| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Department of Obstetrics, Gynecology and Reproductive Sciences, University of Pittsburgh, and Centre for Fetal Medicine, Magee-Womens Research Institute, Pittsburgh, Pennsylvania
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
gene expression; atherosclerosis; endothelial cell; serial analysis of gene expression
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
DNA microarrays have recently been exploited to investigate the transcriptional response of vascular endothelial cells to applied shear stress in vitro (7, 19, 32). One drawback of microarrays is that mRNA quantification relies on indirect measurement via hybridization of labeled cDNA (or cRNA) to the array. This process is known to suffer from numerous potential biases including, for example, cross-hybridization, inter- and intra-array variations, and intersample differences in the specific activity of the labeled target (17, 21). The lack of a direct relationship between signal intensity and mRNA abundance, between different probes on the same array, and between the same probes on different arrays requires that elaborate normalization strategies are employed (12). Furthermore, only genes present on the array can be assayed for their transcriptional response to a given stimulus. Because of these factors, the comparison of microarray data obtained on different platforms, in different labs or even at different times in the same lab, can be problematic.
In contrast to microarrays, direct sequencing methods for determining mRNA concentrations such as serial analysis of gene expression (SAGE) (52, 53) have distinct advantages. For example, SAGE does not demand prior knowledge of genes of interest and therefore does not exclude important genes from analysis. Furthermore, SAGE allows direct measurement of mRNA expression levels by gene-tag counting and thus provides the mRNA concentration of every gene in a library with respect to that of every other gene. In consequence, SAGE data require minimal normalization between libraries undergoing comparison. Therefore, SAGE data are immortal and can be readily shared and analyzed retrospectively in a variety of different contexts (37, 43).
We have utilized SAGE to characterize the transcriptome of human coronary artery endothelial cells (HCAECs) cultured for 24 h under static conditions or 15 dyn/cm2 laminar shear stress (LSS). Using this approach we have observed the differential expression of numerous genes that encode a variety of functional proteins.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
5 x 104 cells/cm2 on untreated 75 x 38 mm (Fisher Scientific, Pittsburgh, PA) glass microscope slides and cultured at 37°C in humidified 5% CO2-95% air. Approximately 48 h later confluent monolayers of cells were then placed in a parallel plate flow chamber [Cytodyne, La Jolla, CA (14)] under aseptic conditions and perfused according to the manufacturer's instructions in EGM2MV at 37°C in humidified 5% CO2 95% air for 24 h. Control cells not exposed to LSS were also cultured in EGM2MV for an identical length of time as LSS-treated cells.
RNA Isolation
Cells were harvested directly into TRIzol reagent (Life Technologies), and total RNA was extracted according to the manufacturer's instructions. RNA integrity was assessed by denaturing agarose gel electrophoresis, and its concentration and purity were determined by UV spectrophotometry. A260/280 ratios were >1.80 for all RNA samples. RNA pooled at a ratio of 1:1:1 from three identical experiments was used for SAGE.
Serial Analysis of Gene Expression
We used 15 µg total RNA as a substrate for SAGE, which was carried out as previously described (47). In brief, after mRNA purification via a biotinylated oligo-dT and streptavidin-conjugated paramagnetic beads (Dynal), double-stranded cDNA was synthesized using the Superscript system (Invitrogen) and then digested with NlaIII. Following ligation of a double-stranded linker, digestion with BsmF1, ditag ligation and purification via PCR, concatomer ligation, and plasmid transformation, SAGE tags were sequenced using a ABI3700 automated DNA sequencer. Primary sequence data were analyzed using the SAGE 2000 software package, which was kindly provided by Ken Kinzler of Johns Hopkins University, and raw tag counts were subject to normalization and statistical analysis as described below.
Confirmation of SAGE Results
For quantitative RT-PCR (QRT-PCR) [hemeoxygenase 1 (HMOX1) and transforming growth factor beta 1 (TGFB1)], total RNAs were purified by the RNeasy Mini Kit (Qiagen, Valencia, CA). Residual genomic DNA was eliminated by the DNA-free kit (Ambion, Austin, TX) according to the manufacturer's protocol and quantified by spectrophotometry (Beckman DU 640). The optimal reverse transcription was carried out in 100 µl volumes as described (10) and two RNA inputs (100 and 400 ng). No-reverse transcriptase controls were carried out with 400 ng of RNA. QPCR was performed on this cDNA on the ABI 7700 Sequence Detection Instrument (Applied Biosystems) using TaqMan MGB probes. PCR primers and probe were ordered from Applied Biosystems (HMOX1: Hs00157965_m1, TGFB1: Hs00998133_m1, GUSB: 4333767T). PCR amplification of cDNA derived from HAECs (n = 2) was performed in duplicate in 50 µl volumes as described (10) 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 β-glucuronidase (GUSB), using the comparative CT method described previously (10). Dot blot analyses [SPARC and polycystin 1 and 2 (PKD1 and PKD2)] were performed as previously described (42). Hybridization probes for SPARC and PKD1 and PKD2 dot blots were prepared by RT-PCR using gene-specific primers. These were radioactively labeled by random-primed DNA labeling in the presence of [
33P]dCTP. Signal intensity was measured using a Storm PhosphorImager (Molecular Dynamics).
[SPARC forward primer; 5'-CCTTTGCAAACACATTATGC; SPARC reverse primer, 5'-TCACACCTGTGACATCTTGC (387 bp): PKD1 forward primer, 5'-GGATGATTCTAAGAGTCTGG; PKD1 reverse primer, 5'-CCTGGACAGCCTCGCTGCCT (408 bp): PKD2 forward primer, 5'-CCGTGGATGACATTTCAGAG; PKD2 reverse primer, CCATCCAACAGCCTTCCCAGC (499 bp)].
Immunohistochemical analysis of HMOX1 and endoglin expression were carried out as follows. Endothelial cells were washed briefly in PBS and fixed in a 1:1 (vol/vol) methanol-acetone mix for 3 min. Hemeoxygenase protein was visualized using a monoclonal antihemeoxygenase antibody (A7811) supplied by Sigma (St. Louis, MO), and endoglin protein was visualized using a monoclonal anti-endoglin antibody (sc-7559) supplied by Santa Cruz Biotechnology (Santa Cruz, CA) and a 1:200 dilution of goat anti-rabbit IgG-FITC (Santa Cruz Biologicals, Sc.2012) secondary antibody. Nuclei were stained with 4',6-diamidino-2-phenylindole.
SAGE Data Analysis
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. Pearson's 
2 statistic is defined as:
 |
 |
Asymptotically, under the null hypothesis that t is not differentially expressed, T has a 
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 study, 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 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 Benjamini and Hochberg's linear step-up multiple-comparison procedure (BH procedure) (5). 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 P(i). Using this procedure, we will consider all the tags whose T statistics are greater than the 1 – P(i) quantile of the distribution to be 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 |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
2 score of at least 3.85. To increase the stringency of our analyses we next considered tags with a total count of at least 7 over the two libraries and controlled FDR at 5%. This approach identified 90 tags that were significantly elevated and 55 tags that were significantly reduced by LSS. All these tags have a 2 score of at least 7.65. Differentially expressed tags identified by this latter approach that correspond to known genes are shown in Supplemental Table S1.1
To avoid the possibility of excluding important data from follow-up experiments we also included a number of differentially expressed tags that have a 2 score of at least 3.85 and whose functions are of specific biological interest. Entire data sets, including gene lists generated with and without FDR control at 5%, are available from the authors on request. 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 OntoEx and FATIGO (2), which are united by their use of the Gene Ontology (GO) database provided by the GO consortium (http://www.geneontology.org). 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 GO information derived from other data sets. Specific terms for these processes were extracted from multiple levels within the GO hierarchy so that only major gene categories are discussed.
Differential Expression of Genes Encoding Factors Involved in Cell Proliferation and Angiogenesis by LSS
Exposure to LSS resulted in the differential expression of genes encoding factors involved in cell proliferation. These include cell division cycle 10 and 20 (CDC10, CDC20), cell division cycle 42 effector protein 2 (CDC42EP2), calpain small subunit 1 (CAPNS1), cyclin B1 (CCNB1), cyclin D1 (CCND1), enhancer of rudimentary homolog (ERH), nucleophosmin (NPM1), and both PKD1 and PKD2 (Fig. 1A). Furthermore, the expressions of a number of genes whose products promote angiogenesis were reduced by LSS. These include connective tissue growth factor (CTGF), cysteine-rich angiogenic inducer 61 (CYR61), phosphoglycerate kinase 1 (PGK1), osteonectin/SPARC (Fig. 1), TGFB1 (Fig. 1), endoglin (ENG) (Fig. 2), and angiopoietin-like 4 (ANGPTL4).
|
|
3 (TUBA3) and 6 (TUBA6) and β2 (TUBB2).
We also observed changes in the expressions of genes encoding integrins and matrix protein components. For example, fibrillin 1 (FBN1), fibronectin (FN1), integrin 5 (ITGA5), integrin E (ITGAE), integrin V (ITGAV), integrin-linked kinase (ILK), vimentin (VIM), laminin receptor 1 (LAMR1), lysyl oxidase-like 2 (LOXL2, which initiates cross-linking of collagens and elastin), and proline 4-hydroxylase (P4HA1, which catalyzes the formation of 4-hydroxyproline in collagen) were all reduced by LSS.
LSS Alters the Expression Levels of Genes Involved in Signal Transduction and the Acute Cellular Response to Stress
A number of genes encoding proteins involved in mediating signal transduction cascades are altered at the level of transcription by LSS. One particular LSS-induced example is GADD45, which mediates activation of the p38/JNK pathway, via MTK1, in response to environmental stresses (48). Consistent with this is the finding that LSS resulted in the induction of genes involved in mediating the cellular response to stress including a variety of heat shock factors (HSPF1, HSJ2, HSP105B, HSPB1, HSPKD1, HSPA1A, HSPA5, HSPCB) and metallothioneins (MT1A, MT2A). Genes whose expression levels are reduced that are involved in signal transduction include phosphatidylethanolamine-binding protein (PBP), which may cause elevation of MEK-, ERK-, and AP1-dependent transcription (62), and ras-related nuclear protein (RAN), which plays a key role in cell division by regulating microtubule polymerization during mitosis (58).
Differential Expression of Inflammatory Markers
We also found that the expressions of a number of genes involved in inflammation and leukocyte adhesion were increased by LSS including intracellular adhesion molecule-1 (ICAM1), interleukin-2 receptor-A (IL2RA), TNF receptor-associated factor 5 (TRAF5), and the tumor necrosis superfamily member TNFRSF1B. Both TRAF5 and TNFRSF1B are involved in mediating CD40 and TNF-directed proinflammatory events via NF-
B activation. A number of groups have previously reported the LSS-induced upmodulation of ICAM1 (35) and TRAF genes family upregulation in response to LSS has been reported in the context of CD40 inhibition (50).
LSS Response of Other Biologically Significant Genes
LSS exposure caused changes in the expression levels of a number of genes encoding factors involved in vasoreactivity. These include, aminopeptidase (ENPEP), which was elevated by LSS, while endothelin-1 (EDN1) and dimethylarginine dimethylaminohydrolase 1 (DDAH1) were reduced. Interestingly, we also found that mRNA levels of the HMOX1 gene were increased by LSS. This is significant given that one of the products of HMOX1 activity is carbon monoxide (CO), which has a vasodilatory effect on vascular endothelium (22). These findings were confirmed both by RTPCR and immunostaining (Figs. 1 and 2, respectively).
It can be seen from Fig. 1B that fold-change between 0 h and 24 h shear stress-exposed samples identified by SAGE was often either under- or overestimated compared with QRT-PCR and/or dot blot analyses. This is likely to be due to experimental variation. However, the LSS-responsive differential expression identified by microarray was qualitatively corroborated by follow-up analysis for the five genes examined.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
One disadvantage of SAGE is that it is a time-consuming and expensive technique. It is most likely because of this that it is common for investigators to generate only a single SAGE library for a given experimental condition (as have we) particularly within the context of an in vitro model system. Although like any experimental technique, there is the possibility that SAGE data will be somewhat compromised by experimental variation, a number of investigators have discovered that SAGE data are reproducible (11, 61). Similarly, although a major advantage of the SAGE method is that, in general, the concentration of a given tag is determined relative to that of every other tag in the library, there are some instances in which this is not the case. For example, a small minority of tags will not match a specific transcript in the genome databases or will have multiple matches. The problem of a single tag matching multiple genes has been partially overcome by the use of different type IIS restriction enzymes for tag generation, such as MmeI, that generate longer tags of 21 bp (45), although this approach tends to result in fewer overall matches and a higher proportion of unmatched tags. Similarly, investigators have utilized anchoring enzymes other than NlaIII to target different parts of the transcriptome, potentially allowing related libraries to be generated from a single RNA pool (54). The occasional problem of multiple tags matching to the same gene is likely to be caused by the detection of alternatively spliced transcripts encoded by the same gene but may occasionally reflect partial digestion of double-stranded cDNA during library preparation. Three such examples are present in Supplemental Table S1 (TMSB4X, ATP5D, HSPF1).
A limitation of our experimental data is that they have been generated using an experimental model system that is unable to determine the biological response to shear stress in the context of pulsatile flow and stretch. Furthermore, in this study we have not explored the response of previously conditioned endothelial cells to changes in shear forces. These are limitations that should be accounted for during interpretation of our findings. Such limitations could be partially avoided by using in vivo-derived tissue samples. However, such approaches would of course result in increased cellular complexity and intersample variation that might be difficult to control for in vivo. Another consideration for interpretation of our data is that, not only have our cells been cultured in vitro, but cells harvested from different vascular beds may vary at the level of transcription and in response to applied shear stress. We have chosen to use coronary artery endothelial cells, but comparison between these and other cell lineages would likely be informative. Indeed, such cell lineage-specific differences in the transcriptomic response to experimental stress have recently been demonstrated by our group in the context of hypoxia in endothelial cells obtained from the pulmonary artery and aorta (43).
It is known that exposure to LSS results in a reduction in endothelial cell rates of DNA synthesis and proliferation (36). Our data provide a global snapshot of the transcriptional characteristics of this antiproliferative molecular phenotype in that a number of genes involved in driving this process were reduced by LSS. For example, CDC20 is required for nuclear movement prior to anaphase and chromosome separation (13), CCNB1 complexes with p34(cdc2) to form the mitosis-promoting factor (44) and CCND1 functions as a regulatory subunit of CDK4 or CDK6, whose activity are required for cell cycle G1/S transition. CAPNS1 inhibition has been shown to decrease the growth rate of mammalian cells (60), ERH is a putative cell cycle gene that has been implicated in pyrimidine biosynthesis (59), and NPM1 is a target of CDK2/cyclin E in the initiation of centrosome duplication (39). The LSS-responsive elevation of PKD1 and PKD2 also leads us to speculate that, given the positive regulatory effect of PKD1 expression on cell cycle arrest (6) and the LSS-responsive increase in p21(waf1), the products of these genes (polycystin 1 and 2, respectively) might play a role in the negative regulation of endothelial cell proliferation.
The relationship between LSS and downmodulation of genes involved in the angiogenic process is also significant given LSS-specific antiproliferative phenotype discussed above. For example, CTGF is cysteine-rich secreted protein that has been shown to induce the proliferation, migration, and tube formation of vascular endothelial cells in vitro and angiogenesis in vivo (46). CYR61 encodes an extracellular matrix-associated protein that has similar features to CTGF. In culture, CYR61 functions through integrin-mediated pathways to promote cell adhesion, migration, and proliferation, cell survival, and tubule formation in human umbilical vein endothelial cells (26). PGK1 participates in the angiogenic process as a disulphide reductase, driving reduction of plasmin via proteolytic cleavage in the kringle 5 domain and release of the tumor blood vessel inhibitor angiostatin (25).
Of particular significance with respect to endothelial cell proliferation and angiogenesis is the almost complete downmodulation in our data of endoglin mRNA expression. Endoglin gene mutations are causative of hereditary hemorrhagic telangiectasia, which is a multisystem vascular dysplasia characterized by telangiectasia and arteriovenous malformations. DNA sequence variation in endoglin has also been associated with sporadic intracerebral hemorrhage (3), and recent reports have addressed the possible association of an intron 7 insertion polymorphism with intracranial aneurysm (IA) with conflicting results (23, 40, 41, 49). Our observation is significant given that IA pathogenesis is thought to have a strong hemodynamic component (33).
Changes in the levels of cytoskeletal genes are consistent with the well-documented alterations in the endothelial cell cytoskeleton following exposure to shear stress (18) and the previously identified LSS-induced reduction in cell proliferation (36). For example, MACF1 is a cytoskeletal linker protein (15) involved in stabilizing actin at sites where microtubules and microfilaments meet. It may function in microtubule dynamics to facilitate actin-microtubule interactions at the cell periphery and to couple the microtubule network to cellular junctions. CNN2 is involved in the structural organization and/or anchorage of actin filaments (31), and FLNA cross-links actin filaments to membrane glycoproteins. Similarly, TMSB4 has a critical role in modulating the dynamics of actin polymerization and depolymerization in nonmuscle cells, and CFL1 is involved in actin filament turnover (16, 28, 51).
It is known that LSS has a vasodilatory effect on endothelial cells in vitro, and, not surprisingly, we observed a transcriptional response among genes involved in these processes. The transcriptional repression of EDN1 has been previously reported, but we also saw a reduction in DDAH1, which inhibits nitric oxide synthase via the regulation of cellular methylarginine concentrations and may be involved in regulating nitric oxide generation in endothelial cells (1). The increase in ENPEP modulation suggests a possible shear stress effect on the angiotensin signaling system since this enzyme is known to be involved in angiotensin degradation (34). Furthermore, the induction by LSS of HMOX1 mRNA synthesis is significant, given that HMOX1 activity results in the production of CO, which can act as a powerful vasodilator (22).
The alteration in expression of PKD1 and PKD2 in our data is notable. Autosomal dominant polycystic kidney disease (ADPKD) is caused by mutations in either the PKD1 or PKD2 (and less commonly, PKD3) genes. ADPKD is characterized by fluid filled renal cysts, but, significantly, ADPKD sufferers exhibit impaired endothelial dependent relaxation (9, 57). Whether this impaired function is related directly to dysfunction of cellular pathways involving PKD1/2 is unclear. Given that PKD1 and 2 are expressed in endothelial cells and are LSS-response at the level of transcription, the possibility that PKD1 and PKD2 are directly involved in the vascular manifestations of ADPKD is worthy of further investigation. This is further supported by evidence that PKD1 and 2 are involved in the homeostatic control of vascular tone in vivo (55, 56). One possibility is that the modulation of PKD1 and 2 are associated with LSS-responsive changes in cell proliferation. Specifically, is has been shown that expression of PKD1 activates the JAK-STAT pathway in renal cells, thereby upregulating cyclin-dependent kinase inhibitor 1A (CDKN1A) and inducing cell cycle arrest in G0/G1 (6). Interestingly, our data and that of others (29) show that p21(waf1) mRNA is modestly increased after 24 h exposure to LSS. Furthermore, it has been shown that absence of a functional PKD1 gene results in angiogenesis of renal vasculature (4), and it has been recently suggested that PKD2 interacts with troponin I, an angiogenesis inhibitor (13, 27). Taken together, these observations suggest that the antiproliferative effect of LSS may involve PKD1- and PKD2-mediated signaling events. This possibility has important implications for efforts to understand the molecular basis of vascular disease associated with ADPKD.
In summary, we have used SAGE to characterize the global response of HCAECs to 24 h applied 15 dyn/cm2 LSS at the level of transcription. This identified numerous LSS-responsive genes encoding a variety of functional classes. In addition to providing information regarding the LSS-responsive endothelial cell transcriptome in vitro, these data provide a foundation for further studies of the molecular mechanisms by which cells respond to LSS. This information has relevance to cerebrovascular disease with respect to our understanding of the mechanisms underlying the atheroprotective effects of LSS. Furthermore, an improved understanding of the endothelial cell response to hemodynamic stress will potentially focus and accelerate the candidate gene discovery process for IA and improve our understanding of the pathogenesis of this disease.
|
GRANTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
1 The online version of this article contains supplemental material. 
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
