1 Division of Pulmonary, Allergy and Critical Care Medicine
2 Lung Translational Genomics Center, Department of Medicine, University of Pittsburgh, Pennsylvania 15213
3 Institute for Human and Machine Cognition, University of West Florida, Pensacola, Florida 32514
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
endothelium; hypoxia; serial analysis of gene expression
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Not surprisingly, adaptation to hypoxia is crucial to the survival of most organisms. Vasoactive substances that modulate vascular tone and blood flow mediate homeostatic response to short-term chronic hypoxia and these adaptations are readily reversed by a return to normoxic conditions (17, 58). In contrast, adaptation to chronic hypoxia involves profound and irreversible changes including remodeling of the vasculature and surrounding tissues via smooth muscle proliferation and fibrosis (17, 58). It is known that the responses, and therefore consequences, of exposure to hypoxia vary in different tissues and that these variations are significant from a clinical perspective (52).
The recent acceleration in the acquisition of genomic sequence information has driven rapid developments in technologies for the massively parallel analysis of gene expression at the level of transcription. One such technique, serial analysis of gene expression (SAGE) (50, 62), allows unbiased genome-wide analysis of gene expression via the generation of short cDNA-derived gene tags from a starting population of mRNA molecules. These are concatemerized, cloned into plasmid vectors and sequenced to generate quantitative catalogs of expressed genes. Unlike microarray approaches, SAGE is unbiased in that it does not require a priori knowledge of genes of interest and is therefore not constrained by sequences on a chip. Furthermore, SAGE data contains expression information for every tag (gene) relative to every other tag in a given library and so requires minimal normalization and, since it generates immortal data, can be readily shared between laboratories.
Despite the intense interest in the cellular response to hypoxia, particularly in the context of vascular biology, there have been no systematic attempts to document the transcriptional response to short-term chronic hypoxia in primary vascular endothelial cells. We have, therefore, utilized SAGE to determine the temporal response to hypoxia in primary cultures of human aortic endothelial cells (HAECs). This preliminary step towards the functional characterization of the endothelial cell genome in response to hypoxic stress serves as both a rich source of information and a general resource for vascular biologists.
![]() |
MATERIALS AND METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
SAGE.
Twenty micrograms of total RNA was used to construct each SAGE library, as recommended in the MicroSAGE Detailed Protocol (version 1.0e) (56) with some minor modifications. In brief, double-stranded cDNA was synthesized from mRNA bound to oligo(dT) magnetic beads (Dynal Biotech, Lake Success, NY), using SuperScript II reverse transcriptase (Invitrogen). The cDNAs were cleaved with NlaIII (anchoring enzyme), and the most 3' terminal cDNA fragments were captured with magnetic beads and divided into two pools. Each pool was ligated to 5' biotinylated linker A or B (62), containing recognition site for the tagging enzyme BsmFI. After ligation, the beads were washed and the SAGE tags released from both pools by digestion with BsmFI. Tags were blunted at their 3' ends and combined to form the 104-bp ditags-linker products, which then were amplified by PCR. The amplified ditags-linkers were redigested with NlaIII to remove the linkers, and the ditags (26 bp) were isolated by gel electrophoresis and purified through Spin X tubes (VWR International, West Chester, PA) and concatemerized by self-ligation. Concatemers with sizes between 500 and 2,500 bp were obtained by gel purification and cloned into the SphI site of vector pZero (Invitrogen) and transformed into Escherichia coli strain DH10B (Invitrogen) by electroporation. For each library, about 1,200 colonies were random picked, and plasmids with concatemer inserts were cycle sequenced with Big Dye Terminator chemistry (Big Dye version 1; Applied Biosystems, Foster City, CA) and analyzed on DNA sequencer (model 3700, Applied Biosystems).
SAGE data analysis.
The sequence file generated by the automated sequencer was analyzed using the SAGE 2000 software (version 4.12; kindly provided by K. W. Kinzler and colleagues). After elimination of linker sequences and the duplicate ditags, the software was used to extract tags from the sequence file and create a report of the sequence and the occurrence of each of the transcript tags. According to Lal et al. (32), the transcript identity of each SAGE tag was obtained by matching the unitag list against the human tag-to-gene "reliable" map from ftp://ftp.ncbi.nlm.nih.gov/pub/sage/ using Microsoft Access. Each specific transcript abundance was then determined by its unique tag count.
Real-time quantitative RT-PCR.
Total RNAs were purified by the RNeasy Mini Kit (Qiagen, Valencia, CA), eliminated of residual genomic DNA by the DNA-free kit (Ambion, Austin, TX) according to the manufacturers protocol, and quantified by spectrophotometry (Beckman model DU 640). The optimal reverse transcription (RT) was carried out in 100-µl volumes as described (9) and two RNA inputs (100 and 400 ng). No-reverse transcriptase controls were carried out with 400 ng of RNA. Quantitative PCR was performed on this cDNA on a sequence detection instrument (model ABI 7700, Applied Biosystems) using TaqMan MGB probes. Quantitative RT-PCR was carried out for four genes that were identified to be differentially hypoxic-induced genes in HAECs by SAGE analyses. Their PCR primers and probe were ordered from Applied Biosystems (MMP2:Hs00234422_m1, SERPINE1:Hs00167155_m1, CAV:Hs00184697_m1, MET:Hs00228845_m1, and CTGF:Hs00170014_ml). PCR amplification of cDNA derived from HAECs (n = 2) was performed in duplicate in 50-µl volumes as described (9) with the optimal primer and probe concentrations used for each gene (300 nM for primer, 100 nM for probe). Gene expressions were measured relative to the endogenous reference gene, human ß-glucuronidase (ß-GUS), using the comparative CT method described previously (9). A calibrator RNA sample, human lung RNA, was amplified in parallel to allow comparison of samples run at different times.
Distribution of the counting of a tag.
The analysis of SAGE data assumes that the distribution of tag counts follows a binomial distribution. Given a SAGE library of size n, the count of a type of tag t has a binomial distribution with parameters (n,p), where p is the relative frequency of tag t, or ideally, the gene represented by tag t in the original tissue/cell population (8).
Test for differentially expressed genes.
Suppose we have s SAGE libraries. Let ni be the size of the ith library, and Xi the counting of tag t in the ith library. Pearsons 2 statistic is defined as
![]() |
![]() |
Asymptotically, under the null hypothesis that t is not differentially expressed, T has a s21 distribution. Simulation studies show that for SAGE data, the asymptotic distribution is a good approximation to the exact distribution of T (under the null hypothesis). In this paper, we use the following level 5% test: a tag t is differentially expressed if the T statistic for this tag is greater than the 95% quantile of the
s21 distribution.
Control of the false discovery rate.
Because we are testing the expression levels of thousands of tags simultaneously, we need to control the false discovery rate (FDR), i.e., among the tags claimed to be differentially expressed, the (average) percentage of the tags that actually are not differentially expressed. We use "linear step up multiple comparison procedure" of Benjamini and Hochberg (BH procedure) (6). The BH procedure first sorts the p values of the test statistics p(1) ...
p(k) in ascending order, where k is the number of tests. To keep the average FDR below a given level
, we search for the largest i such that p(i)
i/k and reject all the null hypotheses whose p values are smaller than p(i). Using this procedure, all the tags whose T statistics are greater than the 1 p(i) quantile of the
s21 distribution will be considered differentially expressed. We apply the BH procedure only to the tags that are at least moderately expressed in one library, because we know in advance that a tag barely expressed in both the libraries is not likely to be differentially expressed. Genes that would not be considered differentially expressed when FDR is controlled at 5%, but would be considered differentially expressed without FDR control, were included in cases where they match genes of potential biological significance.
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Identification of differentially expressed tags.
We identified 321 tags whose expressions varied significantly between the three experimental conditions. Of these, 193 tags were found to match single UniGene entries. Within these 193 tags, 14 and 18 tags matched ESTs (or other uncharacterized cDNA clones) and hypothetical genes, respectively. Thirty-six tags had no match to any UniGene entry. These 193 single match and 36 no-match tags are shown in Supplemental Table S1 (the Supplemental Material for this article is available at the Physiological Genomics web site).1
The remaining 92 tags werefound to match multiple UniGene entries and were excluded from further analysis. The entire list of 321 differentially expressed tags is shown in Supplemental Table S2.
Hierarchical clustering to identify genes whose expression patterns are similarly affected by hypoxia.
We next performed hierarchical clustering analysis to identify clusters of genes whose expressions vary in similar fashion following exposure to hypoxia. These analyses focused only on the 192 tags found to match single UniGene entries plus the 36 tags without any UniGene matches. Although tags matching multiple UniGene entries may be of interest, these were not included in these analyses so as to minimize false-positive results.
Gene and experiment trees were constructed (not shown) using GeneSpring software (Silicon Genetics) for the 228 differentially expressed single UniGene match tags and no-match tags. We identified six major clusters (clusters 16) of genes, and these can be broadly defined as follows. Cluster 1 includes genes whose expressions are dramatically increased within 8 h and then dramatically reduced between 8 and 24 h. Cluster 2 includes gene whose expressions are moderately decreased, unchanging, or slightly elevated between 0 and 8 h and then reduced between 8 and 24 h. Cluster 3 includes genes whose expressions are rapidly decreased between 0 and 8 h and then relatively unchanged between 8 and 24 h. Cluster 4 contains genes that are dramatically reduced between 0 and 8 h and then dramatically increased between 8 and 24 h. Cluster 5 contains genes that are increased between 0 and 8 h and then unchanged, slightly reduced, or slightly increased between 8 and 24 h. Finally, cluster 6 contains genes whose expressions are relatively unchanged between 0 and 8 h and then increased between 8 and 24 h. These data are summarized in Supplemental Table S1.
Functional characteristics of hypoxia-responsive genes.
Given the challenges associated with classifying many genes by function in the context of a single experiment, we utilized a number of online tools that are designed to assist investigators in this task. These include Onto-Express (14) and FatiGO (3), which are united by their use of the Gene Ontology (GO) Database provided by the GO Consortium (http://www.geneontology.org; Ref. 20). Use of the GO terms to functionally classify and analyze the results of massively parallel gene expression experiments has the advantage of providing a standard output format that can readily be compared with gene ontology information derived from other data sets. Table 1 shows a broad functional classification by biological process of the genes listed in Supplemental Table S1 whose expressions were altered by hypoxia. Specific terms for these processes were extracted from multiple levels within the GO hierarchy so that only major gene categories are listed.
|
|
The expressions of genes involved in ATP-derived energy production are coordinately increased by short-term chronic hypoxia.
We also observed an increase in expression of genes involved in metabolic energy production such as ATP synthase H+ transporting mitochondrial F1 complex O subunit (ATP5O), ubiquinol-cytochrome c reductase (UQCR), and ubiquinol-cytochrome c reductase hinge protein (UQCRH), which are components of the mitochondrial electron transport chain. The expressions of other genes encoding enzymes involved in glycolysis and ATP production are also altered by hypoxia. These include adenylate kinase 1 and 2 (AK1 and AK2), glucose phosphate isomerase (GPI), aldolase A (ALDOA), glyceraldehyde phosphate dehydrogenase (GAPDH), hexokinase 2 (HK2), and enolase (ENO1). Lactate dehydrogenase A (LDHA), which catalyzes the conversion of L-lactate and NAD to pyruvate and NADH in the final step of anaerobic glycolysis, is also increased by hypoxia, and this increase is evident at both 8 and 24 h. Furthermore, hypoxic exposure caused elevated expression of the fatty-acid-CoA ligase, long-chain 4 (FACL4) gene (see cluster 1), which is involved in fatty acid degradation.
Exposure to hypoxia results in the modulation of cytoskeletal gene expression.
Given that hypoxia is known to elicit changes in cytoskeletal organization and permeability in endothelial cells, it is significant that we identified an alteration in expression of genes whose products are thought to regulate cytoskeletal remodeling. Genes whose expressions are elevated by hypoxia include: growth factor receptor bound protein 2 (GRB2) (42), pleckstrin 2 (PLEK2) (26), spectrin-ß (SPTB), which is a component of adherens junctions (69), myristoylated alanine-rich protein kinase C substrate (MARCKS), which encodes an actin cross-linking protein whose activity is inhibited by calmodulin (24), profilin 1 (PFN1), which regulates actin polymerization, and ras homolog gene family member A (ARHA), which is involved in regulating remodeling of the actin cytoskeleton during cell morphogenesis and motility (39). Those reduced by hypoxia include p21-activated kinase 3 (PAK3), which is involved in remodeling of the actin cytoskeleton (40).
Hypoxia causes increased expression of genes encoding extracellular matrix factors.
Exposure to hypoxia also resulted in the elevation of tags encoding genes involved in extracellular matrix synthesis and remodeling. These include collagen, type IV, 1 (COL4A1), procollagen-lysine, 2-oxoglutarate 5-dioxygenase (PLOD), procollagen-lysine, 2-oxoglutarate 5-dioxygenase 2 (PLOD2), lysyl oxidase-like 2 (LOXL2), EGF-containing fibulin-like extracellular matrix protein 1 (EFEMP1), and matrix metalloproteinase 2 (MMP2). Matrix metalloproteinase 1 (MMP1), which is expressed at very high levels in our system, is downmodulated by hypoxia at both 8 and 24 h relative to normoxic control. Tissue inhibitor of metalloproteinase 2 (TIMP2) is also elevated in our system at both 8 and 24 h hypoxia relative to the normoxic control. Finally, we also found that transforming growth factor-ß-induced, 68-kDa (TGFBI) was reduced approximately fourfold by hypoxia at 8 h and returned to baseline by 24 h, although this was not statistically significant according to our stringent cutoff criteria.
Hypoxia elicits coordinated changes in the expressions of proapoptotic genes.
Also increased by hypoxia is the HIF1-inducible gene RTP801, which although poorly characterized, is thought to be involved in apoptosis (53). Other proapoptotic factors were also increased by hypoxia including the promyelocytic leukemia gene (PML), the interferon--inducible protein 16 (IFI16), apoptosis-inducing factor (AIF)-homologous mitochondrion-associated inducer of death (AMID), BCL2/adenovirus E1B 19-kDa interacting protein 3-like (BNIP3L), programmed cell death 2 (PDCD2), and death-associated protein kinase 3 (DAPK3). However, we also found that the gene encoding CARD only protein (COP), which has been described as a negative regulator of pro-caspase-1 activation (33), was also elevated and also that the proapoptotic BCL2-associated X protein (BAX) was reduced.
Gene expression changes in other functional groups.
Hypoxia also resulted in an increase in the expression of vascular endothelial growth factor C (VEGFC), although this was not statistically significant. Additionally, we observed a significant increase in the expression of plasminogen activator inhibitor type 1 (SERPINE1). This increase was confirmed by real-time RT-PCR (Fig. 1). We also found that a number of genes encoding proteins involved in mediating a cytoprotective response to oxidant stress were reduced by hypoxia, including metallothionein 2A (MT2A), biliverdin reductase A (BLVRA), and GABAB-related G-protein coupled receptor (GABABL).
|
It can be seen from Fig. 1, A and B, that changes in gene expression level identified by SAGE were reproduced accurately by RT-PCR. The hypoxia-responsive differential expression identified by SAGE was quantitatively corroborated by RT-PCR for these five genes.
Identification of previously described hypoxia-responsive genes.
To analyze our data in the context of known hypoxia-responsive gene expression, we searched our data for genes that are modulated through hypoxia inducible factor 1 (HIF1) mechanisms. Table 3 shows that there are both consistencies and discrepancies between our own and these previously published data. For example, known hypoxia-responsive such as enolase 1, glyceraldehyde phosphate dehydrogenase, heme oxygenase 1, hexokinase 2, and lactate dehydrogenase A are all confirmed as being differentially expressed in our data, whereas phosphoglycerate kinase 1 and pyruvate kinase M are not. However, some known hypoxia-responsive genes are not detected by SAGE in our system (e.g., adenylate kinase 1 and adrenomedullin) and some are expressed at very low levels, making comparison difficult (erythropoietin and hexokinase 1). The discrepancies may reflect genuine differences in tissue-specific gene expression patterns and/or the fact that much of the previously published experiments were performed in cell types other than endothelial cells. For example, previous experiments documenting the hypoxia-responsive gene expression of phosphoglycerate kinase 1 and pyruvate kinase M were performed in HepG2 and embryonic stem (ES) cells (28).
|
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Perhaps the most striking feature of our data is the extensive and coordinated increase in expression by 8 h of a large number of genes encoding HSP proteins followed by dramatic downmodulation by 24 h. Despite the fact that HSPs are widely reported to be induced in response to cellular stress in mammals (for review, see Creagh et al. 2000, Ref. 12) this is the first study describing a coordinated temporal response to hypoxic stress in primary cultures of endothelial cells. Furthermore, it has previously been reported that hypoxia results in downmodulation of HSP genes in transformed hepatocytes (55) and microvascular endothelial cells (44). However, these studies all involved endpoint measurements of HSP gene expression levels after 24, 24, or 20 h (respectively) of hypoxia. Given that our experiments document a dramatic increase in HSP gene expression after 8 h of hypoxia followed by a reduction in expression at 24 h, it is not surprising that these observations conflict with our own.
We also observed a coordinated reduction in the expression of a number of genes that drive cell cycle progression or are biomarkers of this process, suggesting an antiproliferative response to hypoxia. These observations are consistent with a previous comprehensive analysis of endothelial cell proliferation rates under hypoxic conditions in which it was found that endothelial cell division slows but does not arrest and progression through the G1-to-S transition point and/or progression from S to G2/M is altered with an increased percent of endothelial cell in S phase (59). Future experiments will directly determine the time-dependent effects of hypoxia on endothelial cell proliferation rates in vitro. Our data may provide mechanistic information about the molecular events that underlie this process. For example, we found that the positive regulator of cell cycle progression JUND was reduced by 8 h of hypoxia and undetectable at 24 h. The modulation of JUND signaling pathways in the context of hypoxic stress has not been extensively described before, although it has been shown that JUND is induced by hypoxic stress in cardiac myocytes (64). The reduction in expression by 8 h hypoxia of CCND1, which forms a complex with and functions as a regulatory subunit of CDK4 or CDK6 (whose activity is required for cell cycle G1/S transition) also suggests reduced cell proliferation. Similarly, MCM2, described as a "licensing factor" that permits DNA replication during mitosis (38), was expressed under hypoxic conditions in our experiments but undetectable at 8 h and 24 h of hypoxia. Also responsive to hypoxia but poorly characterized is ERH. ERH is implicated in cell cycle regulation through its genetic interaction with the rudimentary gene, which encodes a protein possessing the first three enzymatic activities of the pyrimidine biosynthesis pathway (19), and it has recently been suggested that ERH is a transcriptional repressor (47). Also consistent with an antiproliferative response under hypoxic stress is the increased expression of RBBP1. The protein encoded by this gene is a transcriptional repressor that has recently been shown to inhibit E2F-dependent gene expression and suppress cell growth (31). It is possible that the increase in expression of RBBP1 relates to an inhibition of E2F-mediated cell cycle progression. Additionally, we observed a significant reduction in CALM1 expression after 8 h hypoxia and were unable to detect its expression after 24 h. CALM1 has been shown to activate p42/p44 MAPK under hypoxic conditions (10) and in keeping with the apparent reduction of expression of genes involved in cell cycle progression, p42/p44 MAPK inhibition has been shown to result in cell cycle exit in vascular endothelial cells (63). Furthermore, the p42/44 MAP kinase cascade is negatively regulated by CAV1 (46), which we found to be induced almost fourfold by 24-h exposure to hypoxia.
The increased expressions of genes encoding enzymes involved in glycolysis and ATP production by hypoxia are consistent with previously published data (1, 51). Additionally, the upregulation of FACL4 may indicate utilization of fatty acids for energy production under stress as has been suggested by Dagher et al. 2001 (13). As far as we are aware, this is the first report in which the coordinated modulation of multiple genes involved in these processes has been observed in human endothelial cells. We similarly observed modulation in the expressions of multiple genes involved in cytoskeletal remodeling including downmodulation of SPTB, which is a component of adherens junctions (69), and MARCKS, which encodes an actin cross-linking protein whose activity is inhibited by calmodulin (24). Interestingly, it has recently been shown that the product of the MARCKS gene is phosphorylated during actin cytoskeletal remodeling under oxidative stress in pulmonary endothelial cells (68). The reduction following 8 h hypoxia of ARHA (RhoA) followed by an increase to basal levels by 24 h suggests the possible involvement of AHRA in mediating the remodeling of the actin cytoskeleton in HAECs. Interestingly, we also observed a significant increase in RAC1 gene expression by 8 h hypoxia, which may be significant given the well-documented involvement of both AHRA and RAC1 in regulating endothelial cell permeability under cellular stress (39, 65). Finally, the upregulation by 24 h hypoxia of PAK3, which is involved in remodeling of the actin cytoskeleton (40), may have significance in light of the AHRA/RAC1 observations, since PAK proteins are critical effectors that link Rho GTPases to cytoskeleton reorganization and nuclear signaling (67). Whether AHRA, RAC1, and PAK3 modulation at the level of transcription has any specific functional significance, however, is unclear.
Consistent with previous observations (34), we observed elevation of tags encoding genes involved in extracellular matrix synthesis and remodeling. This synthetic gene expression profile was represented by increases in the expression of COL4A1, PLOD, and PLOD2. COL4A1 encodes the major type IV -collagen chain of basement membranes, whereas PLOD and PLOD2 both catalyze the hydroxylation of lysyl residues in collagens (61). In keeping with the above, LOXL2 (also increased by hypoxia) initiates the cross-linking of collagens and elastin (43). TGFBI (also known as BIGH3), which we found to be transiently downmodulated by hypoxia, may function as a collagen bridging protein (25, 22). However, this reduction in expression of TGFBI was not statistically significant. MMP2, which was elevated by hypoxia, is involved in extracellular matrix remodeling and has been shown to play an important role in angiogenesis (54). Somewhat paradoxically, we also found that MMP1, which is expressed at very high levels in our system, was downmodulated by hypoxia at both 8 and 24 h relative to normoxic control.
In conclusion, we have used SAGE to characterize the global temporal response of HAECs to short-term chronic hypoxia at the level of transcription. This identified numerous hypoxia-responsive genes representing a variety of functional classes. This information can be collated to build up a relatively detailed picture of the way in which HAEC molecular physiology is reprogrammed following exposure to hypoxia. In addition to providing comprehensive data regarding the hypoxia-responsive HAEC transcriptome in vitro, it provides a foundation for further studies of the molecular mechanisms by which cells respond to hypoxic stress. Further experiments will require validation of our findings in experimental systems that more closely represent physiological conditions. Until then, the current data provide a reference point for biologists interested in the genomic response to hypoxia in an in vitro vascular model system.
![]() |
GRANTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
ACKNOWLEDGMENTS |
---|
![]() |
FOOTNOTES |
---|
Address for reprint requests and other correspondence: D. G. Peters, Division of Pulmonary, Allergy, and Critical Care Medicine, 625 NW Montefiore Univ. Hospital, 3459 Fifth Ave., Pittsburgh, PA 15213 (E-mail: dgp{at}imap.pitt.edu).
1 The Supplementary Material for this article (Supplemental Tables S1 and S2) is available online at http://physiolgenomics.physiology.org/cgi/content/full/00221.2003/DC1.
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|