to represent sequences present in circulating
ligands (either secreted proteins or surface receptors ex-
pressed on circulating cells) that home to vascular receptors,
we compiled a panel of candidate human proteins potentially
mimicked by selected peptide motifs (Table 2).
Validation of candidate ligands
It is tempting to speculate on a few biologically relevant ho-
mology hits. For example, a peptide contained within bone
morphogenetic protein 3B (BMP-3B) was recovered from
bone marrow. BMP-3B is a growth factor known to regulate
bone development
26
. Thus, this protein may be mimicked by
the isolated ligand homing to that tissue. We also isolated
from the prostate a potential mimetope of interleukin 11 (IL-
11), which has been previously shown to interact with recep-
tors within endothelium and prostate epithelium
27,28
. In
addition to secreted ligands, motifs were also found in several
extracellular or transmembrane proteins that may operate se-
lectively in the target tissue, such as sortilin in fat
29
. We have
also recovered motifs from multiple organs; one such peptide
is a candidate mimic peptide of perlecan, a protein known to
maintain vascular homeostasis
30
.
To test the tissue specificity of the peptides selected, we de-
veloped a phage-overlay assay for tissue sections. Because of
the availability and well-characterized interaction between
the candidate ligand (IL-11) and its receptor (IL-11Rα), we
chose the motif RRAGGS, a peptide mimic of IL-11 (Table 2),
for validation. We show by phage overlay on human tissue
sections (see Methods) that a prostate-homing phage display-
ing an IL-11 peptide mimic specifically bound to the en-
dothelium and to the epithelium of normal prostate (Fig. 3a),
but not to control organs, such as skin (Fig. 3b). In contrast, a
phage selected from the skin (displaying the motif HGGVG;
Table 2), did not bind to prostate tissue (Fig. 3c); however,
this phage specifically recognized blood vessels in the skin
(Fig. 3d). Moreover, the immunostaining pattern obtained
with an antibody against human IL-11Rα on normal prostate
tissue (Fig. 3e) is undistinguishable from that of the CGRRAG-
GSC-displaying phage overlay (Fig. 3a); a control antibody
showed no staining in prostate tissue (Fig. 3f). These findings
were recapitulated in multiple tissue sections obtained from
several different patients.
Finally, using a ligand–receptor binding assay in vitro,we
demonstrate the interaction of the CGRRAGGSC-displaying
phage with immobilized IL-11Rα at the protein–protein level
(Fig. 4a). Such binding is specific because it was inhibited by
the native IL-11 ligand in a concentration-dependent manner
(Fig. 4b). Preliminary results indicate that serum IL-11 is ele-
vated in a subset of prostate cancer patients (C.J.L., unpub-
lished observations) and that the expression of IL-11Rα in
tumors is upregulated in some cases of human prostate cancer
(M.G.K., unpublished observations); these data may have
clinical relevance.
Discussion
Aside from in vivo phage display, use of methods such as serial
analysis of gene expression (SAGE) clearly shows that the ge-
netic progression of malignant cells is paralleled by epige-
netic changes in nonmalignant endothelial cells induced by
angiogenesis of the tumor vasculature
20
. Because SAGE is
based on differential expression levels of transcripts, it fails to
address functional interactions (for example, binding) at the
protein–protein level. The complexity of the human endothe-
lium is also apparent from recent studies showing that the
profile of certain endothelial cell receptors can vary depend-
ing on ethnic background
31
. In fact, in vivo phage-display in
humans might reveal diversity of receptors expressed in the
blood vessels even at the level of individual patients.
Table 1 Peptide motifs isolated by in vivo phage display screening
Target organ and motif Motif frequency P value
(%)
Unselected library
None NA NA
Bone marrow
GGG* 2.3 0.0350
GFS* 1.0 0.0350
LWS* 1.0 0.0453
ARL 1.0 0.0453
FGG 1.1 0.0453
GVL 2.3 0.0137
SGT 1.1 0.0244
Fat
EGG* 1.3 0.0400
LLV*
1.0 0.0269
LSP* 0.9 0.0402
EGR 1.1 0.0180
FGV 0.9 0.0402
Muscle
LVS* 2.1 0.0036
GER 0.9 0.0036
Prostate
AGG* 2.5 0.0340
EGR 1.0 0.0340
GER 0.9 0.0382
GVL 2.3 0.0079
Skin
GRR* 2.9 0.0047
GGH* 0.9 0.0341
GTV* 0.8 0.0497
ARL 0.8 0.0497
FGG 1.3 0.0076
FGV 1.0 0.0234
SGT 1.0 0.0234
Peptide motifs isolated by in vivo phage display screening. Motifs occurring in peptides
isolated from target organs but not from the unselected phage library (Fisher’s exact
test, one-tailed; P < 0.05). Number of peptide sequences analyzed per organ: unselected
library, 446; bone marrow, 521; fat, 901; muscle, 850; prostate, 1,018; skin, 980.
*, Motifs enriched only in a single tissue. Motif frequencies represent the prevalence of
each tripeptide divided by the total number of tripeptides analyzed in the organ.
© 2002 Nature Publishing Group http://medicine.nature.com

124 NATURE MEDICINE • VOLUME 8 • NUMBER 2 • FEBRUARY 2002
ARTICLES
However, our validation studies show that at
least some ligand–receptor pairs are de-
tectable in multiple unrelated subjects.
Another advantage of the method described
here is that selected targeting peptides bind
to native receptors as they are expressed in
vivo. Even if a ligand–receptor interaction is
mediated through a conformational rather
than a linear epitope, it is possible to select
binders in the screening. Furthermore, it is
often difficult to ensure that proteins ex-
pressed in array systems maintain the correct
structure and folding. Thus, peptides se-
lected in vivo may be more suitable to clini-
cal applications.
Precedent exists to suggest that phage can
be safely administered to patients, as bacte-
riophage were used in humans during the
pre-antibiotic era
32
. Ultimately, it may be-
come possible to determine molecular pro-
files of blood vessels in specific conditions;
infusing phage libraries systemically before
resections of lung, prostate, breast and col-
orectal carcinomas, or even regionally before
resection of limb sarcomas may yield useful
vascular targets. Exploiting this experimen-
tal paradigm systematically with the analyti-
cal tools developed here may permit the
construction of a molecular map outlining
vascular diversity in each human organ, tis-
sue or disease. Translation of high-through-
put in vivo phage-display technology may
provide a contextual and functional link be-
tween genomics and proteomics. Based on
the therapeutic promise of peptide- or pep-
tidomimetic-targeting probes
33
, clinical ap-
plications are likely to follow.
Methods
Patient selection and clinical course. A 48-y-old
male Caucasian patient was diagnosed with
Waldenström macroglobulinemia (a B-cell malig-
nancy) and previously treated by splenectomy, sys-
temic chemotherapy (fludarabine, mitoxantrone
and dexamethasone) and immunotherapy (anti-
CD20 monoclonal antibody). In the few months
preceding his admission, the disease became refrac-
tory to treatment and clinical progression with
retroperitoneal lymphadenopathy, pancytopenia
and marked bone marrow infiltration by tumor cells
occurred. The patient was admitted with massive in-
tracranial bleeding secondary to thrombocytopenia.
Despite prompt craniotomy and surgical evacuation
of a cerebral hematoma, the patient remained co-
matose with progressive and irreversible loss of
brainstem function until the patient met the formal
criteria for brain-based determination of death
34
;
such determination was carried out by an indepen-
dent clinical neurologist not involved in the project.
Because of his advanced cancer, the patient was
considered and rejected as transplant organ donor.
After surrogate written informed consent was ob-
tained from the legal next of kin, the patient was en-
rolled in the clinical study. Disconnection of the
patient from life-support systems followed the pro-
Fig. 2 Identification of extended homing motifs with the Clustal W pro-
gram (European Molecular Biology Laboratory; EMBL). a and b, Peptide se-
quences containing selected tripeptides (Table 1) preferentially isolated
from each single tissue (a) or in multiple tissues (b) were aligned in clusters
to obtain longer motifs shared between different peptides from each clus-
ter. The software registers sequence identities and similarities among multi-
ple peptide sequences and aligns the sequences by placing peptides with
the most similarity or identity next to one another. Similarity between pep-
tides at the level of amino-acid class is color-coded: red, hydrophobic;
green, neutral and polar; pink, basic; blue, acidic. The original and ex-
tended peptide motifs are highlighted in yellow.
a
b
© 2002 Nature Publishing Group http://medicine.nature.com

NATURE MEDICINE • VOLUME 8 • NUMBER 2 • FEBRUARY 2002 125
ARTICLES
cedure. This study strictly adheres to current medical ethics recommen-
dations and guidelines regarding human research
35
, and it has been re-
viewed and approved by the Clinical Ethics Service, the Institutional
Biohazard Committee, Clinical Research Committee and the
Institutional Review Board of the University of Texas M.D. Anderson
Cancer Center.
The University of Texas and researchers (W.A. and R.P.) have equity
in NTTX Biotechnology, which is subjected to certain restrictions under
university policy; the university manages the terms of these arrange-
ments in accordance to its conflict-of-interest policies.
In vivo phage display. Short-term intravenous infusion of the phage li-
brary (a total dose of 1 × 10
14
phage TU suspended in 100 ml of saline) into
the patient was followed by multiple representative tissue biopsies.
Prostate and liver samples were obtained by needle biopsy under ultra-
sonographic guidance; skin, fat-tissue and skeletal-muscle samples were
obtained by a surgical excision. Bone-marrow needle aspirates and core
biopsy samples were also obtained. Histopathological diagnosis was deter-
mined by examination of frozen sections processed from tissues obtained
at the bedside. Triplicate samples were processed for host bacterial infec-
tion, phage recovery and histopathological analysis. In brief, tissues were
weighed, ground with a glass Dounce homogenizer, suspended in 1 ml of
DMEM containing proteinase inhibitors (DMEM-prin; 1 mM phenyl-
methylsulfonyl fluoride (PMSF), 20 µg/ml aprotinin, and 1 µg/ml leu-
peptin), vortexed, and washed three times with DMEM-prin. Next, human
tissue homogenates were incubated with 1 ml of host bacteria (log phase
Escherichia coli K91kan; OD
600
≈ 2). Aliquots of the bacterial culture were
plated onto Luria–Bertani agar plates containing 40 µg/ml tetracycline and
100 µg/ml of kanamycin. Plates were incubated overnight at 37 °C.
Bacterial colonies were processed for sequencing of phage inserts recov-
ered from each tissue and from unselected phage library. Human samples
were handled with universal blood and body fluid precautions.
Table 2 Examples of candidate human proteins mimicked by selected peptide motifs
Extended motif * Human protein Protein description Accession
containing the motif number
Bone marrow
PGGG Bone morphogenetic protein 3B Growth factor, TGF-β family member NP_004953
PGGG Fibulin 3 Fibrillin- and EGF-like Q12805
GHHSFG Microsialin Macrophage antigen, glycoprotein NP_001242
Fat
EGGT LTBP-2 Fibrillin- and EGF-like, TGF-β Interactor CAA86030
TGGE Sortilin Adipocyte differentiation-induced receptor CAA66904
GPSLH Protocadherin gamma C3 Cell adhesion AAD43784
Muscle
GGSVL ICAM-1 Intercellular adhesion molecule P05362
LVSGY Flt4 Endothelial growth factor receptor CAA48290
Prostate
RRAGGS Interleukin 11 Cytokine NP_000632
RRAGG Smad6 Smad family member AAB94137
Skin
GRRG TGF-β1 Growth factor, TGF-β family member XP_008912
HGG+G Neuropilin-1 Endothelial growth factor receptor AAF44344
+PHGG Pentaxin Infection/trauma-induced glycoprotein CAA45158
PHGG Macrophage-inhibitory cytokine-1 Growth factor, TGF-β family member AAB88673
+PHGG Desmoglein 2 Epithelial cell junction protein S38673
VTG+SG Desmoglein 1 Epidermal cell junction protein AAC83817
Multiple organs
EGRG MMP-9 Gelatinase AAH06093
GRGE
ESM-1
Endothelial cell-specific molecule XP_003781
NFGVV
CDO Surface glycoprotein, Ig- and fibronectin-like NP_058648
GERIS BPA1 Basement membrane protein NP_001714
SIREG Wnt-16 Glycoprotein Q9UBV4
+GVLW Sialoadhesin Ig-like lectin AAK00757
WLVG+ IL-5 receptor Soluble interleukin 5 receptor CAA44081
GGFR Plectin 1 Endothelial focal junction-localized protein CAA91196
GGFF TRANCE Cytokine, TNF family member AAC51762
+SGGF MEGF8 EGF-like protein T00209
PSGTS ICAM-4 Intercellular adhesion glycoprotein Q14773
+TGSP Perlecan Vascular repair heparan sulfate proteoglycan XP_001825
For similarity searches, tripeptide motif-containing peptides (in either orientation) selected by in vivo phage display screening were used. *Extended motifs containing at least 4–6
amino acid residues (Fig. 2) were analyzed using BLAST (NCBI) to search for similarity to known human proteins. Examples of candidate proteins potentially mimicked by the pep-
tides selected in the in vivo screening are listed. Sequences correspond to the regions of 100% identity between the peptide selected and the candidate protein. Conserved amino
acid substitutions are indicated as (+). Tripeptides shown in Table 1 are highlighted. TGF, transforming growth factor; TNF, tumor necrosis factor.
© 2002 Nature Publishing Group http://medicine.nature.com

126 NATURE MEDICINE • VOLUME 8 • NUMBER 2 • FEBRUARY 2002
ARTICLES
Statistical analysis. Let p be the probability of observing a particular
tripeptide motif under total randomness, and q = 1 – p. Under such para-
meters, the probability of observing K sequences characterized as
a particular tripeptide motif out of n
3
total tripeptide motif sequences
is binomial (n
3
, p) and may be approximated by the equation
p
K
= Φ[(k + 1)/sqrt(n
3
pq)] – Φ[k/sqrt(n
3
pq)], where Φ is the usual cumu-
lative Gaussian probability. The value p
K
may be treated as a P value in
testing for total randomness of observing exactly K sequences of a par-
ticular tripeptide motif. However, this test requires exact knowledge of
the true value of p, which it is difficult to obtain in practice with cer-
tainty. Therefore, to identify the motifs that were isolated in the screen-
ing, the count for each tripeptide motif within each tissue was compared
with the count for that tripeptide motif within the unselected library.
Immunocytochemistry and phage overlays. Immunohistochemistry
on sections of fixed human paraffin-embedded tissues was done using
the LSAB+ peroxidase kit (DAKO, Carpinteria, California) as described
3
.
For overlay experiments, phage was used at the concentration of 5 ×
10
10
TU/ml. For phage immunolocalization, a rabbit anti-fd bacterio-
phage antibody (B-7786; Sigma) was used at 1:500 dilution. For IL-
11Rα immunolocalization, a goat antibody (sc-1947; Santa Cruz
Biotechnology, Santa Cruz, California) was used at 1:10 dilution. Phage
binding to tissue sections was evaluated by the intensity of immunos-
taining relative to controls.
In vitro protein binding assays. Recombinant (R&D Systems,
Minneapolis, Minnesota) interleukin-11 receptor α (IL-11Rα), vascular
endothelial growth factor receptor-1 (VEGFR1), and leptin receptor (OB-
R) were immobilized on microtiter wells (at 1 µg in 50 µl PBS) overnight
at 4°C, washed twice with PBS, blocked with 3% BSA in PBS for 2 h at
room temperature, and incubated with 1x 10
9
TU of CGRRAGGSC-dis-
playing phage in 50 µl of 1.5% BSA in PBS. An unrelated phage clone
(displaying the peptide CRVDFSKGC) and insertless phage (fd-tet) were
used as controls. After 1 h at room temperature, wells were washed nine
times with PBS, after which bound phage were recovered by bacterial in-
fection and plated as described
3
. Either IL-11 or IL-1 (negative control)
was used to inhibit phage binding to IL-11Rα. Phage were incubated
with the immobilized IL-11Rα in the presence of increasing concentra-
tions of either IL-11 or IL-1. Binding of CGRRAGGSC-displaying phage on
immobilized IL-11Rα in the absence of interleukins was set to 100%.
Fig. 3 Validation of the candidate receptor–ligand pairs resulting from
the in vivo selection. a–f, Phage clones isolated from prostate and from
skin were evaluated for binding to human tissues in an overlay assay.
Shown are paraffin-embedded tissue sections of human prostate (a, c, e
and f) and of human skin (b and d) overlaid with prostate-homing CGR-
RAGGSC-displaying phage (a and b) or skin-homing CHGGVGSGC-dis-
playing phage (c and d). Phage were detected by using an anti-M13
phage antibody. In e, IL-11Rα expression was determined by conven-
tional immunostaining with an anti-IL-11Rα antibody; a and e show
similar immunostaining patterns (brown staining). f, Negative control
antibody on prostate tissue sections. Arrowheads, positive endothelium;
asterisks, positive epithelium. Scale bar, 160 µm (a, c, e and f); 40 µm
(b and d).
Fig. 4 Characterization of CGRRAGGSC-displaying phage binding proper-
ties by using purified receptors in vitro. a, Recombinant interleukin-11 re-
ceptor α (IL-11Rα), vascular endothelial growth factor receptor-1 (VEGFR1),
or leptin receptor (OB-R) were incubated with the CGRRAGGSC
-displaying phage (). VEGFR1 was used as a representative vascular re-
ceptor; OB-R was used because it is homologous to a coreceptor of IL-
11Rα. An unrelated phage clone (displaying the peptide CRVDFSKGC, )
and insertless phage (fd-tet, ) were used as controls. Phage binding
was evaluated and quantified as described (see Methods). b, Specificity of
phage binding to the IL-11 receptor. Phage were incubated with the immo-
bilized IL-11Rα in the presence of increasing concentrations of either IL-11
(native ligand, ) or IL-1 (negative control, ). The experiments were per-
formed three times with similar results. Shown are mean ± s.e.m. from
triplicate wells.
a b
c d
e
f
a
b
© 2002 Nature Publishing Group http://medicine.nature.com

NATURE MEDICINE • VOLUME 8 • NUMBER 2 • FEBRUARY 2002 127
ARTICLES
Acknowledgements
We thank R.C. Bast, Jr., R.R. Brentani, W.K. Cavenee, A.C. von
Eschenbach, I.J. Fidler, W.K. Hong, D.M. McDonald, J. Mendelsohn and
L.A. Zwelling for comments on the manuscript; W.D. Heston for sharing
unpublished data; C.L. Cavazos, P.Y. Dieringer, R.G. Nikolova, C.A. Perez,
B.H. Restel, C.P. Soto and X. Wang for technical assistance. This work was
funded in part by grants from NIH (CA90270 and CA8297601 to R.P.,
CA90270 and CA9081001 to W.A.) and awards from the
Gilson–Longenbaugh Foundation and CaP CURE (to R.P. and W.A.).
M.G.K., J.L. and P.J.M. received support from the Susan G. Komen Breast
Cancer Foundation, R.J.G. from FAPESP (Brazil), M.C.V. from the
Department of Defense, L.C. from the NCCRA and E.K. from the Academy
of Finland.
Competing interests statement
The authors declare competing financial interests; see the Nature
Medicine web site (http://medicine.nature.com) for details.
RECEIVED 19 NOVEMBER 2001; ACCEPTED 2 JANUARY 2002
1. Lander, E.S. et al. Initial sequencing and analysis of the human genome. Nature
409, 860–921 (2001).
2. Venter, J.C. et al. The sequence of the human genome. Science 291, 1304–1351
(2001).
3. Pasqualini, R., Arap, W., Rajotte, D. & Ruoslahti, E. In vivo phage display. In Phage
display: a laboratory manual. (eds. Barbas, C. F., Burton, D.R., Scott, J.K. &
Silverman, G.J.) 1–24 (Cold Spring Harbor Laboratory Press, Cold Spring Harbor,
New York, 2000).
4. Rajotte, D. et al. Molecular heterogeneity of the vascular endothelium revealed by
in vivo phage display. J. Clin. Invest. 102, 430–437 (1998).
5. Pasqualini, R. & Ruoslahti, E. Organ targeting in vivo using phage display peptide li-
braries. Nature 380, 364–366 (1996).
6. Ellerby, H.M. et al. Anti-cancer activity of targeted pro-apoptotic peptides. Nature
Med. 5, 1032–1038 (1999).
7. Koivunen, E. et al. Tumor targeting with a selective gelatinase inhibitor. Nature
Biotechnol. 17, 768–774 (1999).
8. Burg, M.A., Pasqualini, R., Arap, W., Ruoslahti, E. & Stallcup, W.B. NG2 proteogly-
can-binding peptides target tumor neovasculature. Cancer Res. 59, 2869–2874
(1999).
9. Arap, W., Pasqualini, R. & Ruoslahti, E. Cancer treatment by targeted drug delivery
to tumor vasculature in a mouse model. Science 279, 377–380 (1998).
10. Pasqualini, R., Koivunen, E. & Ruoslahti, E. α
v
integrins as receptors for tumor tar-
geting by circulating ligands. Nature Biotechnol. 15, 542–546 (1997).
11. Curnis, F. et al. Enhancement of tumor necrosis factor α antitumor immunothera-
peutic properties by targeted delivery to aminopeptidase N (CD13). Nature
Biotechnol. 18, 1185–1190 (2000).
12. Hong, F. D. & Clayman, G. L. Isolation of a peptide for targeted drug delivery into
human head and neck solid tumors. Cancer Res. 60, 6551–6556 (2000).
13. Trepel, M., Grifman, M., Weitzman, M.D. & Pasqualini, R. Molecular adaptors for
vascular-targeted adenoviral gene delivery. Hum. Gene Ther. 11, 1971–1981
(2000).
14. Rajotte, D. & Ruoslahti, E. Membrane dipeptidase is the receptor for a lung-target-
ing peptide identified by in vivo phage display. J. Biol. Chem. 274, 11593–11598
(1999).
15. Pasqualini, R. et al. Aminopeptidase N is a receptor for tumor-homing peptides and
a target for inhibiting angiogenesis. Cancer Res. 60, 722–727 (2000).
16. Bhagwat, S.V. et al. CD13/APN is activated by angiogenic signals and is essential
for capillary tube formation. Blood 97, 652–659 (2001).
17. Kolonin, M.G., Pasqualini, R. & Arap, W. Molecular addresses in blood vessels as
targets for therapy. Curr. Opinion. Chem. Biol. 5, 308–313 (2001).
18. Bacich, D.J., Pinto, J.T., Tong, W.P. & Heston, W.D. Cloning, expression, genomic
localization, and enzymatic activities of the mouse homolog of prostate-specific
membrane antigen/NAALADase/folate hydrolase. Mamm. Genome 12, 117–123
(2001).
19. Chang, S.S. et al. Five different anti-prostate-specific membrane antigen (PSMA)
antibodies confirm PSMA expression in tumor-associated neovasculature. Cancer
Res. 59, 3192–3198 (1999).
20. St Croix, B. et al. Genes expressed in human tumor endothelium. Science 289,
1197–1202 (2000).
21. Carson-Walter, E.B. et al. Cell surface tumor endothelial markers are conserved in
mice and humans. Cancer Res. 61, 6649–6655 (2001).
22. Vendruscolo, M., Paci, E., Dobson, C.M. & Karplus, M. Three key residues form a
critical contact network in a protein folding transition state. Nature 409, 641–645
(2001).
23. Ruoslahti, E. RGD and other recognition sequences for integrins. Annu. Rev. Cell
Dev. Biol. 12, 697–715 (1996).
24. Koivunen, E. et al. Inhibition of β
2
integrin-mediated leukocyte cell adhesion by
leucine-leucine-glycine motif-containing peptides. J. Cell Biol. 153, 905–916
(2001).
25. Thompson, J.D., Higgins, D.G. & Gibson, T.J. CLUSTAL W: improving the sensitivity
of progressive multiple sequence alignment through sequence weighting, posi-
tion-specific gap penalties and weight matrix choice. Nucleic Acids Res. 22,
4673–4680 (1994).
26. Daluiski, A. et al. Bone morphogenetic protein-3 is a negative regulator of bone
density. Nature Genet. 27, 84–88 (2001).
27. Mahboubi, K., Biedermann, B.C., Carroll, J.M. & Pober, J.S. IL-11 activates human
endothelial cells to resist immune-mediated injury. J. Immunol. 164, 3837–3846
(2000).
28. Campbell, C.L., Jiang, Z., Savarese, D.M. & Savarese, T.M. Increased expression of
the interleukin-11 receptor and evidence of STAT3 activation in prostate carci-
noma. Am. J. Pathol. 158, 25–32 (2001).
29. Lin, B.Z., Pilch, P.F. & Kandror, K.V. Sortilin is a major protein component of Glut4-
containing vesicles. J. Biol. Chem. 272, 24145–24147 (1997).
30. Nugent, M.A., Nugent, H.M., Iozzo, R.V., Sanchack, K. & Edelman, E.R. Perlecan is
required to inhibit thrombosis after deep vascular injury and contributes to en-
dothelial cell-mediated inhibition of intimal hyperplasia. Proc. Natl Acad. Sci. USA
97, 6722–6727 (2000).
31. Wu, K.K. et al. Thrombomodulin Ala455Val polymorphism and risk of coronary
heart disease. Circulation 103, 1386–1389 (2001).
32. Barrow, P.A. & Soothill, J. S. Bacteriophage therapy and prophylaxis: rediscovery
and renewed assessment of potential. Trends Microbiol. 5, 268–271 (1997).
33. Latham, P.W. Therapeutic peptides revisited. Nature Biotechnol. 17, 755–757
(1999).
34. Wijdicks, E.F. The diagnosis of brain death. N. Engl. J. Med. 344, 1215–1221 (2001).
35. Implementing Human Research Regulations; in President’s Commission for the Study
of Ethical Problems in Medicine and Biomedical and Behavioral Research. (The United
States of America Government’s Printing Office, Washington, DC; 1983).
© 2002 Nature Publishing Group http://medicine.nature.com