Knobloch syndrome (MIM# 267750) is an autosomal recessive disease mainly characterized by myopia magna (>20D), vitreoretinal degeneration, and congenital occipital encephalocele (Knobloch and Layer, 1971). High myopia has been diagnosed at birth and vitreoretinal degeneration as early as 3 days after birth (Sertié et al., 2000; Suzuki et al., 2002). Early recurrent retinal detachment, early cataract formation, and anterior segment disease including anterior iritis with posterior synechiae are further features of the eye defect in Knobloch syndrome. While the encephalocele can be corrected surgically within the first year of life, the ocular pathology is progressive in nature and shows poor response to surgical treatment. Patients experience a bilateral loss of vision during childhood.
Knobloch syndrome is caused by homozygous or compound heterozygous mutations in the COL18A1 gene on 21q22.3 (Sertié et al., 2000; Suzuki et al., 2002; Kliemann et al., 2003; Menzel et al., 2004). Collagen XV and collagen XVIII are nonfibrillar collagens characterized by multiple interruptions in their central triple helical region and a unique highly homologous C-terminal domain (Oh et al., 1994). The 20-kDa C-terminal fragment of collagen XVIII can be proteolytically released (Wen et al., 1999; Felbor et al., 2000; Ferreras et al., 2000). It is a potent inhibitor of angiogenesis and tumor growth and was designated endostatin (O'Reilly et al., 1997). The crystal structure of endostatin predicted a prominent heparin/heparan sulfate binding site around arginines 158 and 270. Because heparan sulfate affinity is a feature of many growth factors such as vascular endothelial growth factor (VEGF), it was speculated that endostatin inhibits angiogenesis by binding to heparan sulfate containing proteoglycans on cell membrane surfaces (Hohenester et al., 1998). Subsequently, research has focussed on the characterization of endostatin's interaction with different heparin/heparan sulfate molecules (Sasaki et al., 1999; Dixelius et al., 2000; Karumanchi et al., 2001; Kreuger et al., 2002; Ricard-Blum et al., 2004). However, it remains unclear whether a high affinity endostatin receptor exists and whether heparan sulfate proteoglycans function as coreceptors that are required for local concentration of endostatin and mediate efficient binding of endostatin to a putative high affinity receptor. In addition, the physiological role of endostatin within collagen XVIII remains unknown.
Endostatin is encoded by exons 41 to 43 of the COL18A1 gene (Rehn et al., 1996). Six of nine known independently arisen COL18A1 mutations are located in exons 35, 36, 40, and 41 and lead to a premature stop just before or within the C-terminal endostatin coding region (Suzuki et al., 2002; Menzel et al., 2004). Furthermore, two missense mutations were recently located in the endostatin domain (Kliemann et al., 2003; Menzel et al., 2004). One of these mutations was shown to result in decreased affinity for laminin (Menzel et al., 2004). Thus, it is the most C-terminal endostatin domain that is either specifically deleted, interrupted or functionally altered in these Knobloch patients. It is conceivable that the endostatin domain is required for proper protein folding or protein stability of collagen XVIII. However, a stable truncated protein was isolated after deletion of the entire noncollagenous C-terminal domain of the Caenorhabditis elegans collagen XVIII homologue cle-1 (Ackley et al., 2001). Alternatively, it is possible that binding of collagen XVIII to blood vessels and specific basement membranes is mediated by the endostatin domain. The latter alternative is supported by the observation that the binding pattern of hemagglutinin-tagged murine endostatin on adult human tissue sections closely resembled the distribution of its parent molecule collagen XVIII (Chang et al., 1999).
Histopathologic evaluation of a single enucleated eye from a Knobloch patient was not informative because it only showed unspecific late-stage gliotic changes of the detached retina (Cook and Knobloch, 1982). Thus, mouse models of Knobloch syndrome are important to elucidate early pathogenetic mechanisms. Collagen XVIII knockout mice show delayed regression of the hyaloid arteries and abnormal outgrowth of retinal vessels during the first 3 weeks of life (Fukai et al., 2002). They also develop age-dependent abnormalities in the iris, the ciliary body, and the retina (Marneros and Olsen, 2003; Ylikärppä et al., 2003a; Marneros et al., 2004) in agreement with altered endostatin immunoreactivity in retinae from patients affected with age-related macular degeneration (Bhutto et al., 2004).
The early onset of severe eye disease in Knobloch syndrome suggests that irreversible ocular damage occurs in utero. We localized and characterized endostatin binding partners in murine embryonal development. A systematic screen using alkaline phosphatase (AP) fusion proteins (Flanagan et al., 2000) identified vascular mesenchyme in the developing eye as endostatin's primary target, whereas the VEGF164 affinity probe prominently bound to nonvascular nerve tissues. Alanine in vitro mutagenesis of the heparin/heparan sulfate binding site within endostatin resulted in reduced binding affinity while elimination of the heparan sulfate binding site in VEGF led to complete loss of binding. Finally, inclusion of the non–heparan sulfate-binding collagen XV endostatin analogue into the screen demonstrated a striking spatial restriction and reduction in staining intensity when compared with endostatin.
Generation of AP–Endostatin and AP–VEGF Fusion Proteins
To determine the expression profile of binding partners for endostatin in murine development, five dimeric AP fusion proteins (Fig. 1A) were designed that are expected to produce a pair of ligand moieties facing away from the tag in the same direction. Because endostatin is derived from the C-terminal part of collagen XVIII, murine endostatin (mES) was fused to the C-terminus of heat-stable secreted human placental AP (AP–mES). To reduce nonspecific binding of endostatin to cell surfaces, two arginine residues responsible for heparan sulfate binding (Sasaki et al., 1999) were mutated to alanine by in vitro mutagenesis (AP–mESR158/270A). The highly homologous C-terminal fragment of collagen XV (endostatin-XV) lacks the evolutionarily conserved arginine at position 158 and does not bind to heparin. Whereas endostatin-XV showed anti-angiogenic activity in the chick chorioallantoic membrane angiogenesis assay (Sasaki et al., 2000), it did not inhibit outgrowth of endothelial cells from murine fetal bone explants (Gaetzner et al., in press). To further characterize the role of endostatin-XV, it was included in the present screen (AP–mESXV). Because endostatin can inhibit angiogenesis stimulated by VEGF (Yamaguchi et al., 1999; Kreuger et al., 2002), the murine 164- and 110-forms of VEGF were also fused to AP by means of a proline rich linker (AP–mVEGF164, AP–mVEGF110). Murine VEGF110 corresponds to a plasmin-generated fragment of human VEGF165 (Keyt et al., 1996). Like alternatively spliced VEGF121, it contains the N-terminal receptor binding determinants but lacks the C-terminal heparan sulfate binding domain and, therefore, is freely diffusable. Finally, pAPtag-4 was used to produce unfused AP as a negative control.
AP fusion proteins were produced in the extracellular supernatants of transiently transfected 293T cells. Expression of AP fusion proteins was monitored by quantitative measurement of AP activity in conditioned supernatants of transfected and nontransfected 293T cells. All supernatants of transfected cells demonstrated high and comparable AP activity, whereas the control supernatant of nontransfected cells showed only background activity (data not shown). Furthermore, expression of fusion proteins was assayed by immunoblotting with a polyclonal antibody against secreted human placental alkaline phosphatase. This antibody specifically detected AP fusion proteins of the expected sizes in supernatants of transfected but not of nontransfected cells (Fig. 1B). Ten-μl conditioned supernatants contained approximately 20 ng of fusion proteins (data not shown).
Differential In Situ Binding of Endostatins and VEGFs
The penetration of AP fusion proteins into unfixed whole-mount preparations of murine embryos was limited (data not shown). Therefore, fixed embryos were serial-sectioned in sagittal and transverse orientation. AP staining of midline sagittal sections with AP–mES (Fig. 2) demonstrated that endostatin bound to blood vessels in all organs examined. Nerve tissues in brain, spinal cord, and spinal ganglia did not stain specifically, whereas blood vessels of all sizes clearly did. Similarly, AP–mES detected blood vessels in adult murine brains (Fig. 3). As evidenced by strong staining of endothelial cells in small capillaries, endostatin not only bound to basement membranes and elastic fibers in vessel walls to epithelial basement membranes in a variety of organs (Fig. 2).
Incubation of sagittal sections with AP–mESR158/270A, which lacks the prominent heparan sulfate binding site resulted in a similar staining pattern. However, the staining required longer incubation times with the nitroblue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate (NBT/BCIP) substrate (Fig. 2). Incubation with AP–mESXV and AP (Fig. 2) yielded no specific staining in midline sagittal sections. Unlike endostatin, AP–mVEGF164 staining was not predominantly vessel-associated in the embryonic day (E) 12.5–E14.5 murine tissues analyzed. Staining was prominent in liver tissue with active hematopoiesis but was also seen in forebrain, hindbrain, and basement membranes (Fig. 2). The elimination of the C-terminal heparan sulfate binding site in AP–mVEGF110 produced a perfect negative control (Fig. 2) akin to AP staining alone. That no general background staining was seen with AP, AP–mESXV, and AP–mVEGF110 and only very little with AP–mESR158/270A suggests that the background coloration observed with AP–mES and AP–mVEGF164 might be due to their high affinity for heparan sulfates.
Endostatin Binds to Blood Vessels in the Absence of Endostatin/Collagen XVIII
To rule out the possibility that endogenous endostatin interferes with endostatin binding in situ, collagen XVIII knockout embryos were incubated with AP fusion protein-containing supernatants. Prominent labeling of blood vessels and the pia mater in collagen XVIII null embryos demonstrated that endostatin binding in situ is not due to oligomerization with endogenous endostatin (Fig. 4A). In addition, the differential expression pattern of endostatin and VEGF binding partners was further confirmed because AP–mVEGF164 strongly stained forebrain and hindbrain in these sections (Fig. 4B). AP alone produced no background staining (Fig. 4C).
Non–heparin/Heparan Sulfate-Binding Endostatin Interacts Strongly With Mesenchymal Structures in the Murine Embryonal Eye
Staining of E12.5–E14.5 transverse sections with AP–mESR158/270A gave prominent signals in the hyaloid cavity between the lens and the neural layer of the retina, in the inner limiting membrane, the tunica vasculosa lentis, and the iridopupillary membrane. In addition, AP–mESR158/270A bound to vessels in the choroid (Fig. 5A,E) and in the developing E17.5 ciliary body (data not shown). Whereas incubation with AP–mESR158/270A led to little background coloration even upon overnight incubation with the NBT/BCIP substrate solution, AP–mES showed intense background staining if incubated overnight (Fig. 5B). Thus, elimination of the heparan sulfate–binding epitope resulted in significantly more specific endostatin staining patterns. The only structures that stained with AP–mESXV in our tissue collection were the tunica vasculosa lentis and the iridopupillary membrane (Fig. 5C,F). AP alone showed no specific staining (Fig. 5D).
While incubation with the NBT/BCIP substrate only takes 5 to 10 min to elicit a strong signal, weaker signals are known to take many hours to develop (Flanagan et al., 2000) most likely due to low receptor number resulting in a low amount of bound enzyme. Thus, experiments with varying substrate incubation times were performed (Fig. 6). A reduction of the incubation time with AP–mES fusion protein from 90 to 5 min did not lead to a relevant attenuation of the color reaction, indicating saturation of binding partners (data not shown). The use of higher AP fusion protein concentrations, therefore, was unnecessary.
AP–mESR158/270A labeled ocular structures rapidly. As early as 5 min after the addition of the substrate solution, the eyes showed a distinctive staining pattern (Fig. 6A), whereas blood vessels and basement membranes in other organs required significantly longer incubation times. With AP–mESXV only a faint signal was detected after 1 hr. However, upon overnight staining, a highly specific labeling of the tunica vasculosa lentis was observed (Fig. 6B). Both heparin-binding AP–mES and AP–mVEGF164 led to background staining within less than 1 hr incubation with the substrate solution (Fig. 6C,D). AP–mVEGF110 and AP were negative even upon overnight incubation (Fig. 6E,F), indicating that there is no general background staining with our non–heparin-binding AP fusion proteins. Taken together, the heparan sulfate–binding sites within AP–mES and AP–mVEGF164 increase in situ binding at the expense of staining specificity.
In agreement with the observation that severe ocular alterations are present at birth in Knobloch syndrome, endostatin strongly bound to ocular mesenchyme during murine eye development. Prominent binding was observed within the hyaloid cavity, the future vitreous body, containing the vasa hyaloidea. These vessels supply the lens and the neural retina during embryonal life and regress after birth. In both wild-type and collagen XVIII knockout mice, the endostatin binding pattern (Figs. 5A,E, 6A) strikingly colocalized with the distribution of its parent molecule collagen XVIII (Sasaki et al., 1998) and defects in collagen XVIII knockout mice (Fukai et al., 2002; Marneros and Olsen, 2003; Ylikärppä et al., 2003a; Marneros et al., 2004). Similarly, the collagen XV endostatin homologue showed marked binding to the tunica vasculosa lentis (Figs. 5C,F, 6B), a structure immunopositive with an antibody against collagen XV and defective in collagen XV knockout mice (Ylikärppä et al., 2003b). Taken together with the observation that intact full-length collagen XVIII was localized within ocular basement membranes (Marneros et al., 2004), we conclude that an important tissue-binding site resides in collagen XVIII's C-terminal endostatin domain and that only a properly functioning endostatin domain guarantees correct binding to basement membranes. To further analyze the critical role of endostatin for tissue binding, it will be interesting to see whether mice lacking the endostatin domain will reveal improper localization of collagen XVIII within ocular basement membranes.
That the collagen XV endostatin homologue failed to bind to E12.5–E14.5 capillaries and epithelial basement membranes, the inner limiting membrane of the retina, the iris, and the ciliary body is consistent with a more restricted expression pattern of collagen XV during development and developmental shifts in the expression of collagen XV (Muona et al., 2002). Our data, therefore, emphasize the critical role of collagen XVIII in the eye, which cannot be compensated by collagen XV. They also provide an explanation for the lack of anti-angiogenic activity of endostatin-XV in a murine fetal bone explant angiogenesis model (Gaetzner et al., in press). However, the partially overlapping staining pattern in the eye suggests that the non–heparan sulfate receptor for endostatin-XV and -XVIII could be identical.
VEGF and endostatin both share an affinity for heparan sulfates, and the angiogenic activity of VEGF can be antagonized by endostatin in vitro (Yamaguchi et al., 1999; Kreuger et al., 2002). Therefore, we compared their binding patterns in situ. Endostatin binding was predominantly confined to the vasculature in the murine embryonic nervous system. Thus, the present results do not corroborate a role for endostatin in neuronal development as had been suggested by high expression of the Caenorhabditis elegans collagen XVIII homologue cle-1 in neurons of the avascular nematode (Ackley et al., 2001). In contrast, the AP–mVEGF164 probe labeled forebrain in agreement with the expression of the VEGF receptor Flk-1 in E15 mouse forebrain neurons (Ogunshola et al., 2002). Distinct labeling of further nonvascular tissues such as the optic nerve and the surface ectoderm of the future cornea (Fig. 6D) extends the recent observation that VEGF receptors are found in developing neural retina (Robinson et al., 2001). Prominent staining of E14.5 liver (Fig. 2) coincides with the contribution of Flk-1 to definitive hematopoiesis during fetal development (Shalaby et al., 1997). The lack of VEGF staining in developing heart and lung can be explained with the observation that VEGF mRNA levels increase only in late embryonic stages in these organs (Ng et al., 2001). Both endostatin and VEGF binding colocalized in the hyaloid cavity of the eye. However, as opposed to endostatin, VEGF in situ binding depended on heparan sulfates, because the VEGF110 deletion mutant that contains the receptor binding sites but lacks affinity for heparan sulfates was negative (Fig. 6F). This observation is in agreement with a complete inhibition of VEGF165 binding to endothelial cells after pretreatment with heparinase (Rathjen et al., 1990) and a more than 100-fold decrease in endothelial cell mitogenic activity of VEGF110 and VEGF121 compared with VEGF165 (Keyt et al., 1996).
While RNA in situ hybridization and immunolocalization reveal highly specific molecular distributions in vertebrate embryos, the application of ligand fusion protein probes provides qualitatively different additional information, because novel binding partners can be mapped and characterized. This study addresses the controversial issue of endostatin's affinity for cell membrane heparan sulfates by use of a non–heparan sulfate-binding mutant AP–endostatin fusion protein probe in situ. Thus far, three studies implicated that endostatin binding (Chang et al., 1999; Wickström et al., 2003) and activity (Yamaguchi et al., 1999) are independent of heparan sulfates. In contrast, other studies showed that the recombinant non–heparan sulfate-binding mutant endostatin R158/270A protein is associated with reduced or lost biological activity in vitro (Sasaki et al., 1999; Dixelius et al., 2000; Javaherian et al., 2002; Schmidt et al., 2004) and that the inhibitory effect of wild-type endostatin on endothelial cell migration can be neutralized by heparin 12mers (Kreuger et al., 2002). Timed in situ-staining demonstrated that the AP–mESR158/270A probe specifically and rapidly bound to vascular mesenchyme in the murine embryonal eye (Figs. 5, 6). With lower affinity, it also labeled blood vessels in brain and other organs (Fig. 2). This finding indicates that endostatin interacts with a putative non–heparan sulfate high affinity receptor in situ, which is highly expressed in embryonal ocular mesenchyme. Based on our staining data, a cell line derived from embryonal hyaloid arteries would be ideal for expression cloning of this high affinity binding partner but is currently not available. Other vasculatures may contain the same receptor in lower density requiring heparan sulfates to, e.g., increase receptor binding. Alternatively, they may have different binding molecules. A variety of low affinity basement membrane binding partners such as fibulin-1, fibulin-2, laminin-1, and perlecan have been described (Sasaki et al., 1998). However, competition experiments with excess exogenous murine laminin-1 did not lead to altered endostatin binding in situ although AP–mES bound to laminin in solid phase assays (data not shown). This finding suggests that laminin is not an indispensable binding partner for endostatin in situ. In conclusion, it is conceivable that heparan sulfate binding is not essential for binding and proper anchoring of a structurally functioning collagen XVIII within ocular basement membranes, although it increases the biological activity of the proteolytic cleavage product endostatin in various angiogenesis assays.
Construction of Expression Vectors
Murine α1(XVIII) cDNA clone mc3b (Oh et al., 1994), murine α1(XV) cDNA clone mTS1-3 (Shinya and Olsen, unpublished), plasmid pSS9/TVA-VEGF164, and plasmid pSS10/TVA-VEGF110 (Snitkovsky et al., 2001) were used as templates to amplify the sequences of murine endostatin, murine endostatin XV, murine VEGF164, and murine VEGF110 with Pwo DNA polymerase according to the manufacturer's protocol (Roche). The following oligonucleotide primers were used: AP–murine endostatin: AP–mES5′, 5′-GGTTCCGGACATACTCATCAGGACTTTCAGCC-3′, AP–mES3′, 5′-GGAAGATCTCTATTTGGAGAAAGAGGTCATGAAG-3′; AP–murine endostatin XV: AP–mESXV5′, 5′-GGTCCGGATATGAGAGGCCTGTTCTGCACC-3′, AP–mESXV3′, 5′-GGAAGATCTTCACTTCCTAGTGTCTGTCATG-3′; AP–murine VEGFs: AP–Linkerfor, 5′- GGTTCCGGACCACCACCTGAACTCCTAGG-3′, AP–mVEGF164rev, 5′-ATGCTCGAGTCACCGCCTTGGCTTGTCAC-3′, AP–mVEGF110rev, 5′-ATGCTCGAGTCATGTTCTGTCTTTCTTTGGTC-3′.
In addition to the annealing sequence, forward primers contained a BspEI site at the 5′ end (AP–mES5′, AP–mESXV5′, AP–Linkerfor) and reverse primers contained a stop codon followed by a BglII (AP–mES3′, AP–mESXV3′) or a XhoI site (AP–mVEGF164rev, AP–mVEGF110rev) at the 3′ end. A DNA fragment encoding a proline-rich linker, PPPELLGGP, derived from the hinge region of a rabbit Fc chain had been placed at the N-terminal end of the coding sequences for murine VEGFs (Snitkovsky et al., 2001). After restriction and subcloning into the expression vector pAPtag-4 (obtained from D.A. Feldheim and J.G. Flanagan), the sequences of all inserts were confirmed by using Dye Terminator Cycle Sequencing (ABI). For in vitro mutagenesis with the QuikChange Site-Directed Mutagenesis kit (Stratagene), a PCR product encoding murine endostatin was first subcloned into the EcoRV and NotI restriction sites of pBluescript II SK(+) (Stratagene). After in vitro mutagenesis, it was PCR-cloned into pAPtag-4 as described above.
Expression of Recombinant Proteins
FuGENE 6 transfection reagent (Roche) was used to transiently transfect 12 μg of fusion plasmid DNA into 293T cells (human embryonic kidney cells, a gift from D.A. Feldheim and J.G. Flanagan) plated at 80% confluency on 150-mm tissue culture plates. The medium (DMEM with glutamax-I (GIBCO), 10% fetal bovine serum, 1% penicillin–streptomycin) was replaced 24 hr after transfection. Conditioned supernatants from transfected and nontransfected cells were collected after an additional 48–72 hr; centrifuged at 14,000 rpm (Eppendorf rotor A-4-44); filtered through a 0.45-μm filter (Schleicher & Schuell); buffered with 10 mM HEPES, 0.05% NaN3, pH 7.0; and stored at 4°C for immediate use or at −80°C for long-term usage.
Sodium Dodecyl Sulfate-Polyacrylamide Gel Electrophoresis/Western Blotting
A total of 15 μl of conditioned supernatants were loaded onto 5–15% sodium dodecyl sulfate-polyacrylamide gel electrophoresis gradient gels, run at 50 V for 18 hr, and wet-blotted onto nitrocellulose (Protran; Schleicher & Schuell) by using a TE42 Transphor transfer unit (Amersham Biosciences; 1.5 A for 1 hr at 4°C), because the semi-dry transfer was insufficient. Fusion proteins were detected with a rabbit polyclonal antibody against secreted human placental alkaline phosphatase (1:2,000, WAK-Chemie) followed by a horseradish peroxidase–conjugated anti-rabbit IgG (1:3,000, Santa Cruz) and enhanced chemiluminescence (Perkin Elmer/NEN). AP–mES and AP–mESR158/270A fusion proteins were quantified by Western blot analyses with known amounts of recombinant murine endostatin using a rabbit polyclonal anti-mouse endostatin antibody obtained from Cytimmune Sciences, Inc.
AP Activity Assay
A total of 100 μl unconcentrated supernatants were heat-inactivated for 10 min at 65°C to inhibit endogenous phosphatase activity. After centrifugation at 14,000 rpm (Eppendorf rotor F45-30-11), 20 μl were mixed with 380 μl HBAH buffer (see below) and 400 μl 2× AP substrate buffer (2 M diethanolamine, 1 mM MgCl2, 18 mM p-nitrophenyl phosphate [AppliChem], pH 9.8), and incubated at room temperature. Absorbance at 405 nm was read at 30-sec intervals for 10 min in a spectrophotometer.
Staining of Tissue Sections
Timed-mated NMRI mice were ordered from Harlan Winkelmann. E12.5, E13.5, E14.5, and E17.5 embryos were dissected, fixed in 4% paraformaldehyde (in PBS) at 4°C overnight (E17.5 mice were decapitated and the head was fixed for 2 days; adult mice were perfusion-fixed with 10% neutral buffered formalin), transferred to 20% sucrose (in PBS) at 4°C on a shaker for 1 day, and frozen in OCT embedding medium (Tissue-Tek). Subsequently, E12.5–E14.5 mice were serial-sectioned in sagittal and transverse orientation, thaw-mounted on Polysine slides (Menzel Gläser), and stored at −80°C. For orientation, every 10th section was stained with methylene blue. AP staining of 10-μm cryosections was essentially performed as described in Flanagan et al. (2000). In brief, cryosections were thawed at 37°C for 5 min, washed in HBSS (150 mM NaCl, 20 mM HEPES, pH 7.0) for 10 min and in HBAH (HBSS, 0.5 mg/ml bovine serum albumin, 0.1% NaN3) for 5 min twice. Then, tissues were covered with supernatants containing AP fusion proteins and incubated at room temperature for 90 min. Afterward, the sections were rinsed in ice-cold HBAH 6× for 2 min, fixed in acetone–formalin for 15 sec, washed in HBSS for 5 min twice, placed into preheated HBSS, and incubated in a 65°C water bath for 10 min to heat-inactivate endogenous phosphatase. Then, sections were washed in AP-staining buffer (100 mM NaCl, 5 mM MgCl2, 100 mM Tris-HCl, pH 9.5) for 5 min and incubated with NBT/BCIP substrate (Roche) at room temperature under a shade of aluminium foil. Reactions were monitored under a dissecting microscope and stopped with 10 mM ethylenediaminetetraacetic acid (in PBS). Sections were fixed in 10% neutral buffered formalin (Sigma), and mounted with Kaiser's glycerin gelatin (Merck). Image acquisition was with a SZX9 stereomicroscope (Olympus) and the Spot Insight QE Color imaging software (Visitron). Higher magnifications were taken with a Zeiss Axiophot.
Collagen XVIII Null Mice
Mice with targeted disruption of COL18A1 were a gift from Drs. N. Fukai and B.R. Olsen. Genotypes were determined by PCR amplification as described (Fukai et al., 2002).
Drs. N. Fukai and B.R. Olsen are thanked for providing collagen XVIII knockout mice, Drs. D.A. Feldheim and J.G. Flanagan for the expression vector pAPtag-4, Drs. S. Snitkovsky and J.A.T. Young for plasmids pSS9/TVA-VEGF164 and pSS10/TVA-VEGF110, and I. Berger for subcloning AP–mVEGFs. U.F. receives an Emmy Noether-grant from the Deutsche Forschungsgemeinschaft.