Content uploaded by Olga V Volpert
Author content
All content in this area was uploaded by Olga V Volpert on Mar 21, 2014
Content may be subject to copyright.
196 NATURE MEDICINE • VOLUME 6 • NUMBER 2 • FEBRUARY 2000
ARTICLES
Maspin, a unique member of the serpin family, is a secreted
protein encoded by a class II tumor suppressor gene whose
downregulation is associated with the development of breast
and prostate cancers1,2. Overexpression of maspin in breast
tumor cells limits their growth and metastases in vivo. In this
report we demonstrate that maspin is an effective inhibitor of
angiogenesis. In vitro, it acted directly on cultured endothelial
cells to stop their migration towards basic fibroblast growth
factor and vascular endothelial growth factor and to limit mi-
togenesis and tube formation. In vivo, it blocked neovascular-
ization in the rat cornea pocket model. Maspin derivatives
mutated in the serpin reactive site lost their ability to inhibit
the migration of fibroblasts, keratinocytes, and breast cancer
cells but were still able to block angiogenesis in vitro and in
vivo. When maspin was delivered locally to human prostate
tumor cells in a xenograft mouse model, it blocked tumor
growth and dramatically reduced the density of tumor-associ-
ated microvessels. These data suggest that the tumor suppres-
sor activity of maspin may depend in large part on its ability to
inhibit angiogenesis and raise the possibility that maspin and
similar serpins may be excellent leads for the development of
drugs that modulate angiogenesis.
To study the potential anti-angiogenic properties of maspin,
the mouse protein was produced in Escherichia coli as a recom-
binant glutathione S-transferase (GST) fusion protein and
tested in a variety of angiogenesis assays. Recombinant maspin
blocked endothelial cell migration induced by vascular en-
dothelial growth factor (VEGF) in a dose dependent manner
with a median effective dose (ED50) of 0.2 µM–0.3 µM (Fig. 1a).
Similar results were obtained using basic fibroblast growth fac-
tor (bFGF) as an inducer (Fig. 1b; data not shown). At 1 µM,
maspin completely blocked the response of the endothelial
cells to both angiogenic inducers, whereas the GST control was
inactive. Maspin also inhibited the growth of endothelial cells
(Fig. 1c) and prevented them from forming tubes in a matrigel
assay (data not shown).
Purified maspin effectively inhibited neovascularization
in vivo. Rat corneas were surgically implanted with non-
inflammatory, slow release pellets containing maspin with
bFGF and examined 6 or 7 days later for ingrowth of vessels.
Maspin completely blocked bFGF-induced neovascularization
(Fig. 2; compare i and v).
Maspin, a member of the large family of serine protease in-
hibitors (serpins), has been shown to serve as a substrate rather
than an inhibitor for trypsin-like serine proteinases3, suggest-
ing that it may fall into the growing category of noninhibitory
serpins that lack antiprotease activity. One recent study, using
purified reagents in vitro, suggested that maspin might have
some antiprotease activity as it inhibited tissue plasminogen
activator in vitro4. We were unable to confirm these results in
our laboratory5, but to determine if the anti-angiogenic activity
of maspin depended on the inhibition of some undefined pro-
tease, we constructed, expressed, and tested several mutants.
The RSL (reactive serpin loop) near the C-terminus of serpin
family members is essential for their antiprotease activity.
Mutations at the RSL region of other serpins, especially at the
P1 site, abolish serpin activity6. To disrupt this loop in maspin
two different mutations were introduced in the RSL region: a
C-terminal deletion downstream of P7′residue7(maspin ∆RSL)
and a conversion of the critical P1 arginine of the RSL loop to a
glutamine (maspin*). A third mutant was constructed in which
the first 139 amino acids were removed but the serpin region
was left intact (maspin∆N). We tested recombinant maspin and
two RSL mutants in a quantitative tissue-type plasminogen
activator (tPA) assay. As previously demonstrated5, neither
wild-type maspin nor its derivatives displayed any tPA inhibi-
tion (data not shown).
We used migration assays, with nonendothelial cells, to
demonstrate that the constructed mutants were indeed defec-
tive. An intact RSL region is essential for maspin to block the
migration of breast tumor cells in vitro8. When we repeated this
assay with mutant proteins both maspin* and maspin∆N had
no inhibitory effect (Fig. 1d). All three mutants were also un-
able to inhibit the migration towards bFGF of normal human
fibroblasts (Fig. 1e) or normal human keratinocytes (Fig. 1f)
although wild-type maspin protein produced in the same way
was an effective inhibitor.
When we tested these defective mutants on endothelial cells
(see Fig. 1b and c), they behaved somewhat differently. Those
with RSL defects that were unable to block the migration of
other cells, retained the ability to inhibit endothelial cell mi-
gration and mitogenesis. Protein with mutations in the RSL re-
gion also retained the ability to inhibit neovascularization in
vivo (Fig. 2, compare i and vii). The N-terminal deletion,
maspin∆N, was defective in all assays so it was not possible to
determine if a crucial active region had been deleted or if it was
just a dead protein.
To determine if the ability of maspin to inhibit angiogenesis
is involved in its antitumor activity, we used an athymic mouse
xenograft model. We implanted LNCaP prostate tumor cells
subcutaneously on the bidorsal back of nude mice and moni-
tored tumor growth and neovascularization after systemic
treatment with exogenous maspin. Maspin-treated tumors con-
tained considerably fewer vessels as determined by CD31 im-
munostaining than GST-treated controls (Fig. 3) To determine
whether maspin effects on the tumor-induced vasculature were
maintained during a more prolonged treatment, the above ex-
Maspin is an angiogenesis inhibitor
MING ZHANG1, OLGA VOLPERT2, YIHUI H. SHI1, AND NOËL BOUCK2
1Department of Molecular & Cellular Biology, Baylor College of Medicine, Houston, Texas 77030, USA
2Department of Microbiology-Immunology and Robert H. Lurie Comprehensive Cancer Center, Northwestern
University Medical School, Chicago, Illinois 60611, USA
Correspondence should be addressed to M.Z.; email: mzhang@bcm.tmc.edu
© 2000 Nature America Inc. • http://medicine.nature.com
© 2000 Nature America Inc. • http://medicine.nature.com
NATURE MEDICINE • VOLUME 6 • NUMBER 2 • FEBRUARY 2000 197
ARTICLES
periment was replicated with tumors harvested after 7–8 weeks.
We treated 32 tumor sites with maspin and 37 with GST. At
week 8, the growth of 53% of the maspin-treated tumors had
been completely inhibited. The remaining 15 maspin-treated
tumors were reduced in size to 29.2% of GST control-treated tu-
mors. The effect of maspin was reversible. Among those sites
that exhibited no detectable tumors at week 8, most developed
palpable tumors within 1–3 weeks after cessation of maspin
treatment, indicating that some viable tumor cells remained.
To determine if the reduced size of maspin-treated tumors
coincided with reduced neovascularization, we used 20 repre-
sentative tumors from either maspin-treated (10 sites) or GST-
treated tumors (10 sites) to quantify the density of microvessels
after immunostaining with CD31 antibody (Fig. 3 and Table1).
The density of vessels in maspin-treated tumors was
reduced to 37.8% of control tumors and this differ-
ence was highly significant. We also compared the
treated and control tumors of similar size. There was
also reduction of vessel density (ranging from 33.3%
to 45.5%, n = 4) in the maspin treated samples.
There is little evidence to indicate exactly how
maspin blocks angiogenesis. It could act through a re-
ceptor-mediated event, as does thrombospondin-19.
The discrepancy in the effect of the RSL domain on en-
dothelial and breast tumor cells may be due to the
difference of receptors located on both cells.
Alternatively, it could mimic the protease-indepen-
dent effects of plasminogen activator inhibitor-1 (PAI-
1). PAI-1 has a unique domain located at the
N-terminal that regulates cell motility10. Experiments
are underway to investigate this domain of maspin.
The observed ED50 of 0.2–0.3 µM in the capillary en-
dothelial cell migration assay indicates that maspin is
less potent than inhibitors like angiostatin (10 nM)11,
but more effective than small molecules such as cap-
topril (10 µM)12. However, in epithelial tumor cells,
exogenously added maspin localizes at the cell sur-
face13. If this also happens on endothelial cells, then
the concentration of soluble protein may not be par-
ticularly meaningful. Maspin was effective against
several inducers in vitro, and in vivo it blocked angio-
Fig. 1 Effect of maspin and its derivatives
on cultured cells. a,GST–maspin was
tested at a range of concentrations for its
ability to inhibit endothelial cell migration
induced by 100 pg/ml VEGF. VEGF, migra-
tion towards VEGF alone; BSA, background
migration in the absence of a gradient.
b, GST–maspin and its mutants (1 µM)
were tested for their ability to inhibit en-
dothelial cell migration towards 10 ng/ml
bFGF. Maspin, GST–maspin fusion protein;
maspin ∆RSL, maspin with a deletion at the
C-terminus; maspin*, maspin containing an
R to Q mutation in the P1 residue of the RSL
loop; maspin ∆N, maspin with a deletion at
the N-terminus. Glutathione-S-transferase
tested alone was neutral in this assay. *, P <
0.01 compared with migration towards
bFGF (bFGF-----). c, Maspin and its mutants
(1 µM) were tested for their ability to in-
hibit the growth of endothelial cells after
5 d. *, P < 0.05 compared with GST.
d, Maspin and its mutants were tested at a
concentration of 0.3 µM for their ability to
inhibit the invasion and migration of breast
tumor cells. *, P < 0.03. e, Maspin and its
mutants were tested at a concentration of 1
µM for their ability to inhibit the migration of normal human fibroblasts
towards bFGF (20 ng/ml). *, P < 0.05. f, Maspin and its mutants were
Fig. 2 Maspin inhibition of corneal neovascularization. Pellets containing 10 µM test
substances at with or without 100 ng/ml bFGF were incorporated into Hydron slow-
release pellets and implanted in rat corneas. After 6 or 7 d, rats were perfused with col-
loidal carbon to visualize vessels, and excised corneas were photographed with a ×20
objective (b) and scored for neovascularization (a). +/–, corneas in which only one or two
vessels were induced that did not reach the pellet.
tested for their ability to block the migration of normal human
keratinocytes towards 20 ng/ml bFGF. *, P < 0.03.
a b
c
d
a b
e f
v
i
vii
ix
i
ii
iii
iv
v
vi
vii
viii
ix
© 2000 Nature America Inc. • http://medicine.nature.com
© 2000 Nature America Inc. • http://medicine.nature.com
198 NATURE MEDICINE • VOLUME 6 • NUMBER 2 • FEBRUARY 2000
ARTICLES
genesis induced by bFGF and reduced tumor
angiogenesis in response to the LNCaP, a cell
line that produces VEGF as its major angio-
genic factor (J. D. and N. B., unpublished data).
Complete inhibition of endothelial cell mi-
gration in vitro was achieved between 0.5 and 1
µM, in the same concentration range where
maspin also inhibits tumor cell motility and in-
vasion13, but the mechanisms underlying these
two maspin activities seem to be different. The
former requires that the protein have an intact
RSL (refs. 4, 8), whereas this feature was not es-
sential for the inhibition of angiogenesis. Thus,
even if maspin does inhibit some unidentified
protease, this serpin activity is probably not in-
volved in the inhibition of angiogenesis.
It is probable that maspin produced by
tumor-associated normal tissue as well as that
produced by developing tumor cells them-
selves can influence tumor growth. Maspin is produced selec-
tively and at high levels by myoepithelial cells, which surround
the normal mammary ducts. The myoepithelial cells them-
selves form only low-grade tumors and that they also may
delay the progression of adjacent ductal carcinomas in situ to
invasive carcinomas14. The anti-angiogenic nature of the
maspin they secrete offers a possible explanation for both phe-
nomena.
The ability to inhibit angiogenesis is only one of several activ-
ities associated with the intact maspin protein. Other serpins
also have multiple functions and several of them are linked to
angiogenesis and tumor growth. Plasminogen activator in-
hibitor-1 is involved in modulating both proteolysis and angio-
genesis. Pigment epithelium-derived factor (PEDF), a known
regulator of cell differentiation, is also a very potent anti-angio-
genic factor15. Such results indicate that a variety of molecules,
whose structure places them in the serpin family can be impor-
tant regulators of natural tumor growth through their influence
on neovascularization.
Methods
Cell culture. The human prostate carcinoma cell line LNCaP from
American Type Culture Collection (ATCC; Manassas, Virginia) was grown
in RPMI 1640 with 10% fetal bovine serum. MDA-MB-435 cells were
maintained in DMEM (Life Technologies) supplemented with 10% fetal
calf serum. Normal human foreskin fibroblasts (HFF-S1) were established
in the laboratory by S. Tolsma (Northwester College, Orange City, Iowa)
and maintained in DME (Life Technologies) supplemented with 10% fetal
calf serum. Normal human keratinocytes (NHOK) were a gift from
M.Lingen (Loyola University Medical School, Maywood, Illinois) and
maintained in keratinocyte growth medium (Clonetics Cell Systems, San
Diego, California) with supplements recommended by the manufacturer.
Bovine adrenal capillary endothelial cells, a gift from J. Folkman
(Children’s Hospital, Harvard Medical School, Boston, Massachusetts
02115) were maintained in DMEM supplemented with 10% calf serum
and 100 µg/ml endothelial cell mitogen, and were used at passage 15.
Human dermal microvascular endothelial cells (HMVEC, passage 9) were
from Clonetics Cell Systems and maintained in endothelial cell basal
medium with 5% fetal bovine serum and an EC (endothelial cell) ‘bullet’
kit as recommended by supplier.
Protein production and purification. The pGST-maspin and maspin∆RSL
vectors were constructed as described7. Maspin∆N was generated by di-
gestion of the pGST-maspin vector with BamH1, which was blunt ended
and digested again with Sst1. The adhesive ends were filled by T4 poly-
merase and ligated to remove the N-terminal 139 amino acids. To con-
struct pmaspin∗, oligonucleotides encoding a mutation in the P1 residue
were generated (sense primer 5′–GGGTCCCAGATCTTA–3′and antisense
primer 5′– TAAGATCTGGGACCC–3′). Site-directed mutagenesis accom-
plished using the pGST-maspin vector and the above oligos (Stratagene,
La Jolla, California). All mutant constructs were sequenced to ensure the
in-frame fusion. The GST fusion proteins were produced as described7.
The size and purity of recombinant maspin and mutants were confirmed
by SDS-PAGE gel electrophoresis and western blot analysis using a poly-
clonal AbS4A antibody7.
Endothelial cell assays. Migrations were done using bovine adrenal cap-
illary endothelial cells as described16. Cells were starved overnight in DME
containing 0.1% bovine serum albumin (BSA), collected, re-suspended in
DME with 0.1% BSA, and plated at a concentration of 3 ×104cells/well
on the lower surface of a gelatinized 5.0 µm filter (Nucleopore,
Pleasanton, California) in an inverted, modified Boyden chamber. Cells
were allowed to adhere for 2 h at 37 °C, the chambers were re-inverted,
test samples were added to the top wells and the chambers incubated 4 h
at 37 °C to allow migration. Chambers were then disassembled, mem-
branes fixed and stained and the number of cells that had migrated to the
top of the filter in 10 high-power fields counted (a high power field is
×1000). DME supplemented with 0.1% BSA was used as a negative con-
trol to measure background resulting from random migration. All samples
Fig. 3 Decreased tumor vessels after long-term treatment with exogenous
maspin protein. Tumors were collected between 7 and 8 weeks from GST-
treated mice (aand c) and from maspin-treated mice (band d) and fixed
and stained with hematoxylin and eosin (a and b) or with antibody against
CD31 (c and d).
Table 1 Analysis of tumor volume and vessel number in mice treated with
GST or GST–maspin
GST GST-maspin P
Short Average of 28.6 ± 3.6 15.3 ± 1.8 0.001
term microvessel densityan=10 n=10
Total tumor sites 37 32
Percentage of complete 2.7% (1/37) 53% (17/32) 0.0001
Long inhibitionb
term
Average of tumor 116.3 ± 26.1, n=36 33.9 ± 6.1, n=15 0.027
volume (mm3) of sites (12.1–600) (9.0–87.5)
with tumors
Average of MVD 83.0 ±10, n=10 31.4 ±1.8, n=10 0.002
(38–143) (21–42)
aCollected from three ‘hot fields’ (400×) in each tumor. bComplete inhibition indicates no sign of tumor or
tumor size smaller than 2 mm in diameter after treatment. ±, standard error.
a b
cd
© 2000 Nature America Inc. • http://medicine.nature.com
© 2000 Nature America Inc. • http://medicine.nature.com
NATURE MEDICINE • VOLUME 6 • NUMBER 2 • FEBRUARY 2000 199
ARTICLES
were tested in quadruplicate.
To assess mitogenesis, HMVECs were plated at a concentration of 1 ×
104cells per well in gelatinized, 96-well microtiter plates, allowed to attach
for 24 h and re-fed with endothelial cell basal medium (Life Technologies,
New York) supplemented with 2% serum and GST or GST–maspin fusion
protein and mutants at a concentration of 1 µM where indicated. The cells
were incubated at 37 °C, with 5%CO2for 5 days to allow at least one pop-
ulation doubling. Mitogenesis was assessed using CellTiter nonradioactive
proliferation assay (Promega) according to manufacturer’s instructions.
Baseline (no mitogen) reflects the value, determined before population
doubling, 24 h after seeding.
Migration assays on non-endothelial cells. Breast tumor migration as-
says (MDA-MB-435) were done in quadruplicate as described7. After sub-
traction of background, data were normalized to give percent inhibition
equating GST treated samples to 100% motility. For migration assays on
fibroblasts and keratinocytes, HFF-S1 (0.7 ×106cells/ml) and NHOK (1.2 ×
106cells/ml) were plated in serum-free basal media supplemented with
0.1%BSA on the bottom side of the microporous membrane (8-µm pore
size) in the inverted modified Boyden chamber. The cells were allowed to
attach for 1.5 h, the chambers were re-inverted and test substances in ap-
propriate serum-free basal medium were added to the other side of the
membrane. The migration was assessed, background was subtracted and
data were reported as percent inhibition, as in tumor cell invasion assay,
with bFGF induced motility normalized as 100%.
Corneal neovascularization assay. The assay was done as previously de-
scribed16. Hydron pellets containing bFGF (100 ng/ml), GST (10 µM), GST-
maspin (10 µM) or mutants (10 µM), alone in combination with bFGF
were implanted into a pocket surgically created in avascular corneas of
anesthetized female rats (Fisher 344: Harlan, Indianapolis, Indiana) 1–1.5
mm from the limbus. The compounds were used at concentrations at least
10-fold higher than in migration assay to account for the diffusion rate
from a slow release pellet. Neovascularization was observed on day 6 or 7
after implantation. Vigorous growth of the blood vessels in the direction of
the pellet was noted as a positive response. The animals were perfused
with colloidal carbon, eyes were removed and fixed, corneas excised, flat-
tened and photographed for a permanent record.
Tumor angiogenesis assay. LNCaP tumor cells were grown to 80% con-
fluence, collected and resuspended in sterile HEPES-buffered salt solution
at a concentration of 4 ×10 7cells/ml. This cell suspension was then mixed
with Matrigel (Collaborative Research, Bedford Massachusetts) at a 1:3
ratio and added in 100-µl aliquotes into Eppendorf tubes containing
maspin or GST, mixed on ice for 10 min and subsequently injected subcul-
taneously into the dorsal back of 5-week-old male athymic mice (K & K
Universal, Freemont, California). Each mouse was inoculated at two to four
sites. Ten mice were used for the initial vessel formation assay and 20 mice,
for the tumor inhibition study. Primary tumors at the site of the subcuta-
neous injection were measured using calipers, and tumor volume was cal-
culated according to the algorithm: length x [width]2×0.5 (ref. 17).
Ethylene/vinyl acetate copolymer (Evac; NEN) slow release pellets, con-
taining maspin and GST, were prepared as described18,19. A lyophilized
mixture of BSA and varied amounts of recombinant proteins was dispersed
in 0.125 ml Evac dissolved in dichloromethane (Sigma). This mixture was
frozen, dried, and cut into pellets of appropriate size. Each Evac pellet con-
tained about 210 µg GST-maspin or 70 µg GST. The follow-up treatment
of tumors was carried out by subcutaneous implantation of the pellet at
days 15 and 30 within 0.3 cm of the tumor site and the incision sealed
after implantation. At the end of each experiment (7–8 weeks), the tumors
were measured, excised and fixed in 10% neutral buffered formalin.
Samples were then embedded and sectioned to 5 µm for histology (hema-
toxylin and eosin) and immunohistochemistry. For immunohistochem-
istry, monoclonal antibodies to CD 31 (Pharmingen, Sandiego, California)
was used at 1:50 dilution at 4 °C overnight. After rinsing, slides were incu-
bated with the secondary goat antibody to rat at a dilution of 1:100 for 1 h
at room temperature. Slides were then rinsed and incubated with an
avidin–biotin–peroxidase complex (ABC kit; Vector Laboratories,
Burlingame, California), followed by DNBA (3, 3′-Diaminobenzidine
Tetrahydrochloride) color development. Vessels were counted as described
20 by first scanning the sections at the low power for hot spots or high vas-
cular density (×40), and then counting the areas of microvessels at high
power (×400). Microvessel density was calculated by adding the numbers
from three hot spot fields with a ×400 objective.
Acknowledgments
The authors thank J. Rosen for his advice and equipment support, N.
Greenberg, D. Rowley, and W. Porter for discussion, W. Huss for kindly provid-
ing the protocol for CD31 and factor VIII immunostaining. This work is
supported by National Cancer Institute grants CA 52750 and CA 64239 to
N.B. and a Department of Defense grant (DAMD17-98-1-8028) to M.Z.
RECEIVED 7 SEPTEMBER; ACCEPTED 1 NOVEMBER 1999
1. Zou, Z. et al. Maspin, a serpin with tumor-suppressing activity in human mammary
epithelial cells [see comments]. Science 263, 526–529 (1994).
2. Zhang, M., Maass, N., Magit, D. & Sager, R. Transactivation through Ets and Ap1
transcription sites determines the expression of the tumor-suppressing gene
maspin. Cell Growth Differ. 8, 179–186 (1997).
3. Pemberton, P.A. et al. The tumor suppressor maspin does not undergo the stressed
to relaxed transition or inhibit trypsin-like serine proteases. Evidence that maspin is
not a protease inhibitory serpin. J. Biol. Chem. 270, 15832–15837 (1995).
4. Sheng, S. et al. Tissue-type plasminogen activator is a target of the tumor suppres-
sor gene maspin. Proc. Natl. Acad. Sci. U. S. A. 95, 499–504 (1998).
5. Zhang, M. et al. Maspin plays an important role in mammary gland development.
Dev. Biol. 215, 278–287 (1999).
6. Stringer, H.A. & Pannekoek, H. The significance of fibrin binding by plasminogen
activator inhibitor 1 for the mechanism of tissue-type plasminogen activator-medi-
ated fibrinolysis. J. Biol. Chem. 270, 11205–11208 (1995).
7. Zhang, M., Sheng, S., Maass, N. & Sager, R. mMaspin: the mouse homolog of a
human tumor suppressor gene inhibits mammary tumor invasion and motility. Mol.
Med. 3, 49–59 (1997).
8. Sheng, S., Pemberton, P.A. & Sager, R. Production, purification, and characteriza-
tion of recombinant maspin proteins. J. Biol. Chem. 269, 30988–30993 (1994).
9. Dawson, D.W. et al. CD36 mediates the In vitro inhibitory effects of throm-
bospondin-1 on endothelial cells. J. Cell Biol. 138, 707–717 (1997).
10. Lawrence, D.A., Berkenpas, M.B., Palaniappan, S. & Ginsburg, D. Localization of
vitronectin binding domain in plasminogen activator inhibitor-1. J. Biol. Chem. 269,
15223–15228 (1994).
11. Gately, S. et al. Human prostate carcinoma cells express enzymatic activity that
converts human plasminogen to the angiogenesis inhibitor, angiostatin. Cancer
Res. 56, 4887–4890 (1996).
12. Volpert, O.V. et al. Captopril inhibits angiogenesis and slows the growth of exper-
imental tumors in rats [see comments]. J. Clin. Invest. 98, 671–679 (1996).
13. Sheng, S. et al. Maspin acts at the cell membrane to inhibit invasion and motility of
mammary and prostatic cancer cells. Proc. Natl. Acad. Sci. U. S. A. 93, 11669–11674
(1996).
14. Sternlicht, M.D., Kedeshian, P., Shao, Z.M., Safarians, S. & Barsky, S.H. The human
myoepithelial cell is a natural tumor suppressor. Clin. Cancer Res. 3, 1949–1958
(1997).
15. Dawson, D.W. et al. Pigment Epithelium-Derived Factor: A Potent Inhibitor of
Angiogenesis. Science 285, 245–248 (1999).
16. Polverini, P.J., Bouck, N.P. & Rastinejad, F. Assay and purification of naturally occur-
ring inhibitor of angiogenesis. Methods Enzymol. 198, 440–450 (1991).
17. Tsujii, M. et al. Cyclooxygenase regulates angiogenesis induced by colon cancer
cells [published erratum appears in Cell 1998 Jul 24;94(2):following 271]. Cell 93,
705–716 (1998).
18. Silberstein, G.B. & Daniel, C.W. Investigation of mouse mammary ductal growth
regulation using slow- release plastic implants. J. Dairy Sci. 70, 1981–1990 (1987).
19. Talhouk, R.S., Bissell, M.J. & Werb, Z. Coordinated expression of extracellular ma-
trix-degrading proteinases and their inhibitors regulates mammary epithelial func-
tion during involution. J. Cell Biol. 118, 1271–1282 (1992).
20. Obermair, A. et al. Correlation of vascular endothelial growth factor expression and
microvessel density in cervical intraepithelial neoplasia. J. Natl. Cancer Inst. 89,
1212–1217 (1997).
© 2000 Nature America Inc. • http://medicine.nature.com
© 2000 Nature America Inc. • http://medicine.nature.com