Characterization of TEM1/endosialin in human and murine brain tumors

Background TEM1/endosialin is an emerging microvascular marker of tumor angiogenesis. We characterized the expression pattern of TEM1/endosialin in astrocytic and metastatic brain tumors and investigated its role as a therapeutic target in human endothelial cells and mouse xenograft models. Methods In situ hybridization (ISH), immunohistochemistry (IH) and immunofluorescence (IF) were used to localize TEM1/endosialin expression in grade II-IV astrocytomas and metastatic brain tumors on tissue microarrays. Changes in TEM1/endosialin expression in response to pro-angiogenic conditions were assessed in human endothelial cells grown in vitro. Intracranial U87MG glioblastoma (GBM) xenografts were analyzed in nude TEM1/endosialin knockout (KO) and wildtype (WT) mice. Results TEM1/endosialin was upregulated in primary and metastatic human brain tumors, where it localized primarily to the tumor vasculature and a subset of tumor stromal cells. Analysis of 275 arrayed grade II-IV astrocytomas demonstrated TEM1/endosialin expression in 79% of tumors. Robust TEM1/endosialin expression occurred in 31% of glioblastomas (grade IV astroctyomas). TEM1/endosialin expression was inversely correlated with patient age. TEM1/endosialin showed limited co-localization with CD31, αSMA and fibronectin in clinical specimens. In vitro, TEM1/endosialin was upregulated in human endothelial cells cultured in matrigel. Vascular Tem1/endosialin was induced in intracranial U87MG GBM xenografts grown in mice. Tem1/endosialin KO vs WT mice demonstrated equivalent survival and tumor growth when implanted with intracranial GBM xenografts, although Tem1/endosialin KO tumors were significantly more vascular than the WT counterparts. Conclusion TEM1/endosialin was induced in the vasculature of high-grade brain tumors where its expression was inversely correlated with patient age. Although lack of TEM1/endosialin did not suppress growth of intracranial GBM xenografts, it did increase tumor vascularity. The cellular localization of TEM1/endosialin and its expression profile in primary and metastatic brain tumors support efforts to therapeutically target this protein, potentially via antibody mediated drug delivery strategies.

Conclusion: TEM1/endosialin was induced in the vasculature of high-grade brain tumors where its expression was inversely correlated with patient age. Although lack of TEM1/endosialin did not suppress growth of intracranial GBM xenografts, it did increase tumor vascularity. The cellular localization of TEM1/endosialin and its expression profile in primary and metastatic brain tumors support efforts to therapeutically target this protein, potentially via antibody mediated drug delivery strategies.

Background
Despite advances in neurosurgical techniques, chemotherapeutic regimens and radiotherapy protocols, the prognosis for patients suffering from malignant astrocytoma remains bleak. Novel treatment approaches are required which attack the unique molecular and biological features that contribute to the growth of these tumors. A particularly promising target is the abnormal tumor vasculature, which represents one of the defining characteristics of the most malignant and common astrocytoma, glioblastoma multiforme (GBM). We previously compared the gene expression profile of endothelial cells isolated from freshly resected GBMs with that of endothelium isolated from non-neoplastic temporal lobe [1]. These experiments and subsequent studies identified 21 genes that met the statistical requirements as putative glioma endothelial markers (GEMS). One especially promising candidate was TEM1/endosialin. TEM1 was identified in the tumor endothelium of human colon carcinoma [2]. Soon after, it was recognized that TEM1 encoded endosialin, which corresponded to the tumor vascular endothelial antigen recognized by the FB5 antibody [3,4]. Further analyses revealed selective TEM1/ endosialin expression in tumor endothelium, pericytes and a subset of fibroblast-like cells of tumor stroma in breast carcinoma, anaplastic astrocytoma and GBM [5][6][7][8][9][10]. TEM1/endosialin encodes a transmembrane glycoprotein and putative membrane receptor. Recent evidence suggests it may interact with extracellular matrix components including collagen I, collagen IV and fibronection, as well as Mac-2 BP/90K in promoting vascular migration and invasion [11,12]. Tem1/endosialin knockout (KO) mice are fertile and appear to develop normally. However, when human HCT116 colon carcinoma cells were implanted orthotopically onto the serosal surface of the large intestine of nude Tem1/endosialin KO mice, both tumor take and growth were inhibited while the number of tumor microvessels increased [13]. The selective induction of TEM1/endosialin in malignant gliomas, its cell surface bioavailability and the evidence that lack of TEM1/endosialin can disrupt tumor growth and vascular differentiation in a xenograft model combine to make it a potential target for molecular therapy.
We sought to expand these preliminary findings by examining the specificity of TEM1/endosialin expression in a larger collection of primary and secondary brain tumors. We also wished to determine whether TEM1/endosialin induction in brain tumors was conserved in a commonly used U87MG intracranial xenograft model. Finally, we wished to determine whether TEM1/endosialin expression was required for brain tumor growth in vivo. We confirm that TEM1/endosialin is induced, primarily in the vascular compartment, in a wide range of both low-grade and high-grade cerebral neoplasms and is inversely correlated with patient age. TEM1/endosialin shows limited colocalization with fibronectin, a putative binding partner, in malignant brain tumors. In vitro, TEM1/endosialin expression can be stimulated in human microvascular endothelial cells grown in matrigel. Finally, we show that Tem1/endosialin upregulation occurs in the tumor vasculature of intracranial GBM xenografts and absence of Tem1/ endosialin is associated with increased numbers of microvessels within tumors explanted from Tem1/endosialin KO mice, supporting a role for TEM1/endosialin in the migration and maturation of the brain tumor vasculature. However, despite its conserved expression pattern in the tumor cerebrovasculature, we find that Tem1/endosialin is not required for intracranial tumor growth in Tem1/endosialin KO mice.

Clinical specimens
Clinical specimens were provided by the Brain Tumor Bank of the University of Pittsburgh as approved by the University of Pittsburgh Institutional Review Board and the Neurosurgical Brain Tissue Bank of the University of Rochester as approved by the University of Rochester Research Subjects Review Board. All samples were deidentified according to HIPAA regulations prior to being provided for this research. Astrocytoma array materials were obtained from surgical patients at Tampere University Hospital, Tampere, Finland, in 1983-2002, in accordance with the Research Ethics Committee of Tampere University Hospital.

RNA isolation and reverse transcription
RNA isolations and reverse transcription was performed as previously described [14]. In brief, total RNA was isolated from 10-30 mg of snap frozen, homogenized tissue or approximately 10 6 cultured cells using the Qiagen RNeasy Mini Kit (Qiagen, Valencia, CA) according to manufacturer's protocol. cDNA was generated using the SuperScript First-Strand Synthesis System for RT-PCR (Invitrogen, Carlsbad, CA) according to manufacturer's protocol.

PCR amplification
All PCR primers were generated by IDT (Integrated DNA Technologies, Coralville, IA). Sequences available upon request. PCR reactions were performed using Platinum Taq Polymerase (Invitrogen) as previously described [14]. Reactions were cycled in a PTC-200 DNA Engine thermalcycler (MJ Research, Waltham, MA). Primers for vWF were used for normalization.

Quantitative RT-PCR amplification
qRT-PCR was carried out on triplicate samples for 40 cycles of 10 s at 95°C, 30 s at 60°C after an initial incubation at 95°C for 3 min in an Chromo4 thermal cycler (Bio-Rad, Hercules, CA). Reaction conditions were 1× iQ SYBR Green super mix (Bio-Rad), 400 nM forward and reverse primers, and 2.0 μl cDNA in a total reaction volume of 25 μl. Amplification of vWF was performed for each cDNA in triplicate for normalization of vascular RNA content. Primer sequences were: TEM1 (F)5'-ctacgttggtggcttcgagt-3'; TEM1(R)5'-caggcctcgtcttcatcttc-3'; vWF(F)5'cacatcgtgacctttgatgg-3'; vWF(R)5'-gtgcttcacctcgatggatt-3'. Threshold cycle number (Ct) of amplification in each sample was determined by Bio-Rad software. Relative mRNA abundance was calculated as the Ct for amplification of TEM1/endosialin minus average Ct for vWF, expressed as a power of 2, i.e. 2 -ΔCt . Three individual TEM1/endosialin values were averaged to obtain mean ± SD. Statistical significance was determined using a Wilcoxon rank sum test.

In situ hybridization
Protocol was modified from [14]. Briefly, probe templates blanketing the target cDNA were generated by PCR amplification with incorporation of a T7 promoter into the antisense primer. In vitro transcription was performed using the DIG RNA Labeling Kit (SP6/T7) and T7 RNA polymerase (Roche, Indianapolis, IN) according to manufacturer's protocol. Sections were deparaffinized in xylenes and rehydrated in graded ethanols. DIG-labeled RNA probe cocktails were diluted to 200 ng/ml in RNA hybridization solution (EMD Bioscience, Madison, WI) and hybridized to slides overnight at 55°C. Slides were treated with RNase A/T1 (Ambion, Austin, TX) and stringently washed. Slides were blocked with normal rabbit IgG (Dako) diluted 1:20 in buffer (100 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% casein). For detection, sections were incubated with rabbit HRP-anti-DIG primary (Dako) diluted 1:150 in blocking buffer. Signal was amplified by incubating each slide with one drop of biotinyl-tyramide (Dako). Slides were washed then incubated in goat HRP-antibiotin secondary (1:100, Vector Laboratories, Burlingame, CA) in blocking buffer. Amplification with biotinyl-tyramide was repeated. Sections were incubated with goat APanti-biotin tertiary (1:100, Vector) in blocking buffer and washed. Signal was detected by staining with Fast Red (Sigma-Aldrich, St. Louis, MO) followed by counterstaining with Probe/Hematoxylin counterstain (Biomeda, Foster City, CA). Slides were mounted with Crystal/Mount (Biomeda). Images were captured with an Olympus BH-2 microscope using Spot Insight Color model 3.2.0 camera and Spot Advanced software (Diagnostic Instruments, Sterling Heights, MI).

Immunohistochemistry
Sections were deparaffinized in xylenes and rehydrated in graded ethanols. Slides were washed in 1× PBS and then quenched with 0.3% H 2 0 2 in 0.3% normal serum/1× PBS for 5 minutes. Slides were washed in 1× PBS and blocked with 4% normal goat serum 1×PBS. Sections were exposed to rabbit polyclonal anti-TEM1 (1:50) or anti-vWF (1:500) at 4°C overnight. Sections were washed with 1× PBS and exposed for 1 hour to goat-anti-rabbit secondary, diluted 1:300. Sections were washed in 1× PBS then exposed to Vectastain-Elite ABC (Vector) for 30 minutes. Sections were washed in 1× PBS and incubated with NovaRed substrate (Vector). Slides were rinsed with distilled water and counterstained with Mayer's Hematoxylin (Biomeda, Foster City, CA). Sections were rinsed with distilled water, exposed to 0.08% NH 4 OH for 3 minutes, rinsed with distilled water and mounted with Crystal/ Mount. Images were captured as described above.

Astrocytoma array
Tumor specimens were fixed in 4% phosphate-buffered formaldehyde and processed into paraffin embedded tissue microarrays as previously described [15]. 275 samples were included in the analyses: 39 grade II, 41 grade III, and 195 grade IV. Patient age ranged from 3 to 80 years (mean ± SD: 50.4 ± 15.2). TEM1/endosialin expression was assessed by in situ hybridization as described above. Statistical analyses, including Mann-Whitney and Kruskal-Wallis tests to correlate expression to age, were performed using SPSS 11.0 for Windows (SPSS, Chicago, IL).

Immunofluorescence
Slides were deparaffinzed in xylenes and rehydrated in graded ethanols and rinsed in dH 2 O as described above. For TEM1/CD31, slides were boiled for 15 minutes in citrate buffer (Zymed), cooled for 20 minutes, then washed in PBS for 5 minutes. TEM1/αSMA staining did not require antigen retrieval. For TEM1/FN staining, slides were incubated with 0.1% proteinase K at 37°C for 15 minutes. Slides were washed in PBS and the blocked with 5% normal donkey serum containing 3% BSA in PBS for 30 minutes. Slides were incubated with primary antibodies diluted 1:10 (TEM1/33), 1:20 (CD31), 1:50 (αSMA) or 1:100 (FN) in blocking buffer at 37° for 1 hour. Slides were then rinsed with PBS and incubated with the appropriate secondary in diluted 1:200 in normal host serum for 1 hour at 37°C in the dark. Slides were rinsed in PBS before the addition of ToPro3 nuclear stain (Invitrogen) and then mounted with Prolong Gold mounting media (Invitrogen). Confocal images were captured using Zeiss LSM 510 Axioskop 2 microscope (Zeiss Microimaging, Thornwood, NY) and analyzed with Zen 2007 software (Zeiss).

Cell lines and culture conditions
Human primary brain microvascular endothelial cells (HPBMEC) (Cell Systems, Kirkland, WA) were maintained in CSC-Complete media (Cell Systems). Human dermal microvascular endothelial cells (HMVEC) (Lonza, Walkersville, MD) were maintained in EBM/EGM-2MV growth media with 1× growth supplements (hEGF, hydrocortisone, GA-1000, FBS, VEGF, hFGF-B, R 3 -IGF-1, and ascorbic acid) (Cambrex). U87MG tumor cells (ATCC, Manassas, VA) were maintained in Minimum Essential Medium Eagle with Earle's Salts and L-glutamine (Invitrogen, Carlsbad, CA), supplemented with 10% FBS, 0.1 mM MEM non-essential amino acids, 0.1 mM MEM sodium pyruvate, sodium bicarbonate, 1 unit/ml penicillin, and 1 μg/ml streptomycin (Invitrogen). Cells were maintained in a 5% CO 2 atmosphere at 37°C. Endothelial cells were used at passage 3-7 in all experiments. For growth factor experiments, endothelial cells were cultured in the presence of 25 ng/ml VEGF (Sigma), 10 ng/ml TGFβ(R&D Systems), 100 ng/ml IGF-1 (Upstate, Billerica, MA), 50 ng/ml SDF-1 (Biosource, Carlsbad, CA), 20 ng/ml SCF (Biosource) or 50 ng/ml SF/HGF (Biosource) for 48 hours prior to harvest and RNA isolation. Culture of endothelial cells in matrigel (BD Bioscience, Bedford, MA) was performed as described [14]. Cells were harvested from matrigel using Cell Recovery Solution (BD Bioscience) according to manufacturer's protocol. In brief, the cell/ matrigel layer was transferred to a 50 ml tube and 4 ml of Cell Recovery Solution was added. The mixture was incubated on ice to completely dissociate the matrigel and release the cells, which were then collected by centrifugation, washed twice in PBS and used for RNA isolation.

Intracranial xenografts
U87MG cells were harvested with trypsin (Invitrogen) and resuspended at a concentration of 10 5 cells/10 μl antibiotic-free DMEM. Athymic nu/nu mice were purchased from the National Cancer Institute (Frederick, MD) and the nude Tem1/endosialin KO and WT mice were derived as previously described [13]. Mice were anesthetized using an intraperitoneal injection of 3 ml/kg xylazine/ketamine and tumor cells were stereotactically injected as previously described [16]. Mice were observed daily and survival time was recorded. Mice were euthanized by barbituate overdose when they exhibited symptoms of tumor burden or distress. Brains were surgically removed, formalin fixed and embedded in paraffin. All procedures were approved by the Institutional Animal Care and Usage Committees of the Johns Hopkins University School of Medicine and the University of Rochester.

Vessel counts
Tumor sections were immunostained for vWF, as described above. Vessels were counted in 10 non-overlapping, 400× fields (5 fields each from two anatomically distinct tumor regions), in 6 Tem1/endosialin WT and 6 Tem1/ endosialin KO xenograft tumors. Statistical significance was determined by Wilcoxon rank sum test.

TEM1/endosialin is upregulated in both primary and secondary brain tumors
Our previous SAGE data and studies from others indicated that TEM1/endosialin was upregulated in the microvasculature of human GBMs and a subset of brain tumors [1,5,7]. In order to confirm that TEM1/endosialin induction was common to both primary and secondary brain tumors, we performed quantitative RT-PCR using TEM1/ endosialin specific primers on mRNA isolated from frozen clinical specimens of 2 non-neoplastic brain tissues and 5 GBMs, as well as 5 specimens of metastatic lung adenocarcinoma resected from brain. Reactions were normalized to vWF in order to account for the increased vascularity of the tumor specimens. Representative standard RT-PCR data are shown in Fig 1A. Quantitative RT-PCR results confirmed that upregulation of TEM1/endosialin is detectable in whole tissue from both primary and secondary brain tumor samples as compared to non-neoplastic controls and support a role for TEM1/endosialin as an angiogenic marker for CNS neoplasia (Fig 1B).
We next used in situ hybridization and immunohistochemistry to localize expression of TEM1/endosialin mRNA and protein (Fig 1C) using an independent array of 7 non-neoplastic (reactive) brain tissue specimens, 4 pilocytic astrocytomas, 7 metastatic adenocarcinomas (3 lung adenocarcinomas, 1 lung adenosquamous carcinoma, 1 non-small cell lung carcinoma, 1 esophageal adenocarcinoma and 1 colon adenocarcinoma) and 9 GBMs. TEM1/ endosialin expression was detected in the vasculature of both GBMs and metastatic brain tumors, as compared to the endothelial control marker, vWF. In GBMs, 3/9 tumors or approximately 33% of the tissues demonstrated a robust staining pattern for TEM1/endosialin (nearly 100% of vessels positive), while another 2/9 tumors showed moderate expression (approximately 50% of ves-sels positive) (Table 1). Similarly, 2/7 metastatic tumors (28%) showed robust staining for TEM1/endosialin while another 2/7 (28%) demonstrated a moderate staining pattern. TEM1/endosialin expression was occasionally noted in a subset of tumor stromal cells. Little to no staining was detected in reactive (non-neoplastic) brain tissue. Thus, high-grade primary and secondary brain tumors show similar expression profiles for TEM1/endosialin. These numbers are consistent with previous work [7], and suggest that the role of TEM1/endosialin is critical for vessel development in the brain, regardless of the tumor tissue of origin. Interestingly, 1 of 4 pilocytic astrocytoma speci-Localization of TEM1/endosialin expression Figure 1 Localization of TEM1/endosialin expression. A. Representative RT-PCR for TEM1/endosialin (TEM1) in whole cell lysates from two GBMs (T1, T2), two metastatic brain tumors (T3, T4) and two non-neoplastic controls (N1, N2). Expression was normalized to von Willebrand Factor (vWF), an endothelial positive control. B. Combined quantitative RT-PCR results for TEM1/endosialin in two controls (N), five GBMs (GBM) and five metastatic tumors (Met). Error bars, standard error of the mean. *p = 0.05. C. TEM1/endosialin was visualized by immunohistochemistry (IHC) using the TEM1/33 custom antibody or in situ hybridization (ISH) in glioblastoma (GBM), metastatic brain tumor (MET) and non-neoplastic brain tissues (Normal). Staining for vWF was used as an endothelial positive control. Original magnification for whole spot, 40×; detail, 100×. Pilocytic astrocytomas are low-grade astrocytomas, but can be highly angiogenic and often demonstrate enhancement and blood-brain barrier disruption on MRI scans.
As astrocytomas account for more than half of primary CNS malignancies, we expanded our examination to a larger group of primary astrocytic brain tumors in order to look for associations between TEM1/endosialin expression and clinically relevant pathologic features. We used in situ hybridization to probe expression of 275 arrayed grade IIgrade IV astrocytic tumors. The results are summarized in Table 2. Overall, TEM1/endosialin expression was detected in 79% of tumor samples (Table 2). When we focused on the grade IV astrocytomas (GBMs) alone, which constitute the most aggressive and lethal subtype, 31% of the tumors demonstrated moderate to strong staining for TEM1/endosialin (Table 2). These data confirm that TEM1/endosialin is widely expressed in astrocytic tumors. TEM1/endosialin expression in grade II-IV astrocytomas demonstrated a statistically significant inverse correlation with patient age (p = 0.004) ( Table 2), although it showed no correlation with other common measures such as tumor proliferation index (MIB-1) or apoptotic rate.

Co-localization of TEM1/endosialin and fibronectin
Fibronectin (FN), collagen I and collagen IV were recently identified as potential binding partners for TEM1/endosialin [11]. We therefore stained the same GBM/metastatic brain tumor tissue array for FN to see if there was a correlation between TEM1/endosialin and FN expression patterns. TEM1/endosialin and FN were co-upregulated in some tumors, although the correlation was not consistent (Table 1). A similar pattern of upregulation was noted for collagen I and collagen IV (YL, unpublished data). Using immunofluorescence, we detected limited co-localization of TEM1/endosialin and FN in GBM and metastatic brain tumor sections, suggesting that physiologic interactions between TEM1/endosialin and FN may occur, but that such interactions are not extensive in clinical settings of CNS neoplasia (Fig 2 and Additional file 1). Similar results were obtained using a custom synthesized, affinity purified anti-TEM1 antibody or commercial antibodies, targeting 3 different epitopes within the protein.

Regulation of TEM1/endosialin expression in vitro
TEM1/endosialin was originally characterized as an endothelial cell marker, although recent studies have provided evidence for its expression in endothelial progenitor cells, mesenchymal stem cells, tumor pericytes, all of which contribute to the tumor vasculature, and a subpopulation of tumor stromal cells [8,10,[17][18][19][20]. Co-immunofluorescence for TEM1/endosialin and the pericyte marker alpha-smooth muscle actin (αSMA) in the microvasculature of our GBMs and metastatic brain tumors revealed minimal overlap of the two proteins at the cellular level, although the markers did routinely track along the same vessels (Fig 2). Co-immunofluorescence for TEM1/endosialin and CD31, an endothelial marker, also demonstrated limited co-localization of the two proteins, although again both proteins were found on shared vessels (Fig 2). Our results differ in degree slightly from those of Simonavicius, et al. [5], but this may be due to differences in antibodies, tissue fixation conditions (Rouleau et al. experienced significantly different TEM1/endosialin staining results from FFPE versus frozen tumor tissues [21]) or clinical tumor heterogeneity.
We wished to determine factors that may induce TEM1/ endosialin in the endothelium of human brain tumors in order to further clarify the role of this protein. Little or no Tem1/endosialin induction was detected in human microvascular endothelial cells (HMVECs) or human brain microvascular endothelial cells (HBMECs) after exposure to pro-angiogenic growth factors, including vascular endothelial growth factor (VEGF), transforming growth factor beta (TGF-β), insulin-like growth factor 1 (IGF1), stem cell factor (SCF), stromal cell-derived factor 1 (SDF-1) and scatter factor/hepatocyte growth factor (SF/HGF) ( Fig 3A). Likewise, culturing primary endothelial cells under hypoxic conditions (2% oxygen for 24 hours) did not induce TEM1/endosialin expression (YL, unpublished data). Since it has been proposed that TEM1/endosialin may interact with proteins within the extracellular matrix as well as tumor stromal cells to promote vascular invasion and migration [11], we examined the effects of growth in matrigel on TEM1/endosialin expression in HMVEC cells. After 6 hours in matrigel, cells were harvested and TEM1/endosialin expression was assessed by quantitative RT-PCR using human specific primers. HMVEC cells cultured in matrigel induced TEM1/endosialin ten-fold compared to HMVEC cells cultured on plastic ( Fig 3B). No expression was detected in reverse transcriptase negative samples or in matrigel-only control isolates. This suggests that microvascular endothelial cells can induce TEM1/endosialin when exposed to a complex extracellular environment.

TEM1/endosialin is induced in intracranial GBM xenografts
Previous studies demonstrated robust TEM1/endosialin induction in B16 melanoma and HCT116 colon carcinoma xenografts grown in mice, while very low levels of TEM1/endosialin expression were reported in normal adult mouse brain [9]. We therefore sought to determine whether induction of TEM1/endosialin was conserved in an intracranial GBM xenograft model. Human U87MG glioblastoma cells were implanted intracranially into athymic nude mice. In situ hybridization using mouse specific riboprobes was used to measure mouse TEM1/ endosialin expression in the subsequently harvested tumors. The results demonstrated mouse TEM1/endosialin induction in the vasculature of the tumor, but not in the surrounding normal mouse brain ( Fig 4A). Staining for TEM1/endosialin was often concentrated near the margin of the tumor, although this observation did not reach statistical significance, a characteristic noted in human brain tumors as well (BNW, unpublished data).

Tem1/endosialin expression is not required for brain tumor growth in vivo
In order to investigate whether induction of Tem1/endosialin is required for tumorigenesis and/or angiogenesis in brain tumors, we injected U87MG human GBM cells intracranially into nude Tem1/endosialin KO and WT mice. No significant difference in survival time between the Tem1/endosialin KO and WT animals was detected ( Fig  4B). Neither was there a detectable difference in tumor take or volume. We were able to detect Tem1/endosialin in the vasculature of the WT, but not the KO, tumors, confirming the genotype of the mice.
Since TEM1/endosialin expression has been noted in tumor stromal cells, as well as vascular and perivascular cells, we screened the Tem1/endosialin KO and WT tumors for expression of human TEM1/endosialin. We did not detect appreciable TEM1/endosialin staining by in situ hybridization in the tumor stroma of either the knockout or wildtype xenografts. This suggests that growth of the intracranial xenografts was not supplemented by TEM1/ endosialin expression from the U87MG tumor cells themselves.
Nanda, et al., reported increased numbers of microvessels in xenografted colon carcinomas grown in the intestine of nude Tem1/endosialin knockout mice as compared to their wildtype counterparts [13]. We therefore examined the intracranial U87MG xenografts for a similar phenotype. The tumors from the Tem1/endosialin KO animals had approximately 40% more vessels than the tumors grown in the wildtype mice (Fig. 4C, D). This suggests that although GBM xenografts grew normally in the Tem1/ endosialin knockout animals, tumor angiogenesis was nevertheless altered. Staining with antibodies to αSMA did not demonstrate any difference in pericyte coverage between the WT and KO tumor vessels (MCW, unpublished data). So while Tem1/endosialin is not required for overall growth of this intracranial brain tumor model, it does appear to be required for normal development of the tumor cerebrovasculature, although the mechanism for this remains unclear.

Discussion
TEM1/endosialin/CD248 encodes a 165 kD single-pass transmembrane glycoprotein. It is a C-lectin type receptor with homology to thrombomodulin. Several studies identified TEM1/endosialin as a marker of tumor endothelium and isolated activated fibroblasts within the tumor stroma [7,9]. Little to no expression of TEM1/endosialin was noted in normal adult tissues. Alternate studies have argued that TEM1/endosialin in the tumor vasculature is expressed by vessel-associated pericytes [8,10,19]. Two additional reports identified TEM1/endosialin in populations of VEGFR2 + /CD31 + /CD45 -/VE-cadherin + endothelial precursor cells derived from CD133 + /CD34 + cells or in CD45 + /VE-cadherin + "vascular leukocytes" [20,22]. Both cell types have been proposed to contribute to tumor vascularization, although the degree to which either one incorporates into tumor vessels remains controversial. Human umbilical vein endothelial cells immortalized with SV40 large/small T-antigen, a model of proliferating neovascular endothelial cells, also express TEM1/endosialin [23]. We argue that TEM1/endosialin is expressed to varying degrees by tumor endothelial cells, pericytes and stromal fibroblasts. In normal blood vessels, the pericytes bind tightly to endothelial cells and influence migration, stabilization and maturation of the endothelium. In the cerebrovasculature, pericytes are additionally critical for maintenance of the differentiated blood-brain barrier (BBB) and failure of the BBB in brain tumors is associated with disruption and loss of pericytes lining endothelial cells. Pericytes are of emerging interest as therapeutic targets as vessels may be more sensitive to chemotherapy or radiotherapy when denuded of pericytes. When Bergers, et al., targeted both the endothelial cell and pericyte compartments with combinatorial receptor tyrosine kinase inhibitory (RTKI) therapies, they achieved more favorable responses than with either therapy alone [24]. Cao, et al., demonstrated that systemic overexpression of Ang2 blocked pericyte recruitment to tumor vessels and led to massive vascular destabilization and tumor suppression [25]. The collective data support the hypothesis that TEM1/endosialin has a favorable expression profile for targeting both the endothelial and pericytic portions of the tumor vasculature.
Expression of TEM1/endosialin by a subset of tumor fibroblasts may also influence tumor growth and angiogenesis. Activated tumor fibroblasts contribute to the deposition of fibronectin, collagen I and collagen IV, all of which may bind TEM1/endosialin and guide vessel invasion, migration or differentiation. SDF-1 expression by activated tumor fibroblasts has been shown to drive tumor progression and angiogenesis in breast carcinoma models and has been reported to recruit endothelial progenitor cells to the microvasculature of murine gliomas, where differentiation into both pericytes and endothelium was detected [26,27]. A recent genetic study reported that TGFβ1 treatment of proliferating endothelial cells led to an endothelial-mesenchymal transition, wherein endothelial cells downregulated CD31 and upregulated fibroblast-specific protein-1 (FSP-1). They confirmed the presence of FSP1 + /CD31 + intermediate cells in tumor models [28]. Further work is necessary to fully determine the complete expression profile of TEM1/endosialin and its relationship to these intricately related stromal cell populations.
In the US, primary brain tumors are diagnosed in approximately 40,000 new patients a year. In addition, brain metastases occur in 20-40% of cancer patients, leading to 200,000-400,000 new cases of brain metastases a year. Despite differences in the molecular fingerprints of the tumor cells themselves, it may be possible to identify Immunofluorescence for TEM1/endosialin: co-localization in tumor specimens common changes in the vasculature of these disparate tumor types. Initial studies by ourselves and others reported TEM1/endosialin upregulation in human brain tumors [1,5,7]. We wanted to expand these findings in order to better establish the full extent of TEM1/endosialin expression in brain tumors. This study confirms expression of TEM1/endosialin at the RNA and protein level in astrocytic and metastatic brain tumors, with little to no expression in normal control brain. Expression was primarily limited to the tumor vasculature but occasional Tem1/endosialin expression in mouse brain tumor models stromal cells also stained for TEM1/endosialin. Greater than 30% of GBMs displayed robust TEM1/endosialin expression, while lesser percentages of lower grade tumors expressed TEM1/endosialin. Unfortunately, we did not know whether the individual specimens analyzed were from the tumor center or closer to the tumor margin. Preliminary mouse xenograft data suggested that Tem1/ endosialin staining was increased along the tumor margin, but this needs to be further investigated in both clinical and mouse tumors. It will be especially useful to assess Tem1/endosialin expression in future experiments using xenografted primary glioma spheroids or glioma stem cells, which grow as a more highly invasive intracranial tumor than the U87 model, although with less pronounced angiogenesis. Interestingly, one pilocytic astrocytoma displayed robust expression of TEM1/endosialin. Although pilocytic astrocytomas are considered lowgrade tumors, they are highly angiogenic and often enhance on MRI, which is indicative of vascular leakiness, disruption of the blood-brain barrier and often more aggressive disease. A recent study in breast cancer identified an association between TEM1/endosialin expression and progressive and/or recurrent disease as well as nodal involvement, all hallmarks of a more aggressive disease state [6]. Thus, TEM1/endosialin may have prognostic value for breast cancer. We did detect a statistically significant (p = 0.004) inverse correlation between TEM1/endosialin expression in astrocytomas and patient age, although the implications of this remain unclear. While diagnosis at a younger age can be a predictor of long-term survival for GBM [29], our initial analyses reveal no difference in survival between TEM1/endosialin expressing and non-expressing tumors.
Tomkowicz, et al. recently identified collagen types I and IV and fibronectin as potential ligands for TEM1/endosialin [11]. They showed that CHO cells engineered to overexpress TEM1/endosialin (CHO-TEM1) exhibited greater adhesion to fibronectin than control CHO cells (CHO-K1). They suggested a model by which TEM1/endosialin is induced in cells associated with endothelium and stromal fibroblasts. The subsequent interactions between TEM1/endosialin expressing cells and fibronectin, collagen I or IV present in the extracellular matrix drive cellular migration required for tumor invasion and angiogenesis.
In an independent study, TEM1/endosialin and collagen IV were shown to colocalize in finger-like processes during angiogenic development in the human fetal telencephalon [30]. We had identified collagen I and collagen IV as putative glioma endothelial markers which exhibited a minimum 4-fold induction as compared to normal endothelial cell samples and were induced in both glioma and colon carcinoma endothelium [1]. Therefore, we were curious to determine whether we could detect co-localization of fibronectin with TEM1/endosialin in our primary and secondary tumor samples. We did detect upregulation of both TEM1/endosialin and fibronectin within the same tumor specimens, although only a subset of the samples showed co-localization along the vasculature, supporting a physiologic interaction of these proteins. In addition, Tomkowicz and colleagues showed that the CHO-TEM1 cells, but not CHO-K1 control cells, grew with web-like morphology and formed large clusters when cultured in matrigel. The CHO-TEM1 cells also exhibited enhanced migration through matrigel that was blocked by incubation with a humanized antibody targeted against TEM1/ endosialin [11]. Others have shown that incubation with anti-TEM1/endosialin antibodies can block tube formation by endothelial progenitor cells and mesenchymal stem cells [17,20]. Our in vitro data showed that human endothelial cells grown in matrigel induce TEM1/endosialin themselves in response to the culture environment. Together, the biochemical and localization data support a functional role for TEM1/endosialin in promoting cellcell interactions and migration during tumor angiogenesis.
Since the expression profile of TEM1/endosialin and existent xenograft data suggested TEM1/endosialin might be a viable therapeutic target for brain tumors, we wished to ascertain whether TEM1/endosialin induction was conserved in an orthotopic model of GBM. Intracranial U87MG xenografts demonstrated robust upregulation of Tem1/endosialin in the tumor vasculature with little to no expression in non-neoplastic brain. Again, staining was predominantly present in the vessels, although we could detect staining in occasional tumor stromal cells. Interestingly, when we implanted U87MG cells engineered to stably overexpress VEGF, a small percentage of tumor cells stained positive for human TEM1/endosialin, although there was no significant difference in Tem1/endosialin staining in the mouse vessels (BNW, unpublished data).
When we implanted intracranial GBM xenografts into nude Tem1/endosialin KO and WT mice, we detected no difference in tumor take or survival. This differs from the colon carcinoma data and suggests that the impact of Tem1/endosialin in the cerebral microenvironment may differ from its role in intestinal tumorigenesis. Nanda, et al., reported a significant decrease in tumor take and volume when they implanted colon carcinoma xenografts into the large intestine of nude Tem1/endosialin KO mice [13]. However, they did not detect a difference in tumor growth when identical cells were implanted subcutaneously into KO and WT animals. They argued that this emphasized the importance of an anatomically relevant microenvironment for tumor growth and study. This is a similar argument to that proposed by Camphausen, et al., who compared the expression profiles of U87MG and U251MG GBM cell lines grown in vitro to those from cells grown as subcutaneous or intracranial xenografts [31]. They noted that the expression profiles in tissue culture were significantly different from the subcutaneous profiles, which were, in turn, distinct from the intracranial expression profiles. The disparity between GBM xenograft expression profiles suggested that the orthotopic implantation site can dictate gene expression, presumably due to signaling between the tumor cells and the surrounding stroma. In the current scenario, it would again appear that TEM1/endosialin expression levels within distinct anatomical compartments differentially influence tumor growth and pathology. Importantly, the data suggest that inhibition of TEM1/endosialin function alone in tumors is insufficient for suppression of brain tumor growth.
Despite unchanged tumor growth and equivalent animal survival between Tem1/endosialin WT and KO animals, we did detect an increase in the number of microvessels present in the KO GBM tumors. This strengthens a role for TEM1/endosialin in the maturation of the tumor neo-vasculature [13]. TEM1/endosialin was recently reported to be upregulated by hypoxia via a HIF-2 pathway in a limited number of cell lines, although we could detect no induction in primary human endothelial cells [32]. However, the brain is a highly vascularized site, and it may be that the level of oxygenation is sufficient to promote xenograft growth in the absence of Tem1/endosialin induction. It is more likely TEM1/endosialin is directly or indirectly required for differentiation or maturation of the vasculature, as opposed to strict proliferation, and that lack of TEM1/endosialin protein leads to the overabundance of vessels demonstrated in the knockout mouse tumors. Further studies are needed to determine if the increased vascularity of the Tem1/endosialin KO tumors could render them more sensitive to radiation or chemotherapy, by enhancing the delivery of radiosensitizing oxygen or therapeutic drugs.

Conclusion
The combined tumor data confirm that TEM1/endosialin contributes to the development of the brain tumor vasculature. Although lack of TEM1/endosialin expression did not inhibit brain tumor growth in our model, the cell surface bioavailability of TEM1/endosialin and its localization to tumor and tumor vasculature components in both primary and metastatic brain tumors, both diseases with limited current treatment options, support efforts to target this protein with alternative approaches, including antibody or antibody-cytotoxin based strategies, such as a recently reported anti-TEM1/endosialin IgG conjugated to saporin, and other efforts in pre-clinical development [11,21,33] or as combination therapy.