Subtractive Transcriptomics: Establishing Polarity Drives In vitro Human Endothelial Morphogenesis

The growth of new capillaries from preexisting networks (angiogenesis) is an essential component of embryogenesis and reinitiates in the adult animal during tissue growth and repair processes, such as wound healing and the menstrual cycle. Misregulation of this tightly controlled process has been shown in multiple pathologic conditions in humans, including retinopathy, rheumatoid arthritis, and tumor progression, in which it is instrumental in tumor invasion and metastasis. It is driven by a highly complex and orchestrated genetic program in which endothelial cell proliferation combines with branching morphogenesis, the latter of which requires major changes in both functional and spatial cellular organization. Little is known about the proteins regulating changes in cell shape, long-range migration, and communication necessary for capillary formation that are associated with actual morphogenesis. These proteins are far more likely to be specific to angiogenesis than those involved in endothelial cell proliferation, which currently represent the majority of drug targets of angiogenic-modulating drug candidates. Consequently, a method for identifying morphogenesis-selective proteins is critical to further development of antiangiogenic therapies.

The molecular dissection of complex cellular processes is one of the major challenges in postgenomics biology. Several in vitro models have been developed over the past few years, which have served as high-throughput primary assays to test both proangiogenic and antiangiogenic agents (see the pro and con discussion of these models in ref. 1). Previous microarray analyses of collagen- or Matrigel-induced tube formation in vitro by other groups (2–6) have focused on either proliferation-associated transcripts or used a single time point, plus/minus approach, followed by bioinformatic dissection of statistically high-scoring (i.e., highly expressed) transcripts. Although these studies have provided valuable insight into characteristics that define the phenotype of mature endothelial cells, absent from our knowledge is much of the temporally regulated information inherent to a dynamic process, such as vascularization, and the identification of proteins that are expressed transiently and/or weakly that may be central to the initiation of angiogenesis.

To identify proteins potentially involved strictly in the regulation of endothelial cell “morphogenesis” as opposed to “proliferation,” gene expression in pooled human microvascular endothelial cells (HMVEC) undergoing the two processes in vitro were contrasted and compared. When cultured on gelatin-coated plastic, these cell explants undergo proliferation. Alternately, when seeded at an appropriate density on a gel composed of extracted basement membrane derived from mouse Engelbreth-Holm-Swarm sarcoma (Matrigel), the endothelial cells migrate into the matrix and proceed to form capillary-like lumen-containing structures (7). It has been shown that malignant tumor cells recruit vasculature through both the production and the secretion of growth factors and interaction with locally activated host microenvironment. Tumors activate angiogenesis in a polymorphic manner, with a wide variation in the diameter and density of the formed vessels (see ref. 8 as an example) possibly due to the variation seen in the class and level of growth factors up-regulated as well as the wide variation in local environmental factors present. The Matrigel-driven tubulogenesis model was chosen for this preliminary array analysis to reduce nonendothelial transcript background and to provide initial drug target candidates specific for morphogenesis that could be followed up in later in vivo studies. Comparison of the two temporal processes provides the ability to separate the proliferation phase of angiogenesis from early network formation in a simple in vitro model system and offers the opportunity to identify important proteins up-regulated early on in capillary morphogenesis yet not during the growth process.

Reagents and cell culture. Pooled, neonatal dermis HMVECs were obtained from Cascade Biologics, Inc. (Portland, OR) and maintained as recommended. Recombinant human vascular endothelial growth factor (rhuVEGF) was obtained from Alpha Diagnostics International (San Antonio, TX), Matrigel was from Becton Dickinson Co. (Franklin Lakes, NJ), and TRIzol reagent was from Invitrogen (Frederick, MD).

Formation of three-dimensional tubes on Matrigel. Subconfluent HMVECs at passage 5 to 7 were plated onto the Matrigel-coated dishes (with addition of purified rhuVEGF at 50 ng/mL to the medium and the Matrigel) at 2.25 × 10⁴ cells/cm². Control plates of proliferating subconfluent cells were washed with PBS thrice and overlaid with medium supplemented with the same level of rhuVEGF. Tube formation was monitored via light and Nomarski differential interference contrast (DIC) microscopy. Before DIC microscopy, cells were fixed with 4% formaldehyde (EM grade, Electron Microscopy Sciences, Ft. Washington, PA). DIC microscopy was carried out on a Leica (Allendale, NJ) DMXRE microscope at sector 2 of the Advanced Photon Source at Argonne National Laboratory.

Electron microscopy. Cells grown on Matrigel-coated Permanox-coated chamber slides were fixed, treated with 1% uranyl acetate, dehydrated, and infiltrated with Spurr's resin/propylene oxide at room temperature. Samples were polymerized, serially cut at 90 nm, stained with uranyl acetate and lead citrate, and viewed under a FEI Tecnai F30st-STEM microscope (FEI Co., Eindhoven, the Netherlands) operated at 300 keV.

RNA isolation, quality assessment, and microarray analysis. Quality-control assessments were done at three different stages: before hybridization, during target preparation, and posthybridization. Total RNA was purified from TRIzol suspensions at five time points “each” for proliferating (“VEGF series”) and tube-forming (“tube series”) HMVECs according to manufacturer's instructions. RNA preparations were further purified using RNEasy MinElute Clean Up columns (Qiagen, Valencia, CA). RNA quality was defined as minimum required A₂₆₀/A₂₈₀ and A₂₆₀/A₂₃₀ ratios of 1.8. In vitro transcribed cRNA (15 μg) was used from each of the samples. Hybridization to Affymetrix (Santa Clara, CA) GeneChip Human Genome U133 (HG-U133A and HG-U133B) arrays was carried out by the Functional Genomics Facility at the University of Chicago. Hybridization quality was evaluated by examining the report file for housekeeping gene hybridization, spike control hybridization, percentage of genes called present, 3′ to 5′ ratio, and background and scale factor ratio and by dChip analysis for regional image contamination and/or sample contamination.

Data analysis and filtration and gene annotation. Chip data were analyzed with the Affymetrix GeneChip analysis software (version 3.2) and GeneSpring software (version 5.0.2, Silicon Genetics, Santa Clara, CA). Raw expression scores were normalized using GeneSpring's default Per Chip Per Gene method. Mathematically, a morphogenesis-selective transcript (MRSL) has a normalized expression value at any one tube series time point at least 2-fold greater than all of the normalized expression values in the proliferation series. To create this list, a series of filters were applied to the time series data in terms of fold change or expression levels of normalized data, and the intersection or union of the series of lists was obtained using Venn diagrams. A maximum normalized value of 1.0 for the proliferation series (to eliminate high scores) and a minimum normalized value of 1.0 for the tube formation series (to eliminate low scores) were then applied to this interim sequence list.

Gene mapping and chip annotation was accomplished by using the alignment tools of the ENSEMBL, University of California at Santa Cruz, and National Center for Biotechnology Information (NCBI) Human Genome browsers¹

Quantitative real-time PCR analysis. Quantitative real-time PCR (QRT-PCR) was done using the Brilliant QPCR Core Reagent kit (Stratagene, La Jolla, CA) with rRNA as a control. Gene expression was quantified using the comparative C_T method with 18S rRNA as reference (see ref. 9 for detailed methods). PrimerExpress software (ABI, Foster City, CA) was used to design primers and probes.²

Immunofluorescence and visualization. HMVEC tube structures or proliferating HMVECs on 35-mm tissue culture plates coated with Matrigel and gelatin, respectively, were fixed, permeabilized, and incubated with 1:2,000 to 1:5,000 dilution of one of the following primary antibodies in 1% bovine serum albumin/PBS: rabbit anti–ephrin A1 (Zymed Laboratories, Inc., South San Francisco, CA), mouse anti-nestin (Chemicon International, Temecula, CA), and anti-ADAM19 (Novus Biologicals, Littleton, CO). The cells were then hybridized with a 1:200 dilution of secondary antibodies conjugated with either Alexa Fluor 488 or Alexa Fluor 555 (Molecular Probes, Eugene, OR) and with 4′,6-diamidino-2-phenylindole (DAPI) at a 1:100 dilution from the stock solution (30 g/mL; Polysciences, Inc., Warrington, PA). Cells were mounted using gelvatol (Monsanto, St. Louis, MO) in glycerol and imaged using a Zeiss (Thornwood, NJ) AxioCam microscope with AxioVision 3.1 digital imaging software. DAPI exposure times are 99 ms, whereas FITC channel exposure times ranged from 10.25 seconds for low-expressing proteins, such as nFATc2, to 3 to 4 seconds for higher-expressing proteins, such as ephrin A3.

Western blotting. Total cell protein was isolated by Matrisperse digestion according to the manufacturer's instructions (Becton Dickinson) followed by centrifugation, washing with PBS, and lysis in 1× Laemmli buffer [25 mmol/L Tris-HCl (pH 8.3), 250 mmol/L glycine, 0.1% SDS]. Proteins were separated by SDS-PAGE, transferred to Hybond-C membranes (Amersham Life Sciences, Piscataway, NJ), and blocked with 5% nonfat milk in PBS containing 0.05% Tween 20 with agitation. Tetranectin was visualized by incubation with primary unlabeled rabbit anti-tetranectin antibody (DAKO Corp., Carpinteria, CA) in blocking buffer followed by incubation with horseradish peroxidase–labeled anti-rabbit IgG cross-absorbed conjugates (Jackson Immunoresearch, West Grove, PA) and chemiluminescent detection using SuperSignal West Femto visualization reagents (Pierce Chemical, Rockford, IL).

Monte Carlo simulation of MRSL gene chromosomal location. Correlation coefficients and Monte Carlo simulations were carried out using the number of predicted genes per chromosome in the NCBI Homo sapiens Genome Build 35 version 1. The probability that two MRSL genes would be placed immediately adjacent to one another in a chromosome of n genes containing m MRSL genes was calculated by a Monte Carlo simulation using 10,000 trials, each trial consisting of the random placement of genes within a chromosome and tabulation of the number of times MRSL genes are placed as nearest neighbors.

Morphologically, Matrigel-driven tubulogenesis occurs in two discrete phases: an early attachment, migration, and cord-like formation phase lasting up to 2 hours postplating involving significant process formation (Fig. 1A and B) followed by a reorganization and consolidation phase extending from 2 to 8 hours and later (Fig. 1C; ref. 10). Electron microscopy confirmed progressive vacuolation, increasing in both diameter and number per cell and leading to the formation of single cells containing a lumen of ∼2 μm by 8 hours with the appearance of multicellular lumen of 3 to 4 μm in diameter by 15 hours (Fig. 1D). Studies carried out at lower cell density to slow tube formation show that the earliest morphologic change on attachment to matrix was the formation of long, thin actin-based processes between neighboring cells (Supplementary Fig. S1A; refs. 11, 12).

The microarray analysis was designed to observe early activities (5), including transient transcription factor up-regulation (3), with the first RNA samples isolated 30 minutes postplating. As internal controls for both the validity of the cell culture system as a model for tumor-induced angiogenesis and the subsequent microarray analyses, we analyzed the expression profiles of several transcripts known to be induced by VEGF within endothelial cells. High raw and normalized scores were noted in a time-dependent fashion for HEY1, HEY2, COX2, EGR3, and CXCR4 (Supplementary Fig. S2; refs. 4, 13–15). Serial induction of EGR1 mRNA was followed 1 hour later by up-regulation of tissue factor mRNA, an EGR1-responsive gene, in both time series, showing the utility of the chosen time points for distinguishing distinct expression phases within the system (Supplementary Fig. S3). Analysis of 203 known angiogenesis proteins in the array data set (defined by the MEDIC Breast Cancer Project as being relevant to the angiogenic process) showed that 109 (54%) were detectable in both proliferating and tube-forming cells, 50 (25%) were not detectable in either time series, 10 (5%) were detected only in the proliferating cell data set, and 34 (17%) were tube formation specific (Table 1, top; Supplementary Table S1). These data indicate that only about one-sixth of these sequences defined as angiogenic are associated solely with the morphogenesis aspect of the process.

The majority of microarray studies of endothelial cells have similarly identified transcripts associated with proliferation, not tubulogenesis, as shown by an exhaustive comparative analysis between this data set and previously published studies of endothelial cell gene expression patterns. Previous related microarray and/or cDNA studies showed significant overlap with transcripts up-regulated during the proliferation time series but included almost no MRSLs (2–6, 16–18), emphasizing the value of using temporal expression patterns during proliferation as a background control and examining the process at its earliest initiation in the identification of sequences associated strictly with morphogenesis. Analysis of the MRSL list in the context of endothelial cell specialization (19) and tumor angiogenesis (20, 21) indicated no obvious overlap of genes. This overall lack of concordance of our MRSL list with proteins previously identified as angiogenic/endothelial reflects the emphasis in previous studies placed on mature and/or proliferating endothelial cells, whereas the MRSL proteins identified here are more likely to be involved in regulating and coordinating nascent vasculature.

The multistep informatic filtering algorithm employed as described in Materials and Methods was designed to remove transcripts related primarily to proliferation, nonspecific attachment, or cytokine response in the absence of morphogenesis and to select for putative tubulogenesis process-associated transcripts, essentially subtracting out a background composed of not only noninduced transcripts but also transcripts up-regulated during proliferation. In the heat map set in Fig. 2, it can be seen that the MRSL set of 217 sequences (Table 1; Supplementary Table S2) reflects very low levels of expression during proliferation yet very high normalized levels during tubulogenesis. The 217 MRSLs should not be construed as the complete module of genes operating during endothelial cell morphogenesis, but the set of genes whose change in expression most completely distinguishes endothelial cell morphogenesis from the more functionally simple endothelial cell proliferation.

Temporal expression profiles were validated using QRT-PCR on 52 target sequences: 44 MRSLs, 2 down-regulated sequences, and 6 sequences expressed during both time series. RNA samples used as templates for this analysis were isolated from multiple independent experiments by multiple investigators to generate a robust set of validated gene expression patterns. A representative data set is shown in Fig. 3. Time points to 4 hours postinduction confirmed the overall expression pattern seen in the microarrays in 42 of 44 (95%) MRSL sequences. In multiple instances, QRT-PCR-derived induction values were significantly higher than predicted from corresponding microarray values (see, for instance, ephrin A1 and REPAC/MR-GEF in Fig. 3) as would be anticipated from a method with a wider dynamic range of detection.

Antibodies to MRSL proteins were used as probes to ascertain the pattern of expression of proteins corresponding to microarray-detected transcripts. All 16 MRSL proteins assayed at 7 to 8 hours postinduction were detectable by immunofluorescent microscopy (Fig. 4; Supplementary Fig. S4). Protein presence was highly correlated to transcript presence, as evidenced by undetectable fluorescence in control proliferating HMVECs for 13 of the antibodies, and the low level of fluorescence detected using the anti-nestin, anti-SLIT2, and anti-GSTM1 antibodies, a pattern matched by their corresponding transcript levels. The temporal relationship between transcript up-regulation and protein expression was examined for the C-type lectin tetranectin. An increase above a normalized expression value of 1.0 for the tetranectin transcript using the microarray data is first detected at 1 hour postinduction with Matrigel during tube formation, with transcript levels peaking at 2 hours postinduction (Fig. 5, top). Protein levels lag behind transcript levels as measured by Western blot, attaining maximal levels at 4 hours postinduction (Fig. 5, bottom).

Chromosomal localization of the 217 morphogenesis-specific transcripts indicated that MRSL sequences map to all human chromosomes, except Y, with 23 of the 217 genes lying within 2 or 3 genes of at least one other MRSL gene in 200-kb to 1-Mb length regions at 10 different loci. Seven of the 10 cluster sites consist of nonrelated sequences, whereas the other 3 clusters are at least partly composed of tandem clusters of related genes (Table 1, † genes). Distribution of the MRSL genes among the chromosomes appears random, as the correlation coefficient between the number of MRSL genes on a chromosome and the total number of predicted genes on that chromosome approached 1 (r² = 0.78, 95% confidence interval; Supplementary Fig. S5). However, Monte Carlo simulations of MRSL gene distribution within individual chromosomes indicate that gene clustering by chance is a statistically low probability event (Supplementary Table S3). As an example, the probability of obtaining one of the two MRSL clusters on chromosome 19 by random chance is 0.13 × 0.13 × 0.13 = 0.0022 or 0.22%. Analysis of the temporal expression profiles of genes neighboring the other 194 MRSL genes within ±500 kb with a less stringent mathematical definition of morphogenesis-selective increased the number of identified MRSL gene clusters to a total of 27, including 41 MRSL genes of the original 217.³

It has been postulated that large-scale cellular processes, such as differentiation, may require large-scale nuclear reorganization to position the active genes at the surface of chromosome territories and facilitate access to the transcription and splicing machinery (22, 23). This architectural reorganization would be expedited by the clustering along the chromosome of coregulated sequences into “expression neighborhoods” (24). Several MRSL transcripts were found to involve higher-order regulation of gene expression, including the architectural transcription factors polycomb, PRDM15/ZNF298, NSE1, BAZ2A, BCL6, and HMGIY, which modulate chromatin structure (25). The increase seen in levels of proteins associated with chromatin remodeling and the chromosomal clustering of a subset of MRSL genes suggests that a requirement for dynamic colocalization of active genes may be an underlying factor in the regulation of some aspects of endothelial cell morphogenesis.

Functional annotation of the 217 MRSL transcripts shows that ∼70% of morphogenesis-selective sequences are involved in one of eight functions: regulating gene expression, either as a homeobox or transcription factor protein (H/TF = 17%) followed in order by intracellular trafficking (ICT/S = 10%), motility (M/CT = 9%), cell adhesion (CA = 9%), cell fate determination (CFD/M = 9%), angiogenic factors/guidance cues (AF/GC = 6%), signal transduction (ST = 6%), and cytoskeletal reorganization (CSO = 5%; Table 1; Supplementary Table S2; Supplementary Fig. S6A). Combination of the M/CT and CSO categories to create a larger “cell shape” group resulted in a larger category second in size (14%) only to gene expression regulators. With the exclusion of transcription factors, this translates into 50% of the MRSL products relating to polarity.

Table 1 lists a representative subset of MRSLs in which “early transcripts” refer to initial up-regulation at 30 minutes and 1 hour and “late transcripts” up-regulate at 2 and 4 hours postinduction. The pattern of induction of MRSL transcripts within each function category was analyzed, with representative trends shown in the bar graphs in Supplementary Fig. S6B. A large number of functions reflect a phase shift at 2 hours postinduction, with either maximal up-regulation or a second peak of up-regulation occurring at 2 hours postinduction. Further breakdown of the cell fate determining MRSLs showed that 17 of 19 were associated with negative growth control, with cyclin B2 and Est1A being the exceptions and up-regulating early at 30 minutes. This is consistent with the well-known antithetical relationship between proliferation and differentiation in multicellular organisms.

Early establishment of cellular polarity is characteristic of tube formation, requiring a major reorientation of the cellular actin cytoskeleton and the formation of abundant actin-based filopodia (Fig. 1A; Supplementary Fig. S1A; refs. 10–12). Maintenance of the molecular identity of each of those regions requires intricate sorting and trafficking mechanisms. Consistent with this requirement, 10% of the MRSLs are associated with various aspects of intracellular transport. Components of both the microtubule-based motor complexes (PIM1 and DCTN2 promote dynein/dynactin complex formation; GAKIN/KIF13B is a kinesin-like motor protein) and the actin cytoskeletal motor complexes (MLPH/SLac2a is a cargo adaptor for the actin-based motor protein myosin Va; PDLIM1/ELFIN can localize cargo, such as the MRSL protein STK35 to actin stress fibers) are selectively up-regulated during the tube formation process. It has been shown that microtubule extension occurs preferentially along actin-based filopodia, which act as microtubule polymerization guides in neuronal growth cones, another system that involves the formation of long-range, guided cellular extensions of specialized function (26). It is possible that the motor-related MRSL proteins we have identified may be required to maintain the length of the extended filopodia seen in Fig. 1A during initial tube formation.

Up-regulated motility-associated MRSL transcripts appear at an almost constant rate over the first 4 hours of tube formation (Supplementary Fig. S6B), implying an almost constant regulation of motility during the process. These data are in accord with careful studies carried out by Meyer et al. (27) and Connolly et al. (10), indicating that, subsequent to primitive cord-like structure formation, further cell migration occurs by nonnetworked cells through previously made channels in the matrix and by crawling over preexisting tube structures, resulting in a more mature, higher cell density tubule network. Interestingly, the Sprouty-related protein Spred 1 increases over 2-fold between 2 and 4 hours post-Matrigel stimulation, remaining relatively flat during proliferation, implicating this protein in the inhibition of endothelial cell migration postnetwork formation (Supplementary Fig. S7; ref. 28). The majority of M/CT MRSL genes are signal transduction molecules, not architectural proteins, as would be expected from a transcriptome-based analysis. Between 4% and 5% of the MRSL sequences are G protein regulators, including six guanine nucleotide exchange factors and three GTPase-activating proteins (Table 1). These genes reveal molecular footprints for the small GTPases Rho, Rac, Rap, Ras, and Cdc42 as well as the G protein Gα11. Although the mechanisms by which this family regulates actin microfilament dynamics are complex and not fully understood, our identification of numerous members of these families as MRSLs suggests a prominent role for their signaling during endothelial cell tubulogenesis.

MRSL transcripts associated with guidance (AF/GC) and adhesive (CA) properties each exhibit a single peak of maximal up-regulation at 1 and 2 hours, respectively (Supplementary Fig. S6B). Davis et al. (29) observed that endothelial cells release autocrine factors influencing the invasive properties of nearby endothelial cells, with the use of axon guidance transcripts, such as MRSLs SLIT2, the Semaphorins, and the ephrins as mediators of tissue morphogenesis having been noted by many others (30). The continued up-regulation of angiogenic and guidance factors throughout most of this process suggests that endothelial cells require constant molecular feedback to construct proper tubule structures. The large number of adhesion-modulating molecules produced during tubulogenesis may reflect the fact that changing patterns of surface chemistry are central to the rearrangement of cellular assemblies (31), with lumen formation being a simple consequence of differential adhesion in cells expressing adhesive properties in a polarized fashion (32). Continual trafficking of cadherins (33, 34) and integrins (35) in the regulation of adherens junction adhesiveness and cell motility, respectively, has been documented. A need for continual modulation of cell surface properties, such as adhesiveness, may be a factor in the large number of MRSL transcripts identified as being associated with trafficking.

The small number of proteases noted in the MRSL list (CD10/CALLA, ADAM19/MADDAM, and cathepsin H) may be involved in the exposure of matricryptic sites within the matrix and modulating adhesive and migratory properties. This could be a factor in the ability of “succeeding” endothelial cells to move through the matrix pathway established by “pioneer” endothelial cells via the creation of a molecular trail of attractants. Surprisingly, multiple matrix metalloproteinases (MMP) are up-regulated by proliferating endothelial cells unstimulated by matrix material (such as high levels of MMP2, MMP14, MMP19, and MMP24; data not shown) as well as by tube-forming cells. An early and continuing up-regulation of transcripts associated with lipid metabolism (L-MET), such as ABCG1, ABCA1, and SEC14L2, can also be seen in the MRSL sequences (Supplementary Fig. S6B). Work by Gerritsen et al. (18) showed a similar up-regulation of lipid metabolism genes at this stage and pointed out a requirement for lipid biosynthesis in vacuole/lumen formation.

These data indicate that, at the transcript level, what most distinguishes in vitro endothelial cell tubulogenesis from proliferation is a set of highly choreographed actions of genes and gene products that modulate subcellular localization of cellular and surface components, resulting in higher-order tissue architecture. Expression of these genes is driven by cell/extracellular matrix (ECM) contacts. This work elaborates, at the molecular level, the origin of the biphasic nature of in vitro endothelial cell morphogenesis described in the elegant morphologic studies of Connolly et al. (10). Tube formation in artificial matrices is characterized by a rapid onset, long-range, guided migratory phase resulting in a primitive cellular network followed by a longer consolidation/maturation phase involving short-range migration and lumen formation. The novel morphogenetic genes identified herein can be divided into two broad categories: the one-sixth of the genes that modulate gene expression (H/TF) and the majority of the remaining genes that generate or maintain polarity (ICT/S, CSO, AF/GC, CA, ECM, and P/MMP).

This analysis further shows that few of the molecular targets of angiopreventive drugs in development are specific to endothelial cell morphogenesis, as most were found to be up-regulated at high levels in proliferating endothelial cells not undergoing tube formation. The 217 human proteins identified in this simple in vitro system provide initial information about the kinds of proteins that are involved with such processes as shape change, polarization, and guided migration. The identification of protein targets expressed early within the in vitro process and associated with the morphogenesis component of tube formation may, subsequent to follow-up in vivo studies, eventually provide better targets for pharmacologic intervention, which can interfere in multiple aspects of aberrant capillary growth associated with numerous human disorders.

Although previous efforts have been made to molecularly dissect the individual components of multicomponent processes (e.g., ref. 36), none have taken this type of high-throughput systems biology approach coupling microenvironmental manipulation of cells with a rigorous analysis of subtracted temporal gene expression “patterns.” Our subtractive analysis of transcriptome transformation over time has identified moderately expressed genes up-regulated in the tube-forming cells that are not observed during undirected proliferation. This novel systems biology approach is a generally applicable tool for the molecular dissection of complex biological processes and has the potential to contribute information not readily available by other experimental means.

Grant support: U.S. Department of Energy contract W-31-100-Eng-38.

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.

We thank the staff at sector 2, XOR-CAT, at the Advanced Photon Source, Argonne National Laboratory, for the use of the DIC microscope; the assistance and advice of Dr. X. Li at the Functional Genomics Facility and Y. Chen at the Electron Microscopy Center at the University of Chicago and E.C. Uberbacher at ORNL for helpful discussions. All primary microarray data has been deposited in the Gene Expression Omnibus database at the NCBI in MIAME guideline format upon publication under series record number GSE3891.