Differences in Vascular Bed Disease Susceptibility Reflect Differences in Gene Expression Response to Atherogenic Stimuli
Atherosclerosis occurs predominantly in arteries and only rarely in veins. The goal of this study was to test whether differences in the molecular responses of venous and arterial endothelial cells (ECs) to atherosclerotic stimuli might contribute to vascular bed differences in susceptibility to atherosclerosis. We compared gene expression profiles of primary cultured ECs from human saphenous vein (SVEC) and coronary artery (CAEC) exposed to atherogenic stimuli. In addition to identifying differentially expressed genes, we applied statistical analysis of gene ontology and pathway annotation terms to identify signaling differences related to cell type and stimulus. Differential gene expression of untreated venous and arterial endothelial cells yielded 285 genes more highly expressed in untreated SVEC (P<0.005 and fold change >1.5). These genes represented various atherosclerosis-related pathways including responses to proliferation, oxidoreductase activity, antiinflammatory responses, cell growth, and hemostasis functions. Moreover, stimulation with oxidized LDL induced dramatically greater gene expression responses in CAEC compared with SVEC, relating to adhesion, proliferation, and apoptosis pathways. In contrast, interleukin 1β and tumor necrosis factor α activated similar gene expression responses in both CAEC and SVEC. The differences in functional response and gene expression were further validated by an in vitro proliferation assay and in vivo immunostaining of αβ-crystallin protein. Our results strongly suggest that different inherent gene expression programs in arterial versus venous endothelial cells contribute to differences in atherosclerotic disease susceptibility.
Atherosclerosis is a chronic inflammatory disease with lipid deposition and accumulation in vascular walls over many years. Although systemic risk factors such as hyperlipidemia, diabetes mellitus, smoking, and hypertension contribute to its development, atherosclerosis preferentially affects certain vascular beds. In the normal anatomical locations, atherosclerosis predominantly affects arteries and rarely veins. Differences in hemodynamic environments between veins and arteries may play an important role in the development of atherosclerosis but cannot fully explain the differences in predisposition to atherosclerosis between different vascular beds. Even within the same arterial system, atherosclerosis tends to occur focally in certain predisposed regions. For example, the proximal left anterior descending coronary artery in the coronary circulation, the proximal portions of the renal arteries, and the carotid bifurcation in the extracranial circulation exhibit a particular predilection for atherosclerosis. Atherosclerosis also occurs in the aorta, the largest vessel in the body. Therefore, interplay between circulating factors and the properties of the local vascular wall may be critical for the initiation and development of atherosclerosis.
Vascular endothelial cells maintain the interface between the systemic circulation and soft tissues and mediate many critical processes such as inflammation, coagulation, and homeostasis. Vascular endothelium is also involved in a diverse array of pathological conditions including atherosclerosis and restenosis. To examine whether intrinsic differences between endothelial cells from different vascular beds could contribute to their different propensities to develop atherosclerosis, we compared gene expression profiles of endothelial cells from saphenous veins and coronary arteries in response to various atherogenic stimuli.
Gene expression profiling by high-density microarrays is a powerful tool for exploring complex interactive networks of genes and signaling pathways. However, identifying the appropriate biological context for large-scale changes in gene expression may be difficult, and the possible significance of the identification of a group of related genes as differentially regulated has not been adequately addressed with rigorous statistical methodology. Here, we have applied statistical analysis of gene ontology (GO) as previously published1 and used novel methods for analyzing pathway terms associated with regulated genes to better understand the full picture of gene expression differences observed in this study.
These studies reveal dramatic differences between basal and oxidized LDL (OxLDL) stimulated gene expression profiles for venous and arterial endothelial cells. The prominent array finding of differential proliferation potential was validated with in vitro studies. Further, we described detailed analysis of cellular functions and pathways that may underlie the different propensity for development of atherosclerosis in arteries versus veins.
Materials and Methods
In Vitro Cell Culture
Primary cultures of human coronary artery endothelial cells (CAEC) and human saphenous vein endothelial cells (SVEC) were obtained from Cambrex (Walkersville, Md). All cell types were plated on 100-mm or 150-mm culture dishes in EGM-2MV media (Cambrex) containing 5% fetal sera, vascular endothelial growth factor, basic fibroblast growth factor, epidermal growth factor, hydrocortisone, ascorbic acid, and antibiotics. The purity of endothelial cells was confirmed with acetylated-LDL (Ac-LDL) up-take test (BTI, Stoughton, Mass). Cells at passage 6 and approximately 80% confluence were subjected to serum-free medium for 12 hours and then exposed to OxLDL (40 μg/mL), tumor necrosis factor (TNF) α (10 ng/mL), interleukin (IL) 1β (10 ng/mL), platelet-derived growth factor (PDGF) (10 ng/mL), or vehicle for 24 hours. Fully oxidized LDL was purchased from BTI. TNFα, IL1β, and PDGF were obtained from CalBioChem.
Cells were harvested in TRIzol reagent (Invitrogen) and immediately stored at −80°C. The aqueous layer after phenol-chloroform extraction was applied to an RNeasy column (Qiagen, Valencia, Calif) for further purification as per the instructions of the manufacturer. The integrity of all samples was checked using the BioAnalyzer 2100 (Agilent, Palo Alto, Calif).
Oligonucleotide Microarray Expression Profiling
The Agilent human 1A oligo arrays used for these studies had approximately 22 000 features representing 16391 unique gene sequences. Detailed technical specification for these arrays can be found at http://www.chem.agilent.com/scripts/pds.asp?lPage=2433.
Total RNA from primary cell cultures and human universal RNA (Stratagene, La Jolla, Calif) used as common reference for all samples were labeled with the Agilent Direct Fluorescent labeling kit. The slides were scanned on an Agilent G2565AA Microarray Scanner System at 5% and 100% PMT settings to quantitate genes of both high- and low-signal intensities. Images were quantified using Agilent Feature Extraction Software (Version A.7.1.2).
Selection of Differentially Expressed Genes
Data from the images scanned at two different PMT settings were combined, and imported into the Resolver System (Rosetta Inc, Seattle, Wash). Combined gene expression ratios were computed by using an error-weighted approach where expression log ratios with less error contribute more to the combined result than those with greater error. Error weighted 1-way ANOVA was used in the comparisons between 2 different cell states. Fold changes of gene expression >1.5-fold and ANOVA probability values <0.005 were considered significant and were used in the following analyses.
Statistical Analysis of GO and Pathway Annotation Terms
The detailed methodology used for the GO and pathway annotation (PA) analyses can be found in the online data supplement available at http://circres.ahajournals.org. Briefly, all genes represented on the microarray were assigned to corresponding GO terms and PA membership using Biomolecule Naming Service (BNS).2 Of the total 16 400 genes on the array, 9828 genes annotated with GO terms were analyzed, with 8679 genes of cellular component, 7809 genes of biological process, and 7562 genes of molecular function. The pathway database was collected at Stanford University from sources such as KEGG, BioCarta, and SPAD and contained 360 curated pathway terms associated with 1698 genes. An additional 1357 genes were added to these pathways through scientific text association by searching more than 250 000 PubMed abstracts (supplementary Table X).3 Given a list of differentially expressed genes, we scored each GO or PA term using the hypergeometric distribution by comparing the number of genes in the list annotated by this term to the total number of genes annotated by this term. Probability values for each GO term were computed using the method described in the expanded Materials and Methods section in the online data supplement. Z scores for PA terms were computed as defined by Doniger et al4 and further converted into probability values for consistency with GO analysis. GO and PA terms with a probability value <0.005 were considered significant. To explore the impact of multiple testing correction, we applied Bonferroni correction to the probability values (supplementary Tables VIII and IX). However, Bonferroni correction assumes independence of the tests performed,5 and, given the complex dependency structure between terms in the GO hierarchy, Bonferroni adjusted probability values are likely overly conservative. Indeed, we observed that many important biological terms with significant corresponding gene counts were excluded in the corrected data.
Cell Proliferation Assay
CAEC and SVEC at 80% confluence were cultured in serum-free medium overnight and treated with OxLDL (40 μg/mL) or 5% FBS for 24 hours in the presence of 5-bromodeoxyuridine (BrdUrd). The Cell Proliferation ELISA kit (Roche) was used according to the instructions of the manufacturer.
Measurement of Gene Expression by QuantiGene Assays
The QuantiGene (branded DNA) assay was used to measure gene expression in aliquots of the same RNA preparations used for the microarray experiments. The QuantiGene assay kit was purchased from Genospectra (Fremont, Calif). Four hundred nanograms of total RNA for each sample were used in the assay according to the protocols of the manufacture.
Frozen sections of mouse aorta and vena cava obtained from Zyagen (San Diego, Calif) were stained with von Wildbrand factor (vWF) using Blood Vessel Staining kit (Chemicon), according to the protocol of the manufacturer. Polyclonal antibodies against αβ-crystallin were from QED Bioscience.
Gene Expression Profiling
Five separate cultures of CAEC and SVEC treated with OxLDL, TNFα, IL1β, or PDGF from 2 different donors were prepared to ensure that variations between donors and cell cultures did not significantly impact the data. All sample RNAs were labeled as red (cy5) channel and universal human reference RNA as green (cy3) channel. Each treatment and cell type had at least 4 microarray hybridizations. Differentially expressed genes were selected on strict criteria of probability value <0.005 and fold change >1.5 to reduce false positives. Reproducibility of microarray results as measured by the coefficient of variation between arrays for signal intensities in the reference RNA channel was 10.5% for all probes and 6.8% for probes with signal intensities in the top 75%. Furthermore, expression level of 5 genes was selected to compare the results between microarrays and QuantiGene assays in untreated and TNFα-treated cells. The results from the 2 methods were consistent with correlation of 0.98 (Table 1).
Baseline Differences Between Gene Expression Profiles of Coronary Artery and Saphenous Vein Endothelial Cells
We first compared gene expression profiles of the 2 cell types under basal conditions, in which cells were cultured with required survival factors without serum for 36 hours. The gene expression pattern was dramatically different between the 2 cell types. There were 1129 genes differentially (P<0.005) expressed between CAEC and SVEC (&7%) (supplementary Table I). When an arbitrarily chosen 1.5-fold cutoff was also applied, 285 genes were more highly expressed in SVEC and 111 genes more highly expressed in CAEC.
Enrichment of Atheroprotective Genes in Venous Endothelial Cells
The 111 genes overexpressed in CAEC were noticeably associated with inhibition of cell proliferation (eg, IL6, interferon [IFN] β1, and necdin [NDN]), lipid metabolism (eg, lipoprotein lipase [LPL], fatty acid binding protein 4 [FABP4], and collectin subfamily member 12 [COLEC12]).
Regulation of cell growth. These genes included ESP8 and MIG-6, which are known to regulate EGFR signaling,6,7 and angiogenic factors such as ANGPT1 and VEGFC. In addition, many important transcription factor genes involved in cell proliferation and differentiation, such as homeobox (HOX) D1, HOXD8, HOXA9, HOXA10, and HOXB7, were expressed at higher levels in SVEC (supplementary Table I).
Oxidoreductase activity and stress. Genes associated with the inhibition of reactive oxygen production were identified to be more highly expressed in SVEC, including RODH-4, COX6A2, and LOX (Table 2). In addition, CRYAB, a stress response gene, was 28-fold higher in SVEC compared with CAEC.
Immune and antiinflammatory responses. Several important genes directly or indirectly involved in the antiinflammatory response and atherosclerosis were also more highly expressed in SVEC (Table 2). A factor encoded by 1 such gene, PAFAH2, hydrolyzes oxidized phospholipids and inactivates the strong proinflammatory mediator PAF8 and may reduce inflammatory injuries of SVEC. The gene for apolipoprotein E, a molecule important for clearance of circulating cholesterol and antiinflammation, was >6-fold higher in SVEC than CAEC. Furthermore, 5 of 9 pregnancy-specific β-glycoprotein (PSG) isoforms represented on our arrays (PSG1, -3, -6, -7, and -9) were more highly expressed in SVEC than CAEC. These factors have been recently linked to the antiinflammatory responses.9,10
Fibrinolysis. Thrombogenesis is an important process in coronary atherosclerosis as well as in deep vein thrombosis. A group of genes important for fibrinolysis and inhibition of thrombin formation, including PLAT, F2, TFPI, and TFPI2, were more highly expressed in SVEC. The expression ratios were 1.8- to 6-fold (Table 2).
Statistical analysis of GO terms confirmed the overrepresentation of atheroprotective pathways such as “oxidoreductase activity” and “extracellular matrix” in SVEC (Figure 1). PA term analysis also found that genes involved in “fibrinolysis” and “extrinsic prothrombin activation” pathways were overrepresented in this list.
Markedly Different Gene Expression Responses to OxLDL in CAEC Versus SVEC
We found OxLDL induced dramatically different gene expression responses between CAEC and SVEC (Figure 2). In CAEC, 122 genes were upregulated and 175 genes downregulated by OxLDL, while in SVEC, 267 genes were upregulated and 550 genes downregulated (P<0.005 and >1.5-fold). The lists of all genes with P<0.005 can be found in supplementary Tables II and III.
Interestingly, only 36 genes induced by OxLDL were common between CAEC and SVEC. Bone morphogenetic protein 6 (BMP6) was among the genes upregulated in both CAEC and SVEC, although the role of this potent growth and differentiation factor in vascular function is not known. Common downregulated genes included several genes related to cholesterol/lipid synthesis like stearoyl-coenzyme A desaturase (SCD) and 3-hydroxy-3-methylglutaryl-coenzyme A synthase (HMGCS1).
Although little is known about the effect of OxLDL on SVEC, this stimulus had a potent effect on gene expression. There were 616 genes altered only in SVEC that were not significantly changed in CAEC and 160 genes altered by OxLDL only in CAEC without concurrent change in SVEC. A gene without change was defined as up- or downregulated with fold change <1.3 and probability value >0.005. For example, OxLDL significantly inhibited TNFRSF5 (CD40) expression in SVEC but not in CAEC, which may reduce the inflammatory response to atherosclerotic stimuli in SVEC.11 In addition, OxLDL downregulated THBS4 in SVEC but upregulated it in CAEC. Increased THBS4 activity has been associated with premature coronary artery disease and myocardial infarction.12
OxLDL-Induced Expression of Cellular Proliferation Genes in CAEC but Not in SVEC
OxLDL-induced sets of genes that regulate cellular proliferation only in CAEC. These genes included upregulated genes known to promote cell growth, such as protein tyrosine kinase SYK (SYK), VGEF receptors neuropilin (NRP) 1 and NRP2, and downregulated genes that inhibit cell growth, such as IGFBP3,13 fibroblast growth factor-inducible 14 (FN14), and IFN1β. Interestingly, we found OxLDL inhibited 8 of 21 genes from the H1 histone family present on the arrays. Histone H1 is a linker of histone proteins involved in the condensation of chromatin. Downregulation of H1 histone genes might decompress chromatin and facilitate cell proliferation. Paradoxically, we also found that OxLDL upregulated proapoptotic genes in CAEC, such as CASP3, TNFSF10, THBS1, and THBS4, and downregulated antiapoptotic genes such as TNF receptor–associated factor 4 (TRAF4), inducible T-cell costimulator (ICOS), and PIM1.
Identification of OxLDL-Activated Pathways That Distinguish CAEC and SVEC
By analyzing GO and PA terms, we found some pathways significantly overrepresented by OxLDL-regulated genes in CAEC but not in SVEC (Figure 3). For example, cell-adhesion pathways were not significantly upregulated in SVEC but were in CAEC, consistent with the previous report that OxLDL does not induce adhesion molecule expression in SVEC but does in arterial ECs.14 Furthermore, terms for the regulation of focal adhesion and inflammatory response in SVEC were associated with OxLDL-induced downregulated genes (Figure 3A). Similarly, genes downregulated by OxLDL in SVEC were significantly associated with pathways known to be important in atherosclerosis, such as “induction of apoptosis,” the “NF-KB pathway,” and “regulation of cell cycle” (Figure 3B). These results further support the idea that some genes related to protective pathways are activated in SVEC.
In arterial cells, we found that OxLDL activated pathways related to Alzheimer’s disease (Figure 3B). These genes included SYK, pyruvate dehydrogenase kinase 4 (PDK4), microtubule-associated protein 2 (MAP2), calmodulin 3 (calm3), CASP3, and aldehyde dehydrogenase 6 (ALDH1α3). The potential link between Alzheimer’s disease and atherosclerosis has been recently suggested in a mouse model.15
Strong Gene Expression Responses to TNFα and IL-1β in Both CAEC and SVEC
TNFα and IL1β are classical proinflammatory cytokines. The robust and similar gene expression response of CAEC and SVEC to TNFα and IL1β was in direct contrast to their significantly different response to OxLDL. Overall, CAEC had 1108 and 1130 genes that showed significant (P<0.005 and >1.5-fold) differential expression responses to TNFα and IL1β, respectively, whereas SVEC had 968 and 829 genes that responded (supplementary Tables IV through VII). Of these, a common set of 210 genes was upregulated and 42 genes downregulated by both TNFα and IL1β in CAEC and SVEC. Reassuringly, many genes known to be induced by TNFα and IL1β were near the top of the lists when sorted by fold change, including IL8, SERPINE1 (PAI1), E selectin (SELE), superoxide dismutase 2 (SOD2), TNFα inducible protein 6, ICAM1, VCAM1, IL6, and chemokine ligand 3 (CXCL3). Prominent TNFα and IL1β downregulated genes included matrix Gla protein (MGP), thrombomodulin (THBD), and CXC chemokine receptor 4 (CXCR4).
GO term analysis yielded a similar pattern of significant terms for IL1β and TNFα in both CAEC and SVEC, including pathways known to be associated with TNFα and IL1β. These terms included “inflammatory response,” “apoptosis,” “cell proliferation,” “immune response,” and “cytokine/chemokine” (Figure 3A). Similarly, PA term analysis found many pathways known to be associated with TNFα were significantly overrepresented, including “apoptosis,” “cytokine and inflammatory response,” “NF-KB signaling,” “Toll-like receptor,” “IL6, IL1, and IFN signaling” pathways (Figure 3B).
Further comparison of TNFα-induced gene expression profiles between CAEC and SVEC revealed a set of differentially responsive genes between the 2 cell types. There was a set of 810 genes in common regulated by TNFα in CAEC and SVEC, whereas 283 and 162 genes were altered by TNFα only in CAEC or SVEC, respectively. These differences may be important. For example, in CAEC, TNFα upregulated apoptosis genes, eg, CASP1, and downregulated antiinflammatory genes, eg, PAFAH2, but this effect was not seen in SVEC. GO terms related to antiapoptosis, an antiatherogenic process, were significantly overrepresented in genes regulated by TNFα and IL1β only in SVEC. Several pathways that mediate antiatherogenic processes, such as IL4 and inhibition of cellular proliferation, were significantly represented only in TNFα-regulated genes in SVEC. The finding that some antiatherogenic pathways are activated by TNFα and IL1β in SVEC but not in CAEC suggests that relatively subtle differences in the ability of venous versus arterial cells to respond to these cytokines may contribute to the differential propensity to develop atherosclerosis.
Confirmation of Microarray Results by Cellular Proliferation Assays and Immunostaining
The differential cell proliferation in basal state and in response to OxLDL between CAEC and SVEC revealed by gene expression profiling was further corroborated by a functional assay (Figure 4). When the same number of CAEC or SVEC was plated on wells of the same 96-well plate, SVEC grew faster (&1.7-fold) than CAEC in basal conditions (Figure 4). In OxLDL-treated cells, OxLDL increased cell proliferation in CAEC by approximately 3-fold. In contrast, OxLDL induced no significant change in SVEC proliferation. However, FBS added after overnight serum-free culture to SVEC induced a 3-fold increase in proliferation. This verified that SVEC were viable and capable of undergoing cell division.
The differential expression of αβ-crystallin (CRYAB) was further confirmed at the protein level in vivo by immunostaining of endothelial cells from sections of mouse aorta and vena cava (Figure 5). The vWF, an EC-specific marker, was stained specifically in endothelial cells of both vein and artery (Figure 5B and 5E). However, CRYAB protein levels were much higher in the endothelial cells of vena cava than of aorta (Figure 5C and 5F), which is in agreement with our microarray results (Table 1). CRYAB was also stained in smooth muscle cells (Figure 5F).
Atheroprotective Gene Expression Programs in Venous Endothelial Cells May Contribute to Atheroresistance
Veins and arteries are of different embryonic origin and serve different hemodynamic functions.16 Atherosclerosis commonly develops in arteries and is very rarely observed in veins under normal conditions. Although this difference in disease susceptibility is widely attributed to pressure and flow differences between the arterial and venous circulations, the molecular mechanisms underlying this phenomenon are poorly elucidated. In this study, we provide 2 lines of evidence to strongly suggest that intrinsic differences between venous and arterial endothelial cells at a gene expression level contribute to their dramatically different susceptibilities to atherosclerosis.
First, we identified differences in the basal gene expression patterns between SVEC and CAEC. Dysfunction of endothelial cells is a hallmark of the initiation of atherosclerosis.17 Many important genes that guard against endothelial dysfunction, including those linked to cell growth, oxidoreductase activity, antiinflammatory responses, fibrinolysis, extracellular matrix, and cellular communications, are more highly expressed in venous than arterial endothelial cells. Also, gene expression and functional assays reported here show that untreated venous endothelial cells have higher tendency for proliferation than arterial cells. The higher venous endothelial cell growth might be important in maintaining the integrity of the endothelium in the normal vascular wall and contribute to natural resistance to atherogenesis in veins.
Importantly, some genes differentially expressed between venous and arterial endothelial cells under in vitro culture conditions were also found to be differentially expressed in vivo in intact arteries and veins. In a separate series of profiling experiments, gene expression was compared between intact saphenous vein and intact (radial and mammary) arteries. Although these experiments were on a different microarray platform, and not globally comparable, a number of genes identified through the in vitro experiments were found to be differentially expressed in the vein to artery comparison. These genes included brain-derived neurotrophic factor, clusterin, integral membrane protein 2, and fatty acid binding protein 4, which had very similar vessel-specific relative ratios of expression (data not shown).
The observation that SVEC express a greater number of genes at high levels than CAEC suggests that venous ECs perform a greater number of cellular functions than arterial ECs. Recently, Chi et al compared gene expression profiles of ECs from various vascular beds and also found a greater number of genes expressed in venous ECs than arterial ECs.18 However, the venous ECs used in their studies were derived from the embryonic and atypical umbilical vein, and could not be considered representative.
The second line of evidence suggesting that differential gene expression contributes to vascular bed–specific atherosclerotic susceptibility is derived from studies reported here showing differential response to atherogenic stimuli, primarily OxLDL. OxLDL appears to play a critical role in the initiation of atherosclerosis by inducing gene expression patterns related to cell adhesion, cell stress, scavenger receptor activity, proliferation, apoptosis, and thrombogenesis.19 Although studies on the effects of OxLDL on arterial ECs or human umbilical vein ECs have been reported, this is the first side-by-side comparison of the differential OxLDL response of SVEC and CAEC. Our results clearly show that OxLDL can induce dramatically different gene expression responses in CAEC compared with SVEC. The observation that both CAEC and SVEC primary cultures used for these studies were capable of responding to TNFα and IL1β in a robust fashion suggests that the differences observed in the response to OxLDL were not attributable to an intrinsic abnormality of the cells.
One of the most important differences between CAEC and SVEC is the different proliferative response to OxLDL. Dual effects of OxLDL on cell proliferation and apoptosis have been reported in human umbilical endothelial cells previously, with the response depending on concentration and exposure time.20 In our studies using the medium dose range of 40 μg/mL, we found OxLDL activated genes related to both cell proliferation and apoptosis in CAEC but not in SVEC. Importantly, functional cell-proliferation assays clearly confirmed the differential proliferative response of CAEC to OxLDL compared with SVEC. In the context of vascular disease, this may be an important observation, because the promotion of both proliferation and apoptosis in EC by OxLDL has been suggested to result in high EC turnover, an important pathological process in atherosclerotic lesions.21
In a mouse atherosclerotic model, Shi et al reported that ECs from mice with different genetic backgrounds prone to diet-induced atherosclerosis have differential responses to OxLDL,22 suggesting that genetic differences in endothelial cells influence atherosclerosis susceptibility.22 Our results suggest that vascular bed–specific gene expression patterns also play an important role in modulating susceptibility to atherosclerosis. The fact that such patterns are maintained in vitro, in the absence of pressure and flow considerations, argue that the gene expression patterns are fundamental to the EC origin and differentiation.
Statistical Analyses of GO and PA Terms Are Effective Tools for Interpretation of Microarray Data
Application of GO terms in the analysis of microarray data has been published,4 but statistical analysis of PA terms has not. GO and PA terms are associated with different sets of genes, with a total of 9828 genes associated with GO terms, and 3055 genes associated with PA terms. There are 2667 genes that are associated with both GO and PA terms. Furthermore, the gene set associated with a PA term need not necessarily be associated with a GO term and vice versa. For example, genes associated with PA term “Apoptosis_Homo _sapiens” were associated with multiple GO terms, such as “induction of apoptosis,” “anti-apoptosis,” “regulation of apoptosis,” etc. Clearly, the PA and GO terms provide different kinds of biological information.
The effectiveness of the GO and PA term analysis methods can be assessed in the data set of TNFα and IL1β, because both TNFα and IL1β are classical proinflammatory cytokines and induce strong responses in ECs with well-defined signaling pathways. Accordingly, our GO and PA term analyses yielded a similar pattern of significant terms for IL1β and TNFα in both CAEC and SVEC. Pathways known to be associated with TNFα and IL1β, such as “inflammatory response,” “apoptosis,” “NF-KB,”“immune response,” and “cytokine/chemokine” were clearly identified. Furthermore, the signaling pathways identified by GO and PA term analyses overlapped considerably, but did provide some complementary information. For example, that TNFα- and IL1β-activated antiapoptosis pathways in SVEC were clearly identified by GO analysis, whereas TNFα- and IL1β-activated pathways related to Huntington’s disease in CAEC were more evident in PA analysis.
Taken together, these results suggest that GO and pathway term analyses provide useful tools for objectively analyzing regulated gene lists from gene expression profiling experiments and capture the complex interactions among different genes and signaling pathways. However, there are 2 main limitations of our current analysis. First, a pathway is more complex than a simple list of genes. For example, a specific gene could have multiple functions and could activate or inhibit the pathway with which it is associated. We have not incorporated this important factor into our statistical analysis yet. Second, based on current existing pathway databases, many important genes did not have informative GO and/or PA terms. We attempted to improve the representation of genes from our data with specific pathways by inputting gene/protein associations from the scientific literature.
Support for this study was provided by Agilent Technologies. We thank Drs Phil Tsao and Josh Spin for comments on the early version of this manuscript.
D.X.-F.D., A.T., A.V., A.B.-D., R. Kincaid, Z.Y., and L.B. are employed by and own equity in Agilent Technologies Inc, a company that sells a DNA microarray platform.
Original received November 5, 2004; revision received April 4, 2005; resubmission received November 21, 2005; accepted December 8, 2005.
Kincaid R, Kleusing D, Vailaya A. BNS: an LDAP-based biomolecule naming service. In: OiBC. Washington DC; 2002.
Vailaya A, Bluvas P, Kincaid R, Kuchinsky A, Creech M, Adler A. An architecture for biological information extraction and representation. Bioinformatics. 2005; 21: 430–438.
Blomberg LA, Cohn ML, Cahill RA, Chan WY. Effect of human pregnancy-specific beta1-glycoprotein on blood cell regeneration after bone marrow transplantation. Proc Soc Exp Biol Med. 1998; 217: 212–218.
Topol EJ, McCarthy J, Gabriel S, Moliterno DJ, Rogers WJ, Newby LK, Freedman M, Metivier J, Cannata R, O’Donnell CJ, Kottke-Marchant K, Murugesan G, Plow EF, Stenina O, Daley GQ. Single nucleotide polymorphisms in multiple novel thrombospondin genes may be associated with familial premature myocardial infarction. Circulation. 2001; 104: 2641–2644.
Bayes-Genis A, Conover CA, Schwartz RS. The insulin-like growth factor axis: a review of atherosclerosis and restenosis. Circ Res. 2000; 86: 125–130.
Amberger A, Maczek C, Jurgens G, Michaelis D, Schett G, Trieb K, Eberl T, Jindal S, Xu Q, Wick G. Co-expression of ICAM-1, VCAM-1, ELAM-1 and Hsp60 in human arterial and venous endothelial cells in response to cytokines and oxidized low-density lipoproteins. Cell Stress Chaperones. 1997; 2: 94–103.
Chi JT, Chang HY, Haraldsen G, Jahnsen FL, Troyanskaya OG, Chang DS, Wang Z, Rockson SG, van de Rijn M, Botstein D, Brown PO. Endothelial cell diversity revealed by global expression profiling. Proc Natl Acad Sci U S A. 2003; 100: 10623–10628.
Shi W, Haberland ME, Jien ML, Shih DM, Lusis AJ. Endothelial responses to oxidized lipoproteins determine genetic susceptibility to atherosclerosis in mice. Circulation. 2000; 102: 75–81.