Abstract
To gain a molecular understanding of tumor angiogenesis, we compared gene expression patterns of endothelial cells derived from blood vessels of normal and malignant colorectal tissues. Of over 170 transcripts predominantly expressed in the endothelium, 79 were differentially expressed, including 46 that were specifically elevated in tumor-associated endothelium. Several of these genes encode extracellular matrix proteins, but most are of unknown function. Most of these tumor endothelial markers were expressed in a wide range of tumor types, as well as in normal vessels associated with wound healing and corpus luteum formation. These studies demonstrate that tumor and normal endothelium are distinct at the molecular level, a finding that may have significant implications for the development of anti-angiogenic therapies.
Tumors require a blood supply for expansive growth (1–3), an observation that has stimulated a profusion of research on tumor angiogenesis. However, several basic questions about tumor vessels remain unanswered. For example, is endothelium that lines blood vessels in tumors qualitatively different from endothelium in vessels of normal tissue? What is the relation of tumor angiogenesis to angiogenesis associated with wound healing or other physiological processes? The answers to these questions critically impact the potential for new therapeutic approaches to inhibit angiogenesis in a tumor-specific manner.
To determine if tumor-specific endothelial markers exist, we compared gene expression profiles in endothelium derived from normal and tumor tissue. Human colorectal cancer was chosen for these studies because it has a high incidence, tends to grow slowly, and is often resistant to chemotherapeutic drugs. Importantly, the progressive growth of this tumor type appears to be angiogenesis-dependent (4).
Global analysis of gene expression in tumor and normal endothelium is difficult because (i) the endothelium is enmeshed in a complex tissue consisting of vessel wall components, stromal cells, and epithelial cells, and (ii) only a small fraction of the cells within these tissues are endothelial. Thus, we needed to develop methods for isolating highly purified endothelial cells (ECs) and for evaluating gene expression profiles from relatively few cells.
To overcome the first obstacle, we attempted to purify ECs from dispersed human colorectal tissue using CD31, an endothelial marker commonly used for this purpose (5–8). This resulted in a substantial enrichment of ECs but also in contamination of the preparations by hematopoietic cells, most likely due to expression of CD31 by macrophages (9). We therefore purified ECs from human tissues using P1H12, a recently described marker for ECs (10). Unlike CD31, CD34, and VE-cadherin, P1H12 specifically localized to ECs of all vessels including microvessels of normal and cancerous colorectal tissue (Fig. 1A). Our purification protocol also optimized the detachment of ECs from neighboring cells, leaving cell surface proteins intact, and included positive and negative affinity purifications using a mixture of antibodies (Fig. 1B). The ECs purified from normal colorectal mucosa and colorectal cancers were essentially free of epithelial and hematopoietic cells as judged by reverse transcription–polymerase chain reaction (RT-PCR) (Fig. 1C) and subsequent gene expression analysis (see below).
Purification of ECs from human normal and malignant tissue. (A) Vessels (red) of frozen sections were stained by immunofluorescence with anti-P1H12 (Chemicon, Temecula, California) and detected with a biotinylated goat anti-mouse IgG secondary antibody followed by rhodamine-linked strepavidin. The vessels stained are from within the lamina propria of normal colonic mucosa. E-cadherin–positive epithelial cells (green) at the edge of the crypt were simultaneously visualized with a rabbit polyclonal antibody (Santa Cruz Biotechnology, Santa Cruz, California), followed by a goat anti-rabbit IgG secondary antibody labeled with Alexa (Molecular Probes, Eugene, Oregon). Bar, 50 μM. (B) For isolation of pure EC populations from collagenase-dispersed tissues, the epithelial and hematopoietic cell fractions were sequentially removed by negative selection with magnetic beads. The remaining cells were stained with P1H12, and ECs were isolated by positive selection with magnetic beads (39). (C) RT-PCR analysis used to assess the purity of the EC preparations. Semiquantitative PCR analysis was performed on cDNA generated directly from colorectal cancer tissue (Unfractionated Tumor) or from purified ECs isolated from normal colonic mucosa (Normal Endothelial Fraction) or colorectal cancer (Tumor Endothelial Fraction). Expression of the epithelial-specific transcript cytokeratin 20 (CK20) was limited to the unfractionated tumor. Two endothelial-specific transcripts, vWF and VE-cadherin (VE-Cad), showed robust amplification only in the endothelial fractions, whereas transcripts corresponding to the ubiquitous housekeeping enzyme glyceraldehyde phosphate dehydrogenase (GAPDH) were amplified in all samples. No signal was detected in the no-template (N.T.) control. cDNA templates were diluted 1:10, 1:100, 1:1000, 1:4000, and 1:40,000, as indicated by the declining wedge. (D) The relative expression level of select genes was determined by measurement of the tag abundance from several SAGE libraries combined into four groups. The first was composed of ∼193,000 tags from the two in vivo–derived EC preparations (Endothelial Cell Fraction), whereas the second contained a single library of ∼57,000 tags containing macrophages and other leukocytes derived from the negative selection (Hematopoietic Fraction). The third group contained ∼401,000 tags from cultured HUVEC and HMVEC (Endothelial Cells in Culture), and the fourth consisted of ∼748,000 tags from six colon cancer cell lines in culture (Epithelial Cells). After normalization, the library with the highest tag number for each marker was given a value of 100%, and the corresponding relative expression levels of the remaining three libraries were plotted on the ordinate. A high level of CD31 is apparent on hematopoietic cells, the likely cause of the impurity of the initial endothelial selection, compared with the selectivity of P1H12.
To evaluate gene expression, we used a modification of the serial analysis of gene expression (SAGE) technique (11). SAGE associates individual mRNA transcripts with 14–base pair (bp) tags derived from a specific position near their 3′ termini (12). The abundance of each tag provides a quantitative measure of the transcript level present within the mRNA population studied. SAGE is not dependent on preexisting databases of expressed genes and therefore provides an unbiased view of gene expression profiles. This feature is particularly important in the analysis of cells that constitute only a small fraction of the tissue under study, as transcripts from these cells are unlikely to be well represented in extant expressed sequence tag (EST) databases.
We generated a SAGE library of ∼96,000 tags from the purified ECs of a colorectal cancer and a similar library from the ECs of normal colonic mucosa from the same patient. These ∼193,000 tags corresponded to over 32,500 unique transcripts (13). The expression pattern of hematopoietic, epithelial, and endothelial markers confirmed the purity of the preparations (Fig. 1D).
We next identified pan endothelial markers (PEMs), that is, transcripts that were expressed at substantially higher levels in both normal and tumor-associated endothelium compared with other tissues. Tags expressed at similar levels in both tumor and normal ECs were compared with ∼1.8 million tags from a variety of cell lines derived from tumors of nonendothelial origin. This simple comparison identified 93 transcripts that were expressed at levels at least 20-fold higher in ECs in vivo compared with nonendothelial cells in culture (14). Among the most abundant transcripts, there were 15 tags corresponding to characterized genes, 12 of which had been previously shown to be preferentially expressed in endothelium (10, 15–26), and the other 3 genes not previously associated with endothelium (Table 1). These data also revealed many novel PEMs, which became increasingly prevalent as tag expression levels decreased (Table 1). We validated the expression of selected PEMs in vivo using a highly sensitive nonradioactive in situ hybridization method that allowed the detection of relatively rare transcripts from frozen sections of human tissues (27). Expression of PEM3 and PEM6 was limited to vascular ECs in both normal and neoplastic tissues (Fig. 2, A and B). For other PEMs, their endothelial origin was confirmed by SAGE analysis of ∼401,000 transcripts derived from primary cultures of human umbilical vein endothelial cells (HUVEC) and human dermal microvascular endothelial cells (HMVEC) (Table 1). These data also suggest that ECs maintained in culture do not completely recapitulate expression patterns observed in vivo. For example, Hevin and several other PEMs were expressed at high levels in both tumor and normal ECs in vivo, but few or no transcripts were detected in cultured HUVEC or HMVEC (Table 1). The source of the Hevin transcripts was confirmed to be endothelium by in situ hybridization in normal and malignant colorectal tissue (Fig. 2C).
Expression of PEMs is limited to ECs. (Ato C) The endothelial origin of PEMs identified by SAGE was confirmed by a highly sensitive in situ hybridization assay (27). Localization of novel PEMs to the ECs (red stain) was demonstrated by examining two representative PEMs, PEM3 (A) and PEM6 (B), in lung cancer and colon cancer, respectively. Hevin expression was readily detected in the ECs of a colon tumor (C) despite its low level of expression in cultured ECs. Bars, 50 μM.
SAGE analysis reveals previously characterized and novel pan endothelial markers. The most abundant characterized or novel tags derived by summing the tags from normal EC (N-ECs) and tumor EC (T-ECs) SAGE libraries are listed in descending order. For comparison, the corresponding number of SAGE tags found in HUVEC and HMVEC endothelial cell cultures, and several nonendothelial cell lines (14), are shown. Tag numbers for each group were normalized to 100,000 transcripts. A description of the gene product corresponding to each tag is given, followed by alternative names in parentheses. Some uncharacterized genes have predicted full-length coding sequence. The sequence CATG precedes all tags, and the 15th base (11th shown) was determined as described (38).
We next identified transcripts that were differentially expressed in endothelium derived from normal or neoplastic tissues. This comparison revealed 33 tags that were elevated at least 10-fold in normal endothelium and 46 tags that were elevated 10-fold or more in tumor endothelium (28). Because transcripts expressed at higher levels in tumor endothelium are most likely to be useful for diagnostic and therapeutic purposes, our subsequent studies focused on this class. Of the top 25 tags most differentially expressed, 12 tags corresponded to 11 previously identified genes, 1 containing alternative polyadenylation sites (Table 2). Of these 11 genes, 6 were previously recognized as markers of angiogenic vessels (16,22, 29–33), and at least 7 encode proteins involved in extracellular matrix formation or remodeling. These matrix-related processes are likely to be critical to the growth of new vessels. The remaining 14 tags corresponded to uncharacterized genes, most of which have been deposited as ESTs (Table 2).
SAGE tags elevated in tumor endothelium. The top 25 tags with the highest tumor EC (T-ECs) to normal EC (N-ECs) tag ratios are listed in descending order. To calculate tag ratios, we assigned a value of 0.5 in cases where zero tags were observed. The SAGE libraries are the same as those listed in Table 1. Tag numbers for each group were normalized to 100,000 transcripts. A description of the gene product corresponding to each tag is given, followed by alternative names in parentheses. TEM9 was uncharacterized at the outset of these studies but was recently characterized as a lectin present on ECs in culture and on microvascular vessels of human placenta in vivo (41).
To validate the expression patterns of these genes, we focused on nine tumor endothelial markers (TEM1 through TEM9) for which EST sequences but no other information was available (Table 2). RT-PCR analysis was used to evaluate the expression of the corresponding transcripts in purified ECs derived from normal and tumor tissues of two patients different from the original one used to construct the SAGE libraries. As expected on the basis of the SAGE data, von Willebrand Factor (vWF) and PEM6 were expressed at similar levels in normal and tumor ECs from both patients. These controls were not detected in purified tumor epithelial cells (Fig. 3A). In contrast, all nine TEMs chosen for this analysis were prominently expressed only in tumor ECs, but were absent or barely detectable in normal ECs (Fig. 3A). These RT-PCR assays were sensitive indicators of expression, and the absence of detectable transcripts in the normal endothelium, combined with their presence in tumor endothelial RNAs even when diluted 100-fold, provides important confirmation of their differential expression. These results also show that these transcripts were not simply expressed differentially in the ECs of the original patient, but were characteristic of colorectal cancer endothelium in general.
Expression of TEMs. (A) RT-PCR analysis confirmed the tumor-specific expression of novel TEMs. Semiquantitative PCR analysis was performed on cDNA generated from purified tumor epithelial cells as a negative control (Control) or from purified ECs isolated from normal colonic mucosa (Normal ECs) or colorectal cancer (Tumor ECs) from two different patients. Two endothelial-specific markers, vWF and PEM6, showed robust amplification only in the endothelial fractions, whereas GAPDH was observed in all samples. TEM1, TEM7, and TEM9 were specifically expressed in tumor ECs. The cDNA template was diluted 1:10, 1:100, 1:1000, and 1:10,000, as indicated by the declining wedge. (B to J) The endothelial origin of TEMs identified by SAGE was confirmed by in situ hybridization, as in Fig 2. Expression of TEM1 (B) and TEM7 (C) was highly specific to the ECs in colorectal cancers; sections were imaged in the absence of a counterstain to show the lack of detectable expression in the nonendothelial cells of the tumor. TEM7 was also expressed in ECs from a metastastic liver lesion (D) arising from a primary colorectal cancer, and primary tumors derived from lung (E), breast (F), pancreatic (G), and brain cancer (H), as well as in a sarcoma (I). TEM7 expression was also localized to vessels during normal angiogenesis of human corpus luteum (J). Bars, 50 μM.
To exclude the possibility that the differentially expressed transcripts were derived from contaminating nonendothelial cells, we performed in situ hybridization on normal and neoplastic colon tissue. In every case where transcripts could be detected (TEM 1, 3, 4, 5, 7, 8, and 9), they were specifically localized to ECs (34) (Fig. 3, B and C). Although caution must be used when interpreting negative in situ hybridization results, none of the TEMs were expressed in vascular ECs associated with normal colorectal tissue even though vWF and Hevin were clearly expressed (34).
To determine whether TEMs were specifically expressed in the endothelium from primary colorectal cancers, or whether they were characteristic of tumor endothelium in general, we studied the expression of a representative TEM (TEM7) in a liver metastasis from a colorectal cancer, a primary sarcoma, and in primary cancers of the lung, pancreas, breast, and brain. As shown in Fig. 3, the TEM7 transcript was expressed specifically in the endothelium of each of these cancers, whether metastatic (Fig. 3D) or primary (Fig. 3, E to I). Analysis of the other six TEMs (TEM 1, 3, 4, 5, 8, and 9) revealed a similar pattern in lung tumors, brain tumors, and metastatic lesions of the liver (34).
Finally, we investigated whether these transcripts were expressed in angiogenic states other than that associated with tumorigenesis. As assessed by in situ hybridizations, these transcripts were generally expressed both in the corpus luteum and in the granulation tissue of healing wounds (34) (Fig. 3J). One possible exception is TEM8, which we failed to detect in corpus luteum. In all tissues examined, expression of the genes was either absent or confined to the EC compartment.
The studies described above provide a definitive molecular characterization of ECs in an unbiased and general manner. They lead to several important conclusions that have direct bearing on long-standing hypotheses about angiogenesis. First, normal and tumor endothelium are highly related, sharing many endothelial cell–specific markers. Second, the endothelium derived from tumors is qualitatively different from that derived from normal tissues of the same type and is also different from primary endothelial cultures. We identified 46 transcripts that were expressed at substantially higher levels (>10-fold) in tumor endothelium than in normal endothelium, and 33 transcripts that were expressed at substantially lower levels in tumor than in normal endothelium. Most of these genes were either not expressed or were expressed at relatively low levels in ECs maintained in culture. Third, these genes are characteristically expressed in tumors derived from several different tissue types, demonstrating that tumor endothelium, in general, is different from the endothelium in surrounding normal tissue. Fourth, most of the genes expressed differentially in tumor endothelium are also expressed during angiogenesis of corpus luteum formation and wound healing. This is consistent with the idea that tumors may recruit vasculature by means of the same signals elaborated during other physiological or pathological processes. Indeed, the notion that tumors represent “unhealed wounds” is one of the oldest ideas in cancer biology (35). However, the fact that TEM8 expression was not detectable in developing corpus luteum suggests that there may be discrete differences between tumor angiogenesis and normal angiogenesis.
Finally, it is perhaps not surprising that so many of the endothelial-specific transcripts identified in this study, whether expressed only in neovasculature or in endothelium in general, have not been previously characterized, and some are not even represented in EST databases. ECs represent only a minor fraction of the total cell population within normal or tumor tissues, and only those EC transcripts expressed at the highest levels would be expected to be represented in libraries constructed from unfractionated tissues. The genes described in the current study should therefore provide a valuable resource for basic and clinical studies of human angiogenesis in the future.
↵* To whom correspondence should be addressed. E-mail: kinzlke{at}jhmi.edu
