Identification of a Core Set of Genes That Signifies Pathways Underlying Cardiac Hypertrophy

Although the molecular signals underlying cardiac hypertrophy have been the subject of intense investigation, the extent of common and distinct gene regulation between different forms of cardiac hypertrophy remains unclear. We hypothesized that a general and comparative analysis of hypertrophic gene expression, using microarray technology in multiple models of cardiac hypertrophy, including aortic banding, myocardial infarction, an arteriovenous shunt and pharmacologically induced hypertrophy, would uncover networks of conserved hypertrophy-specific genes and identify novel genes involved in hypertrophic signalling. From gene expression analyses (8740 probe sets, n = 46) of rat ventricular RNA, we identified a core set of 139 genes with consistent differential expression in all hypertrophy models as compared to their controls, including 78 genes not previously associated with hypertrophy and 61 genes whose altered expression had previously been reported. We identified a single common gene program underlying hypertrophic remodelling, regardless of how the hypertrophy was induced. These genes constitute the molecular basis for the existence of one main form of cardiac hypertrophy and may be useful for prediction of a common therapeutic approach. Supplementary material for this article can be found at: http://www.interscience.wiley.com/jpages/1531-6912/suppmat


Introduction
Cardiac hypertrophy is a compensatory mechanism to augment cardiac output after biomechanical stress. Sustained hypertrophy leads to cardiac dysfunction, heart failure and arrhythmia, hence hypertrophy is an independent risk factor for cardiac morbidity and mortality (Agabiti-Rosei et al., 1997). Understanding the genes that govern cardiomyocyte hypertrophic properties has implications for both clinical medicine and basic cell biology.
Cardiac hypertrophy appears in different phenotypes, depending on the eliciting stimuli; however, it is unclear whether this is the result of divergent transcriptional responses or if a common hypertrophy gene program exists (Aronow et al., 2001;Chien et al., 1991;Wollert et al., 1996). The structural changes in the left ventricle including eccentric and concentric hypertrophy can be mimicked in animal models by volume and pressure overload procedures, respectively. Eccentric hypertrophy is characterized by left ventricular dilation 460 C. C. Strøm et al. and decreased wall thickness, while in concentric hypertrophy the ventricular wall is thickened and the left ventricle reduced. Typically, arteriovenous shunts and myocardial infarction induce eccentric hypertrophy, while aortic banding elicits the concentric phenotype.
However, a comprehensive analysis and direct comparison of the expression changes in different forms of cardiac hypertrophy is lacking. Such an approach may identify shared molecular networks in hypertrophy and point to new therapeutic strategies that interrupt the underlying disease pathways. DNA microarrays allow simultaneous analysis of thousands of genes, and have been useful in analysis of cellular responses to different stimuli, animal models of human disease and cancer classification (Golub et al., 1999;Roberts et al., 2000). Using this technology, we attempted to define a unified gene response that characterizes cardiac hypertrophy by comparing transcriptional profiles from multiple models, including aortic banding, myocardial infarction, arteriovenous shunting and pharmacologically induced hypertrophy. Our results provide a detailed view of the shared gene expression patterns in different hypertrophic phenotypes. Out of 8740 analysed genes on 46 microarrays, we identified a set of 139 genes with common regulation in all hypertrophy models, supporting the notion of a common hypertrophic gene program.

Cardiac hypertrophy models
Male Wistar rats were used in all experiments. All operations were performed during anaesthesia with fentanyl/fluanisone and midazolam. The investigation conforms to the Guide for the Care and Use of Laboratory Animals, published by the US National Institutes of Health (NIH Publication No. 85-23, revised 1996). Animals were subjected to echocardiographic examination under isoflurane anaesthesia before being sacrificed. Thereafter, the hearts were excised, rinsed in ice-cold saline, weighed, dissected into left and right ventricles, frozen in liquid nitrogen and stored at −80 • C before mRNA extraction.
Neonatal hearts were harvested 3 days postpartum.

Aortic banding (AB)
A titanium clip with an inner diameter of 0.6 mm was placed around the ascending aorta, using a custom-made applicator (Weck Closure Systems, USA). Control animals underwent the same procedure except for placement of the clip. The animals were sacrificed after 6, 12, 16 or 30 weeks.

Myocardial infarction (MI)
The LAD coronary artery was occluded with a 6-0 silk ligature immediately below the left atrial appendage. In sham-operated animals, the ligature was passed under the artery and removed. Animals were sacrificed 3 or 9 weeks after surgery. Only hearts with large transmural infarctions were considered for microarray analysis. The scar tissue was removed and discarded.

Aorto-caval fistula (shunt)
The aorta was punctured caudal to the left renal artery with an 18-gauge needle that was subsequently advanced into the vena cava. The needle was withdrawn and the aortic puncture point was sealed with cyanoacrylate glue. The patency of the shunt was verified visually before closure. Control animals underwent the same procedure except for puncture of the vessels. The animals were sacrificed after 3 or 8 weeks.

Hormone treatment
Angiotensin II (AngII; 200 ng/kg/min) was administered subcutaneously for 2 weeks, using implanted miniosmotic pumps (ALZET  , USA). AngII was diluted in 0.9% NaCl with 0.01 M acetic acid. Age-matched control animals received pumps containing vehicle only. The thyroxin analogue Echocardiography Echocardiography was performed during anaesthesia with 1-1.5% isoflurane using a Vivid Five Echocardiograph from GE Medical Systems Ultrasound. Recordings were stored digitally for off-line analysis. Left ventricular cavity and wall dimensions were measured in 2D short axis recordings.

Gene expression profiling
GeneChip RGU34A from Affymetrix containing 8740 genes (and 59 control genes, which were excluded from further analysis) was used for all hybridizations. Approximately 6000 known genes are represented on the chip, the rest being ESTs (see www.affymetrix.com for a more detailed description). Standard protocols for chip hybridizations were used. At each time point in each model, both diseased and control animals were randomly split into two to four groups, and RNA isolated from the left ventricle from animals in each group were pooled and hybridized to the oligonucleotide arrays. Thus, each time point in each model is represented by four to six (two to four diseased, and two controls) independent RNA pools, each hybridized to an array, in total 46 arrays. Raw data are available at www.ncbi.nlm.nih.gov/geo as Series No. GSE 738.

Array data analysis
Array data were normalized using the non-linear invariant rank fitting method of Li and Wong, available at www.dchip.org (Li et al., 2001a(Li et al., , 2001b. Model-based expression values (MBE) were calculated for each gene using dChip (perfect match only model). For each hypertrophic sample, a log fold change was calculated for each gene by dividing the MBE value by the average MBE value of the two relevant control samples and then applying log 2 . This approach was taken to minimize variance not related to hypertrophy between samples.

Statistical significance
Genes with common regulation in different hypertrophic samples were identified using Wilcoxon's rank sum test on the log fold changes, while ANOVA was used to identify differences between the models. The p values calculated for each gene were Bonferroni corrected by discarding all genes with a p value higher than 0.05 divided by the number of genes. The Bonferroni correction is a method to control type 1 error when doing multiple comparisons. The principle is to divide the desired type 1 error (usually 0.05) by the number of comparisons.

Clustering
For calculation of distances between genes, fold change was calculated by divided individual MBE by the average MBE of all control samples and applying log2. Vector angle distance on means normalized data is identical to the Pearson correlation. So the only difference between Pearson correlation and vector angle is a normalization step. Hierarchical clustering was applied (weighted pair-group average linkage) and visualized using ClustArray (C. Workman, unpublished; www.cbs.dtu.dk).

Real-time PCR
cDNA was synthesized from single samples previously analysed on GeneChips. Reverse transcription was performed using Superscript II RT (Invitrogen). 1 µg total RNA and 1 µl 50 pmol/µl (dT) 24primer in a total volume of 12 µl was incubated for 10 min at 70 • C and chilled on ice. After adding 4 µl 1st Strand Buffer (from supplier), 1 µl DTT (0.1 M), 2 µl dNTP mix (10 mM) and 1 µl Super-Script RT II (200 U/µl), the reaction was incubated for 1 h at 42 • C and finally for 5 min at 95 • C. The cDNA was diluted 1 : 20 for use in real-time PCR. Real-time PCR analysis was performed on selected genes using the primers shown in Figure 1. Primers were designed using the Primer3 software available at www-genome.wi.mit.edu/cgibin/primer/primer3 www.cgi. Triple determinations were performed on the ABI PRISM  7000 Sequence Detection System using the SYBR  Green PCR Master Mix (Applied Biosystems, USA). To determine the relative gene expression, a standard curve was made with serial dilutions of a reference sample. The gene expression in the sample of interest was then determined relative to the expression in the reference sample via the standard curve (built-in feature of the analysis software, ABI Prism 7000 SDS Software 1.0.1). The PCR reaction consisted of 12.5 µl SYBR Green PCR Master Mix, 300 nM forward and reverse primers, and 2.5 µl 1 : 20 diluted template cDNA in a total volume of 25 µl. The reaction was thermocycled using the default settings of ABI Prism 7000 SDS Software 1.0.1: 2 min at 50 • C, 10 min at 95 • C, followed by 40 rounds of 15 s at 95 • C and 1 min at 60 • C. A dissociation protocol was added after thermocycling, determining dissociation of the PCR products from 65 • C to 95 • C. All samples were normalized to GAPDH. For each sample a triple determination was made for each gene of interest and the normalization gene (GAPDH). An average was calculated for the triple determinations. To normalize the gene expression for a specific sample, the average expression value was divided by the average expression value for the normalization gene in the same sample. According to the GeneChip data, GAPDH is consistently expressed in our samples. Signals from GeneChip analyses were compared to the normalized real-time PCR data.

Animal models for cardiac hypertrophy
All experimental models of cardiac hypertrophy resulted in significant left ventricular hypertrophy (Table 1). In animals with a myocardial infarction (MI), echocardiographic examination revealed massive dilation of the left ventricular cavity, thinning of the anterior wall due to infarction, and decreased systolic function (Table 2). Pressure overload induced by aortic banding (AB) resulted in wall thickening but no dilation of the left ventricular cavity consistent with concentric hypertrophy. Systolic function was preserved in these animals. Volume overload induced by a shunt between the aorta and the caval vein was characterized by ventricular dilation and no or marginal increases in wall thickness consistent with eccentric hypertrophy. Systolic function was marginally depressed in the volume-overloaded animals.
A unified gene expression response to cardiac hypertrophy To identify genes with a common gene expression pattern in hypertrophy, we analysed cardiac gene expression by DNA microarrays. After normalization to the appropriate controls, consistent gene expression changes between the different phenotypes of cardiac hypertrophy were identified by comparing gene expression changes in the different models using a Wilcoxon test. To avoid false positives as a result of multiple testing, p values were Bonferroni-corrected, such that the risk of one or more false positive results was less than 5%. This procedure identified 179 genes, representing 139 known genes, 13 ESTs and 27 gene duplicates that were significantly regulated in hypertrophic hearts vs. controls (Table 3 in this paper, and  Supplemental Table A). Among the 179 genes, 137 were upregulated and 42 downregulated. Of the 139 known genes, 61 have previously been reported to change expression in response to hypertrophy, while 78 have not previously been connected with hypertrophy. A literature search revealed that 30 of the 137 genes have previously been reported to change at the protein level (Supplemental Table  A). This confirms that many of the transcriptome changes actually result in changed cellular protein, considering that only 61 genes had previously been associated with hypertrophy. Figure 2 depicts a hierarchic cluster analysis of the commonly regulated genes. All duplicate genes were equally regulated and clustered tightly together. The genes fell into two major clusters, representing up-and downregulated genes, respectively.
Classification into seven major categories based on biological function revealed that the core hypertrophy-related genes were predominantly involved in metabolism, signal transduction, cytoskeletal/ecm organization, and cell defence/inflammation (Figure 3). Most notably, in the cluster of downregulated genes ( Figure 2) the vast majority of genes were involved in β-oxidation of fatty acids and oxidative phosphorylation.
Cardiac hypertrophy is associated with extensive remodelling of the extracellular matrix. In line with this notion, we observed enhanced expression of several ecm genes previously associated with hypertrophy (fibronectin, vimentin, biglycan, tissue-inhibitor of metalloproteinase-1 and -2, and osteopontin) and genes not previously connected with myocardial remodelling (protease nexin-1, profilin-1 and osteoprotegerin). Several cytoskeletal genes were also upregulated (skeletal and vascular smooth muscle α-actin). Several genes involved in inflammation were upregulated (Hsp27, Psme1, Pai1, and several complement genes), confirming the notion that inflammation is important in the hypertrophic process.
Genes involved in cell growth and proliferation included cyclin D2, FSTL1 and S100A4, confirming previous observations that cyclin D2 protein is upregulated in aortic-banded rats (Busk et al., 2002). FSTL1 and S100A4 are secreted proteins that have been implicated in cancer cell growth.

Differences in gene expression patterns between hypertrophic phenotypes
We next looked for the transcriptional responses specific to each hypertrophy model by asking which genes showed expression differences between the different hypertrophic models, using ANOVA. This analysis identified 44 genes with differential expression between the models (Supplemental Table B). The genes represented 40 known genes, two ESTs and two replicates of ANP. Of these genes, 15 differed only in fold change between the models but were consistently either upor downregulated in all models compared to controls (Supplemental Table B). A hierarchic cluster analysis of the 44 genes showed tight clustering of the disease models but separate clustering of the pharmacologically treated animals (Supplemental Figure A). Thus, the differences between the models primarily resulted from differences between the disease models and the pharmacologically treated animals. After exclusion of the pharmacologically treated animals, only seven genes were differentially expressed: Mapk7, skeletal α-actin, Csrp2, acetyl-CoA acyltransferase, ANP (two genes) and an EST.

Confirmation by real-time PCR
To independently confirm the microarray data, mRNA levels of eight genes were analysed by realtime PCR (Figure 1). The real-time PCR analyses all confirmed the microarray findings, although the fold change tended to be greater when determined by real-time PCR.

Discussion
Our work addresses global aspects of gene regulation in surgical and pharmacological models of cardiac hypertrophy in mammals. Is there a common set of genes that largely defines cardiac hypertrophy in these models? To address this question, we compared transcriptome responses among five models of cardiac hypertrophy. We identified a set of 139 genes and 13 ESTs that were commonly regulated in cardiac hypertrophy in response to diverse    Figure 3. Known genes (n = 139) with common expression in all hypertrophic models were categorized based on function, as determined from database searches. Genes involved in the cytoskeleton/extracellular matrix (ecm) organization, cell death/apoptosis, and cell growth/proliferation were exclusively upregulated. The vast majority of downregulated genes were involved in the β-oxidation of fatty acids and naturally occurring pathological stimuli, such as pressure overload, volume overload and myocardial infarction. Overall, this outcome is important for two reasons. First, these transcriptome changes are likely to signify molecular events necessary for, or a consequence of, the hypertrophic process in these models. Second, using a global approach, our data support and expand the notion of a universal gene program underlying cardiac hypertrophy in response to multiple diversified stimuli. This paradigm was previously based on transcriptional changes in a small set of genes, including immediate early genes, ANP, BNP, and sarcomer genes constituting the socalled 'fetal gene program' (Chien et al., 1991). Many genes in our core set including ANP, BNP, β-actin, procollagen, fibronectin, GATA4, ribosomal proteins, biglycan, vimentin, Hsp27, TSC22, PAI1, and osteopontin were consistently regulated in previous microarray studies analyzing the myocardium after MI in rats, in failing hearts from hypertensive rats, in neonatal and adult mice and in failing and hypertrophic human hearts (Supplemental table A) (Anversa et al., 1983;Barrans et al., 2002;Friddle et al., 2000;Hwang et al., 2000Hwang et al., , 2002Schoenfeld et al., 1998;Sehl et al., repressed, as depicted in Supplemental Figure B. Many genes involved in the mitochondrial respiratory chain, e.g. OSCP, SDHA, NNT, COX8h and ETFA, were downregulated, indicating a general defect in energy metabolism in hypertrophic hearts, as also noted by others (Stanton et al., 2000;Yang et al., 2000).
A notable finding was that upregulation of three annexins (I, II and V), members of a family of proteins that have anticoagulant effects, inhibits leukocyte infiltration, and form Ca 2+ channels. Annexin upregulation might have beneficial effects. Annexin I peptides protect against ischaemia-reperfusion injury of the myocardium, most likely by inhibiting leukocyte infiltration (La et al., 2001). Annexin channel formation might be required for changes in Ca 2+ homeostasis and gene expression in hypertrophic cardiomyocytes. In hypertrophic chrondocytes, annexin-mediated Ca 2+ influx activates gene expression of Cbfa1, ATPase and osteocalcin, which are differentiation markers in these cells (Wang et al., 2003).
Another interesting finding was increased expression of granulin, a growth factor that resembles the EGF/TGFα family, although granulin does not activate EGF receptors (Bateman et al., 1998). Granulins regulate differentiation/proliferation of multiple cells, including epithelial and haematopoietic cells and carcinoma and breast cancer cell lines. Granulin was found upregulated in human glioblastomas in a cDNA microarray study (Liau et al., 2000). Further work will reveal whether granulin plays a role in cardiomyocyte hypertrophy. It should be noted that TSC-22 (TGFβcontrolled transcription factor), FSTL1, and PAI-1 (TGFβ-induced effectors) were upregulated, suggesting that activation of TGFβ family signalling is involved in hypertrophy, as also noted by others (Stanton et al., 2000).
In conclusion, we have identified a set of genes common to several different clinically relevant phenotypes of cardiac hypertrophy. These genes are likely to be central to hypertrophic phenotype, irrespective of the initiating cause.