Molecular Cloning and Characterization of Endosialin, a C-type Lectin-like Cell Surface Receptor of Tumor Endothelium*

Endosialin, the antigen identified with monoclonal antibody FB5, is a highly restricted 165-kDa cell surface glycoprotein expressed by tumor blood vessel endothelium in a broad range of human cancers but not detected in blood vessels or other cell types in many normal tissues. Functional analysis of endosialin has been hampered by a lack of information about its molecular structure. In this study, we describe the purification and partial amino acid sequencing of endosialin, leading to the cloning of a full-length cDNA with an open reading frame of 2274 base pairs. The endosialin cDNA encodes a type I membrane protein of 757 amino acids with a predicted molecular mass of 80.9 kDa. The sequence matches with an expressed sequence tag of unknown function in public data bases, named TEM1, which was independently linked to tumor endothelium by serial analysis of gene expression profiling. Bioinformatic evaluation classifies endosialin as a C-type lectin-like protein, composed of a signal leader peptide, five globular extracellular domains (including a C-type lectin domain, one domain with similarity to the Sushi/ccp/scr pattern, and three EGF repeats), followed by a mucin-like region, a transmembrane segment, and a short cytoplasmic tail. Carbohydrate analysis shows that the endosialin core protein carries abundantly sialylated, O-linked oligosaccharides and is sensitive toO-sialoglycoprotein endopeptidase, placing it in the group of sialomucin-like molecules. The N-terminal 360 amino acids of endosialin show homology to thrombomodulin, a receptor involved in regulating blood coagulation, and to complement receptor C1qRp. This structural kinship may indicate a function for endosialin as a tumor endothelial receptor for as yet unknown ligands, a notion now amenable to molecular investigation.

The endosialin antigen was identified in a survey of normal and neoplastic human tissues conducted at the Ludwig Institute for Cancer Research in pursuit of new targets for antibodybased cancer therapies (37). The hallmark of monoclonal antibody (mAb) 1 FB5, the probe used to discover endosialin (1), is its distinctive pattern of reactivity with human tissues. Thus, in a detailed study of biopsy and surgical specimens representing diverse cancer types, FB5 immunostaining was found primarily in tumor blood vessels and not in malignant tumor cells. Significantly, the antigen was not observed in all cancer samples examined, and even in cancers showing FB5-immunoreactive endothelial cells, the antigen was frequently detected with a heterogeneous pattern in the tumor vascular bed. Such a mixed pattern might be expected for a molecule involved in the reorganization of blood vessels in tissues such as cancers, in which areas of stable blood supply and histology are juxtaposed to regions of necrosis, hypoxia, excessive growth, tissue invasion, and remodeling. The normal tissues examined were unreactive with mAb FB5, including the blood vessel endothelium present in the respective organs.
The expression of the FB5 antigen by cultured normal and tumor cells was also investigated (1), revealing that the standard test cells for normal endothelial differentiation markers, cultured human umbilical vein endothelial cells, and microvascular endothelial cells derived from bone marrow and dermis are antigen-negative. Human umbilical vein endothelial cell cultures stimulated with a range of mediators known to induce the expression of endothelial activation antigens (2, 3) maintain their FB5-negative phenotype (1). Among a host of cultured epithelial, neuroectodermal, mesenchymal, and hematopoietic cell types tested, most were FB5 antigen-negative. Notable exceptions are short term cultures of normal fibroblasts and neuroblastoma cell lines, which consistently express the antigen in tissue culture (1), allowing the chromosomal assignment of an FB5 coding gene to human chromosome 11q13-qter, based on the serologic analysis of mouse-human neuroblastoma cell hybrids. The reason why cultured fibroblasts and neuroblastomas are FB5 positive in vitro, yet the corresponding cells in uncultured tissue sections are FB5 negative, is not known. Presumably, some of these findings reflect cell type-restricted, adaptive changes of the cell surface phenotype triggered by tissue culture factors (38); the mechanism underlying such induction and its relationship to FB5 antigen induction in tumor blood vessels are not known.
Methods are not generally available to examine the biochemical nature of antigens with restricted expression in tumor endothelium, because of to three properties of these cells: (i) they constitute a very minor fraction in most tumor tissues; (ii) they are not readily purified in sufficient numbers for direct * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM /EBI Data Bank with accession number(s) AJ295846.
** To whom correspondence should be addressed: Boehringer Ingelheim Pharma KG, 88397 Biberach and der Riss, Germany. E-mail: martin.lenter@bc.boehringer-ingelheim.com. 1 The abbreviations used are: mAb, monoclonal antibody; bp, base pair(s); EGF, epidermal growth factor; EST, expressed sequence tag; OSGE, O-sialoglycoprotein endopeptidase; PAGE, polyacrylamide gel electrophoresis; HPLC, high performance liquid chromatography; MOPS, 4-morpholinepropanesulfonic acid; RACE, rapid amplification of cDNA ends; SAGE, serial analysis of gene expression; TEM1, tumor endothelial marker 1. protein analysis; and (iii) no cell culture model of tumor endothelium exists that faithfully reproduces its distinctive in vivo phenotype. With no source for antigen characterization in tumor endothelium, the target molecule for mAb FB5 was studied in cultured neuroblastoma and sarcoma cells and fibroblast cultures. In these cell types, the antigen is a unique 165-kDa glycoprotein, comprised of a core protein that migrates as a 95-kDa species on SDS gels and carries the mAb FB5-defined epitope, and abundant, highly sialylated O-linked carbohydrate moieties. There has been no proof that the same protein carries the FB5 epitope in tumor endothelium, but in recognition of the characteristic biochemical properties and the tumor endothelial expression pattern the molecule was designated endosialin.
In the absence of more detailed information about the molecular structure of endosialin, no clues were available regarding its potential function in cancer. Therefore, considering the keen interest in angiogenesis as a determinant of cancer progression and metastasis (4) and as a target for novel cancer therapies (5), this study aimed to clone the endosialin gene and provide the requisite probes and structural information to explore endosialin function in suitable model systems and in human cancers.
Northern Blot Analysis-The SV Total RNA Isolation System (Promega) was used for total RNA isolation according to the manufacturer's instructions. Oligo(dT)-cellulose (Life Technologies, Inc.) was taken for poly(A) ϩ RNA isolation as described (8). For Northern blots, 4 g of poly(A) ϩ RNA were electrophoresed on a 0.8% (w/v) agarose gel containing 20 mM MOPS, 5 mM sodium acetate, pH 6.6, and 1.11% formaldehyde. The RNA was blotted in 10ϫ SSC (1.5 M NaCl, 0.15 M sodium citrate, pH 7.0) on a Hybond Nϩ membrane (Amersham Pharmacia Biotech) for 16 -20 h and UV cross-linked. To detect endosialin mRNA, a 287-bp polymerase chain reaction fragment, comprising positions ϩ 2237 to ϩ 2524 of endosialin cDNA, was used.
RACE-Polymerase Chain Reaction -For reverse transcription 2 g of total RNA were added to a mix of 4 l of the display THERMO-RT buffer (Display Systems Biotech), 2 l of a 5 mM dNTPmix, 7 l of 3 M betain (Sigma), 1 l of the display THERMO-RT Initiator Mix (total volume, 20 l), and 1 M of primer Est2243DN2 (ACAGGTAGCCGTC-GACAGCCAGCGTGC). For cDNA synthesis the mixture was incubated for 10 min at 65°C, cooled to 42°C, and incubated with 2 l of the display THERMO-RT terminator solution for 40 min. The temperature was raised to 65°C for 15 min, cooled to 37°C, and then 1 l of RNase Mix (5Ј RACE System for Rapid Amplification of cDNA Ends Reagent Assembly, version 2.0; Life Technologies, Inc.) was added and incubated at 37°C for 30 min. The cDNA was purified via GlassMax (Life Technologies, Inc.), and the subsequent tailing reaction was performed in the presence of 1 M betain according to the manufacturer's instructions (Life Technologies, Inc.) using primer EST 1987DN1 (AGAGGCTG-GCTGGGCCCAGTTGGTG) and the Abridged Anchor Primer (Life Technologies, Inc.). The cycling parameters were 94°C for 40 s, 62°C for 40 s, and 72°C for 105 s, for a total of 40 cycles, with a final extension cycle of 72°C for 10 min. An aliquot containing 1% of the specifically amplified tailed cDNA was subjected to a second nested polymerase chain reaction using the endosialin primer GSP4 (AC-CCACCAGGCTGTCCACACGCTGG) and the Abridged Universal Amplification Primer (Life Technologies, Inc.) under the cycling conditions described above. The products were resolved on a 2% (w/v) agarose gel and sequenced on an ABI PRISM 310 Genetic Analyser (PerkinElmer Life Sciences).
Transfection and Immunocytochemistry-Transient transfection of HeLa-S3 cells plated in chamber slides (Becton Dickinson, Heidelberg, Germany) was carried out with the transfection reagent FuGENE 6 and expression vector pMH (Roche Diagnostics), containing the complete endosialin coding sequence or as empty vector. Cells were analyzed for antigen expression 24 h after transfection, using immunocytochemistry with mAb FB5 or control IgG essentially as described (9). Antibody binding was detected with Alexa Fluor 488-conjugated goat anti-mouse IgG F(abЈ) 2 fragment (Molecular Probes, Eugene, OR).
Immunoprecipitation Assays and Carbohydrate Analysis-Immunoprecipitations were performed as described (8). For glycosylation studies, cells were metabolically labeled with 400 Ci of [ 35 S] methionine/ 400 Ci of [ 35 S]cysteine in 1.2 ml of methionine/cysteine-free minimal essential medium (Life Technologies, Inc.) for 16 h. Cell lysates were precleared with protein A-Sepharose (Amersham Pharmacia Biotech) and precipitated with mAbs or mIgG coupled to CNBr-Sepharose (Amersham Pharmacia Biotech) at a ratio of 2.8 mg/ml of gel. The precipitate was washed, split into equal aliquots, and incubated for 16 h at 37°C either with 1 unit of PNGase F (N-glycanase F) from Flavobacterium meningosepticum, 10 milliunits sialidase from Arthrobacter ureafaciens, or a mixture of 10 milliunits sialidase and 0.5 milliunits O-glycosidase from Diplococcus pneumoniae (Roche Diagnostics) or with 25 l of reaction buffer alone (50 mM sodium phosphate, pH 7.0, 0.5% Triton X-100). Precipitates were washed, eluted in Laemmli buffer, and reduced with 50 mM dithiothreitol. Signals were detected after SDS-PAGE by exposure to Kodak BIOMAX MR film. For Osialoglycoprotein endopeptidase digestion, equal aliquots of LA1-5s cell detergent lysates were precipitated with mAb FB5 coupled to CNBr-Sepharose (2.8 mg/ml of gel) or with mAb 9EG7 against integrin ␤ 1chain coupled to protein A-Sepharose (1 mg/ml of gel) via rabbit anti-rat IgG (1 mg/ml of gel) for 2 h at 4°C. The precipitates were washed, split into equal aliquots, and treated either with 4 l of O-sialoglycoprotein endopeptidase from Pasteurella hemolytica (activity: 1 l cleaves 10 g of glycophorin A/h at 37°C; Cedarlane Laboratories, Hornby, Canada) or reaction buffer only (50 mM Tris/HCl, pH 7.4), with incubation for 1 h at 37°C. Proteins were eluted and analyzed by SDS-PAGE as described above.
Data Base Mining and Sequence Analyses-Amino acid sequence information for endosialin peptides was used to perform iterative BLAST, FASTA, and Prosite pattern data base searches against public sequence data bases. The same algorithms, including gapped BLAST and PSI_BLAST (10), have been applied for further analysis of endosialin cDNA and amino acid sequence. The SEG program was applied for searching low complexity regions in protein sequences with the following search parameters: window length, 25; trigger complexity, 3.0; and extension complexity, 3.3 (11). Known functional sequence domains matching sequence segments in endosialin were examined in the PFAM and SMART (13) domain libraries.

Protein Purification of Human Endosialin and Identification
of the cDNA-We used immunocytochemistry and enzymelinked immunosorbent assays with mAb FB5 to confirm that human umbilical vein endothelial cell cultures and cultures of microvascular endothelial cells from dermis and bone marrow lack endosialin expression. Therefore, the FB5 antigen-positive neuroblastoma cell line LA1-5s was selected as a source for endosialin purification. Protein was isolated from detergent extracts of 6 ϫ 10 9 cells using mAb FB5 conjugated to CNBr-Sepharose as the immunoaffinity matrix. To confirm the purity of the resulting preparation, an aliquot of the eluted material was separated on an SDS-polyacrylamide gel, and a major 165-kDa band, consistent with the size of endosialin, was visualized by silver staining (Fig. 1). The remaining eluate was subjected to quantitative SDS-PAGE and yielded about 10 pmol of specific protein as determined by Coomassie staining. Endosialin-containing gel fragments were digested with trypsin, and the resulting peptides, separated by HPLC and analyzed by Edman degradation and matrix-assisted laser desorption ionization-time of flight, gave rise to six unique peptide sequences. In a Prosite pattern search, five out of these six peptides matched with closely spaced protein sequences deduced from the EST clone AI371224 (Fig. 2, double underlined). The complete cDNA sequence of this EST revealed a single open reading frame of 1,394 bp. No potential start and stop codons were found, prompting us to conduct an iterative data base search that identified two additional ESTs, H19258 and AA595199, respectively, which extend the 3Ј end of the open reading frame to 2,239 bp with a stop codon at position ϩ 2272 and a polyadenylation signal at position ϩ 2512 followed by the poly(A) tail (Fig. 2). The cDNA data were verified by sequencing two independent EST clones, H12800 and H12557, that cover the sequence of HS19258 and AA595199. Because Northern blots of poly(A) ϩ RNA isolated from the two endosialinexpressing cell lines LA1-5s and MF-SH show a single, major transcript of about 2,600 bp (Fig. 3), only a short 5Ј segment of the endosialin mRNA appeared to be missing in the ESTderived cDNA sequence. Using RACE, we succeeded in extending the 5Ј sequence by 52 bp, including a putative start ATG. Thus, the complete cDNA of endosialin comprises an open reading frame of 2,274 bp and codes for a protein of 757 amino acids with a predicted molecular mass of 80.9 kDa.
Endosialin cDNA Encodes the FB5 Cell Surface Antigen-Transient transfection studies with the endosialin-negative epithelial cell line HeLa-S3 were used to confirm that the putative endosialin cDNA indeed specifies the cell surface antigen defined by mAb FB5. Although the parental or mock-transfected HeLa-S3 cells grown in plastic chamber slides show no detectable binding of mAb FB5 in indirect immunofluorescence tests, the endosialin transfectants generated with an expres-sion vector containing full-length endosialin cDNA show strong staining with mAb FB5 (Fig. 4).
Endosialin Is a C-type Lectin-like Membrane Receptor-Bioinformatic analysis identifies endosialin as a typical precursor sequence for a eukaryotic cell surface protein, resulting evolutionarily from domain shuffling. The endosialin protein displays a considerable number of disparate functional segments (Fig. 5A). First, there is an N-terminal sequence of about 20 amino acids that is similar to a signal leader peptide for export to the endoplasmic reticulum and can be detected by the SIGNALP2 suite of Nielsen et al. (14). Next, there are two major segments, a more N-terminal globular portion with five distinct domains comprising amino acids 30 -360, and a more C-terminal portion comprising amino acids 361-757 and classified almost exclusively as low complexity region by the SEG program (11). A more detailed analysis of domain structure architecture in the N-terminal globular portion is accomplished by comparison with hidden Markov models in libraries of described sequence domains, such as PFAM (12) and SMART (13), as well as sequence similarity searches in data bases with the BLAST and PSI-BLAST tools (10) that rely on the concept of a common evolutionary ancestor among sequentially homologous sequences. By this approach, we found statistically significant hits for a C-type lectin domain (residues 29 -157, E Ͻ  10 Ϫ5 with PFAM) and three EGF-like domains (residues 235-271, E Ͻ 10 Ϫ5 with PFAM; residues 274 -311, E Ͻ 10 Ϫ2 with SMART; residues 316 -350, E Ͻ 10 Ϫ3 with PFAM). In accordance with SMART assignments, the latter two EGF repeats may be Ca 2ϩ binding. The region 176 -230 is probably a Sushi/ SCR/CCP domain. Although the E value in a PFAM search is only 0.72, the typical motifs in the Sushi seed alignment with the endosialin segment are well conserved, including all necessary cysteines for disulfide bridges and a number of critical prolines, glycines, and a tryptophan.
A more detailed analysis of the C-terminal, low complexity portion of endosialin, covering amino acids 360 -757, showed that the local amino acid composition is dominated by the preferential occurrence or absence of certain residue types. In accordance with analyses of known three-dimensional protein structures (15), there is little reason to expect such polypeptide segments to assume stable tertiary structures with sterically predefined binding or catalytic activity. It is not surprising that secondary structure tools (16,17) predict almost exclusively coil preferences for this region. Segment 363-380 is highly negatively charged, with 6 glutamate and 7 aspartate residues, and it may serve a nonspecific role in facilitating a binding contact with a charged ligand molecule. Using the software tools DAS (18) and TOPPRED2 (19), the existence of a single hydrophobic transmembrane region (amino acids 686 -706) can be detected with strong significance; thus, endosialin appears to be a type I membrane protein. The extracellular region spanning amino acids 391-660 is much richer than the data base average (20) in prolines (21.9%), threonines (9.3%), and serines (9.3%), but it contains surprisingly few charged residues (DEKR total only 9.6%), glycines (2.2%), asparagines (1.1%), and phenylalanines (1.5%). There is no truly sequentially similar protein sequence in the data base. The small intracellular portion 707-757 is compositionally rich in proline (11.8%) but depleted of phenylalanine (none) and hydrophobic residues in general (LVIFM total only 17.6%). This sequence region is also unique compared with other known protein sequences.
Although the prediction of O-glycosylation sites is not very reliable even with the most advanced algorithms (21), we supplied the endosialin sequence to the NetOGlyc 2.0 server offered by J. E. Hansen. This program predicts as probable Oglycosylation sites Thr-60 (in the C-type lectin domain) and a series of 34 serine and threonine residues located in segment 400 -669, which probably does not have a globular structure.
Endosialin Is Related to Thrombomodulin and Complement Receptor C1qRp-The N-terminal endosialin segment between amino acids 1 and 360 shares 39% sequence identity with the corresponding region in the precursor protein of human thrombomodulin (also referred to as fetomodulin, CD141, or emb: CAA29045.1) and 33% sequence identity with the human complement receptor C1qRp (also known as lymphocyte antigen 68, antigen AA4, or emb:CAC00597.1), respectively. The matching region encompasses the C-type lectin domain, the putative Sushi domain, and the three EGF-like repeats (Fig. 5B). Moreover, the positions of cysteine residues critical for disulfide bridges are highly conserved among the three proteins, consistent with a shared three-dimensional structure in this segment. Finally, the three proteins show a conserved WIGL consensus motif, a feature found in several additional cell surface proteins that modulate endocytosis (22).
Endosialin Is Conserved in Nonhuman Species and Maps to Human Chromosome 11q13-A number of mouse and rat ESTs, originating from embryonic tissue sources and comprising fragments of 186 -521 bp were identified in public data base searches. These ESTs show 82-92% identity to the corresponding human endosialin cDNA sequence, thus suggesting the presence of a close homologue in nonhuman species. Further data base searches identified a human genomic clone (RP11-755F10, accession number AP000759) assigned to chromosome 11q13, which contains more than 75% of the endosialin coding sequence (positions ϩ521 to ϩ2533) in a putative single exon and in this portion is 99% identical with the endosialin cDNA. This chromosomal assignment for the endosialin confirms and refines the assignment to 11q13-qter made for the FB5 antigen using somatic cell hybrids (1).
Carbohydrate Analysis of Recombinant Endosialin-In several human cell types, endosialin shows a high degree of Oglycosylation and sialic acid content (1), and we have examined the glycosylation pattern of recombinant endosialin in transiently transfected HeLa-S3 cells (Fig. 6A). When tested by immunoprecipitation with mAb FB5 immobilized to CNBr-Sepharose, a 165-kDa protein is detected in these cells, but not in mock-transfected HeLa-S3 cells. Pretreatment of matrixbound antigen with O-glycosidase or sialidase led to a marked reduction in the size of the endosialin band on SDS gels, with a pattern identical to endogenous endosialin in LA1-5s cells (Fig. 6A). Thus, sialidase treatment alone yields a 120-kDa protein species, and combined treatment with sialidase and O-glycanase generates the putative 95-kDa core protein. No shift in the mobility of endosialin on SDS gels was observed  after PNGase F treatment, consistent with the notion that minor, if any, N-linked sugar moieties are present at the single Asn-Xaa-(Ser/Thr) consensus site at position Asn-628. It is noteworthy that endogenous endosialin in LA1-5s cells does not show the marked heterogeneity in glycosylation, revealed by broad bands on SDS gels, that is typical for a number of other glycoproteins with abundant O-glycosylation. The appearance of two closely spaced endosialin bands in the HeLa-S3 transfectants may be indicative of limited heterogeneity of O-glycosylation in this particular cell type, because the two bands merge into a single band after deglycosylation.
Mucin-like Endopeptidase Susceptibility-Because O-sialoglycoprotein endopeptidase (OSGE) has been shown to cleave specifically highly O-glycosylated mucin-like molecules (23,24), we examined whether endosialin is susceptible to this enzyme. To this end, detergent extracts of metabolically labeled LA1-5s cells were purified with mAb FB5 and treated with OSGE for 1 h at 37°C prior to SDS-PAGE. We found that OSGE completely degraded endosialin (Fig. 6B), whereas a control glycoprotein, ␤ 1 -integrin, was largely insensitive to OSGE under these conditions. DISCUSSION We have cloned the gene coding for the human endosialin core protein based on several lines of evidence. First, transfection of endosialin cDNA into HeLa-S3 cells leads to induction of cell surface reactivity with the cognate mAb FB5. Secondly, mAb FB5 detects the expected 165-kDa sialoglycoprotein in the transfectants, with a core protein migrating as a 95-kDa species on SDS gels. Third, the endosialin gene maps to chromo-some 11q13, refining the previous assignment to 11q13-qter based on serologic analysis of somatic cell hybrids with mAb FB5. Finally, the amino acid sequence provides the requisite attachment sites for abundant O-glycosylation of endosialin.
The present study is not the only evidence linking endosialin induction to vascular endothelial cells in human cancer. Rather, an independent line of investigation, aimed at dissecting comprehensive gene expression profiles in cancers with the serial analysis of gene expression (SAGE) method, has implicated endosialin-specific ESTs in tumor angiogenesis. St. Croix et al. (25) employed SAGE to survey about 32,500 unique gene transcripts for differential expression in purified tumor endothelium versus endothelium of nonmalignant tissue. The study identified 46 transcripts elevated at least 10-fold in tumor endothelium, and the highest and most consistent up-regulation was noted for a hitherto uncharacterized gene, deposited as ESTs of unknown function in data bases and designated tumor endothelial marker 1 (TEM1) by St. Croix et al. (25). We show here that TEM1 is the endosialin gene.
In their study, St. Croix et al. (25) confirmed the SAGE data and showed by in situ RNA hybridization that TEM1/endosialin expression in vivo is selective for tumor endothelium, with no detection in the malignant epithelial cells of colorectal and other cancer, a range of normal organs, and a panel of cultured tumor cell lines. Nevertheless, TEM1/endosialin mRNA was not unique to tumor blood vessels because endothelial cells associated with wound healing and corpus luteum formation also showed gene expression by in situ RNA hybridization.
Taking together the endosialin expression data generated The alignment shows the N-terminal 360 amino acids of endosialin (end marked by black slash), which include the C-type lectin domain, the Sushi-like domain, and the three EGF-like repeats. In this region, endosialin shares 39% identity with the matched region of thrombomodulin and 33% identity with C1qRp (white letters on blue background), respectively. Nearly all cysteine residues in the five domains are conserved among the three proteins (red). The WIGL consensus motif is depicted in green.
with mAb FB5 (1) and the RNA expression data derived from SAGE and in situ RNA hybridization studies (25), a consistent picture of endosialin induction in tumor endothelium emerges. Such a close agreement between epitope and protein/mRNA expression is instructive for a molecule like endosialin, which falls into the group of sialomucin-type glycoproteins. There is precedent for sialomucins and other glycoconjugates for which mAb-defined epitope expression and protein/mRNA distribution sharply diverge. Several levels of complexity have been described (26 -29), including (i) epitopes on core proteins that are masked by glycosylation or specific sialylation, (ii) epitopes on carbohydrate moieties shared by several unrelated glycoproteins or glycolipids, and (iii) epitopes on distinct splice variants or processing variants of the same core protein. In the case of endosialin, the observed cotyping for the FB5 epitope and TEM1/endosialin mRNA suggests that mAb FB5 binds to an invariant epitope on the 95-kDa core protein. Nevertheless, this conclusion needs to be confirmed for each cell type under investigation, because epitope masking and mimicry can show exquisite cell type selectivity.
A persistent question in the exploration of tumor angiogenesis is the extent of phenotypic and functional heterogeneity that may distinguish subsets of tumor endothelial cells in a macroscopic tumor lesion with regard to their derivation from pre-existing blood vessels or new vessel sprouting, and their function and state of activation in response to varying milieus of tumor-derived, paracrine angiogenesis inducers and inhibitors. Generally, the capillary bed of a locally advanced or metastatic tumor mass will extend into areas of relatively stable blood supply and tissue architecture and closely juxtaposed or intermingled regions of hypoxia, necrosis, excessive proliferation, tissue invasion, fibroblastic remodeling, or inflammatory cell infiltration. Placed in this context of malignant tissue disruption, markers of tumor angiogenesis might be expected to display a heterogeneous pattern of expression as a rule rather than exception. Considering the marked heterogeneity in FB5 staining in tumor vessels (1), it may be surprising that the report on TEM1/endosialin mRNA in situ hybridization (25) does not address this aspect. One explanation would suggest that endosialin mRNA is uniformly expressed in tumor endothelium and that only subsets of endothelial cells accumulate the endosialin protein or protein with the FB5 epitope; however, as outlined above, there is currently no evidence supporting such a mechanism. Alternatively, it may be argued that mRNA in situ hybridization is the more sensitive detection method for endosialin and that FB5 immunostaining marks tumor endothelium with the highest levels of protein/epitope density rather than a distinct tumor endothelial cell type. Finally, there may simply be differences in the design of the two available studies on endosialin expression.
Our analysis shows that the N-terminal extracellular portion of endosialin is likely to consist of five domains with globular structure, and in this region homology to thrombomodulin and the complement receptor C1qRp, with 39 and 33% amino acid sequence identity, respectively, is observed. Importantly, a pattern of cysteine residues is highly conserved among the three proteins and provides a scaffold of anchoring points for disulfide bridges, presumably imposing similar three-dimensional structures on each protein. Thrombomodulin exhibits a C-type lectin domain, six EGF-like repeats, and a serine/threoninerich region in the extracellular domain. Unlike endosialin, thrombomodulin is expressed on a variety of normal cell types and on normal vascular endothelium serves as a receptor for thrombin, modulating the coagulation cascade and triggering the thrombin-activable fibrinolysis inhibitor (30). The thrombin interaction depends on the EGF repeats 5 and 6 of thrombomodulin (31,32), which are not present in endosialin, making it unlikely that the two molecules overlap in this particular function. The complement receptor C1qRp is also expressed on a wider variety of cell types than endosialin, including macrophages, neutrophilic granulocytes, and normal vascular endothelium. The extracellular portion of C1qRp is composed of a C-type lectin domain, a tandem of five EGF-like repeats, and a mucin-like region. The molecule binds the complement factor C1q, mannose-binding lectin, and pulmonary surfactant protein A (22). On myeloid cells, C1qRp mediates phagocytosis of antibody-coated particles, but the function of C1qRp on normal endothelial cells is largely unknown (33).
Predictions about the membrane-proximal extracellular domain focus on fewer structural features. Thus, the high content of proline, threonine, and serine residues indicates likely attachment points for more than 30 O-linked carbohydrate side chains consistent with the sialomucin-like features of endosialin (28,34). Certain sialomucins have been implicated in sequestering growth factors to the plasma membrane and presenting these factors to endothelial cell surface receptors (34,  (HeLa transfectant), or LA1-5s cells (LA1-5s), respectively, were precipitated with mAb FB5 conjugated to CNBr-Sepharose or with control mouse IgG on CNBr-Sepharose. Immunoprecipitated endosialin was either mock-treated (none) or pretreated with PNGase F, sialidase, or with a combination of O-glycosidase and sialidase. B, SDS-PAGE analysis of endosialin following pretreatment with O-sialoglycoprotein endopeptidase. To test the sensitivity of endosialin to O-sialoglycoprotein endopeptidase, detergent lysates of the labeled LA1-5s cells were precipitated with mAb FB5 (FB5) or anti-␤1 integrin mAb 9EG7 (anti-␤1) as a control. Immunoprecipitated antigens were treated with O-sialoglycoprotein endopeptidase from P. hemolytica for 1 h at 37°C (OSGE) or mock-treated without enzyme (none). Note complete digestion of endosialin by OSGE and only limited degradation, presumably nonspecific, for ␤ 1 -integrin. Molecular size markers (in kDa) are indicated to the left. 35); endosialin may be a candidate for such a function unique to tumor endothelium.
Apart from further biochemical investigations that will benefit from the availability of mAb FB5 and recombinant endosialin protein, study of defined alterations in endosialin gene expression and structure are now possible, including genetic manipulations in mice and rats, which harbor a close homologue of endosialin. In conclusion, key experiments are now within reach to explore the endosialin system in human physiology and disease and to extend these investigations into experimentally more amenable rodent models. A better understanding of endosialin may even enable new approaches to cancer detection and treatment.