1 Beth Israel Deaconess Medical Center, Harvard Medical School, Boston 02115
2 Deputy Editor, Physiological Genomics
3 Editor-in-Chief, Physiological Genomics
4 Department of Medicine, Brigham and Womens Hospital, Harvard Medical School, Boston, Massachusetts
THE DEVELOPMENT AND USE of microarray technology in cell biology and physiology has tremendous potential in the elucidation of pathways involved in normal and disease states. However, the very magnitude of these studies means that an investigator may find an overwhelming number of genes that appear to be differentially regulated between two or more states. Therefore, an important starting point in designing and interpreting array experiments is to minimize the frequency of false-positive results. In general, this is accomplished through the tight control of culture conditions and/or animal subjects, the use of biological and/or technical replicates, or the use of various filtering strategies. Other strategies include tightening the level of significance, imposing fold-change criteria, demanding 100% "present" calls and discarding genes where replicates exhibit less than 100% concordance. While decreasing the number of hits and eliminating false-positive results, these latter criteria also increase the likelihood of false-negative results. In short, an investigator must choose to either analyze a vast pool of candidate genes or risk discarding potentially useful, biologically relevant hits.
A report by Gerritsen et al. (6) in this online release of Physiological Genomics (Ref. 6; see page 13 in this release) illustrates the magnitude of this problem and offers a potential solution that may prove applicable to a wide number of studies. This manuscript involves the identification of genes expressed in endothelial cells that are involved in the induction of angiogenesis.
Endothelial cells display remarkable heterogeneity in health and disease (5, 7, 10, 13). Such heterogeneity provides unique opportunities for developing site-directed therapies. As a gatekeeper to the underlying tissue, the endothelium is a highly accessible and attractive therapeutic target. The presence of phenotypic differences, for example, at the level of cell surface receptors, may be exploited to deliver drugs or genes to defined vascular beds. Therefore, an important goal is to map endothelial cell phenotypes under normal and pathophysiological conditions.
Endothelial cell phenotypes are governed largely by signals residing in the extracellular environment (15). When endothelial cells are harvested and grown in tissue culture, they are uncoupled from these critical extracellular cues and undergo phenotypic drift. The net result is a loss of complexity and site-specific signatures.
Given that the cultured endothelial cells are phenotypically modulated with culture, how does one identify genes relevant to an in vivo phenotype? There are several approaches to this problem. One is to utilize proteomics to identify novel cell surface receptors within intact endothelium. For example, antibody and subfractionated strategies have been employed to generate monoclonal antibodies that specifically target the caveolae in one vascular bed or another (9). Others have used phage display peptide libraries to select for peptides that home to specific vascular beds in vivo (3, 11). These latter studies have uncovered a vascular "address system" that allows for site-specific targeting of biologically active compounds, for example, to the endothelial lining of tumor blood vessels (2, 4). These proteomic approaches are valuable in that they are conducted in the native environment of the endothelial cell and they select for cell surface molecules that may be amenable to therapeutic targeting.
In general, proteomic strategies do not yet have the power or breadth of genomic screens. How then can one apply recent advances in genomics to an understanding of cell type-specific function, when that function is so tightly coupled in both time and space to the environment in vivo? This question has received increasing attention in the angiogenesis field. One approach has been to compare gene expression profiles in endothelial cells from normal tissue and tumor tissue (14). This strategy involves the isolation of relatively pure populations of endothelial cells but is limited by the potential for contamination with nonendothelial cells and by the inability to control for altered gene expression in the ex vivo setting. Another approach has been to screen for genes that are expressed in endothelial cells undergoing tube formation in vitro, a model that mimics certain steps in angiogenesis, including endothelial cell migration, proliferation, and extracellular matrix interactions. Two independent groups have successfully employed a three-dimensional collagen gel system to identify genes that are upregulated during the process of in vitro angiogenesis (1, 8). Interestingly, the transcriptional profiles in these latter two studies differed to some extent, suggesting that subtle changes in the conditions, such as the absence or presence of phorbol ester or perhaps the source of endothelial cells, may influence gene expression.
Gerritsen and colleagues (6) have extended these studies to identify genes that are differentially expressed in tumor endothelium. The authors employed three different models of human umbilical vein endothelial cell (HUVEC) tube formation. The assays differed in their matrix (collagen vs. fibrin) and growth factor composition [vascular endothelial growth factor (VEGF) + basic fibroblast growth factor + phorbol ester vs. VEGF + hepatocyte growth factor]. Affymetrix microarrays were used to compare mRNA transcripts differentially expressed in the three models to reference expression profiles generated from the RNA of endothelial cells in monolayer. The experiments were carried out in triplicate at five distinct time points and repeated in HUVEC derived from three independent donors. Cognizant of the importance of the microenvironment in modulating cell phenotype, the investigators took care to employ identical lot numbers of growth factors, plastic ware, and matrices.
Several years ago, we hypothesized that the use of different but related models would provide a convenient filter for large databases (12). By demanding concordance across multiple models, model-specific genes within the gene set would be predicted to drop out, as might many of the associated, noncausally related genes. Using a similar approach, Gerritsen et al. (6) began by identifying those genes that were upregulated in all three in vitro models of tube formation by at least a factor of 2. Despite the improved level of stringency, the studies initially yielded an overwhelming list of 1,038 genes, pointing to the need for additional filters.
To further refine the list of regulated genes, the authors followed several critical steps. First, bioinformatics programs were utilized to narrow the list to those genes with the predicted structure of either transmembrane or secreted proteins (n = 397). To identify genes that may be involved in tumor angiogenesis, total RNA was obtained from six different colon adenocarcinoma samples for microarray analysis compared with normal colon mucosa. By overlapping the resulting list of colon tumor-specific genes with the list of (predicted) transmembrane and secreted proteins, Gerritsen et al. (6) generated a panel of 128 transcripts. The last step was to subtract all those transcripts expressed in the colon epithelial tumor cell lines to enrich for endothelial and stromal cell-specific genes, resulting in a final list of 24 tumor angiogenesis-associated genes.
As a means to validate the patterns of expression, Gerritsen et al. (6) examined one of the upregulated genes, STC-1, which is homologous with a fish protein involved in calcium and phosphate regulation, to determine its profile in angiogenesis. Pellets containing VEGF, which stimulates angiogenesis, were implanted in rat corneas. After 6 days, the mRNA levels of STC-1 were compared with control corneas, revealing significantly higher levels in the corneas undergoing angiogenesis. Furthermore, in situ hybridization with antibodies against STC-1 revealed that the protein is highly expressed in the blood vessels of colon adenocarcinomas.
This study demonstrates the rational use of data filters and higher order screening to provide a bridge between the ex vivo biology of cultured endothelial monolayers and the complexity and heterogeneity of the intact endothelium. As a starting point, the authors focused on a model of in vitro tube formation as a crude approximation of angiogenesis. The identification of genes that are common to three models of tube formation provides a high degree of stringency and thereby reduces the likelihood of false-positive markers. The selection of genes that encode for cell surface receptors or secreted proteins is not only helpful in narrowing down the number of candidate genes, but, as with the proteomic approaches, facilitates the identification of markers with therapeutic or diagnostic relevance. Finally, by overlapping the genes with those from colon adenocarcinoma and subtracting out the transcripts from the tumor cells themselves, the investigators theoretically enriched for genes that are specific for tumor-associated angiogenesis. Continued efforts to validate these putative angiogenesis markers may lead to new diagnostic and/or therapeutic avenues while at same time providing mechanistic insight into neovessel formation.
Several important questions remain to be answered. For example, it will be interesting to compare the level of expression of these 24 markers in blood vessels of tumors, healing wounds, corpora lutea, and embryos. In view of the importance of the microenvironment in dictating endothelial cell phenotypes, it is also reasonable to predict that the repertoire of genes expressed in the endothelium of colon tumors will differ from that of other tumors. Indeed, the approach described by Gerritsen et al. (6) should prove useful in interrogating transcriptional profiles of other tumor beds.
In addition, it will be very important to determine which of these 24 genes may encode proteins that are playing a causal role in angiogenesis. Gain-of-function or loss-of-function experiments both in cell culture and in vivo will be necessary to address this issue.
FOOTNOTES
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).
Address for reprint requests and other correspondence: W. C. Aird, Beth Israel Deaconess Medical Center, 330 Brookline Ave., RW-663, Boston, MA 02115 (E-mail: waird{at}caregroup.harvard.edu).
10.1152/physiolgenomics.00071.2002.
REFERENCES