| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Departments of Surgery
2 Statistics
3 Medicine, Duke University Medical Center, Durham, North Carolina 27710
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONDISCLOSURESReferences |
|---|
genomics; microarrays
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONDISCLOSURESReferences |
|---|
ßARK1, a member of the G protein-coupled receptor (GPCR) kinase (GRK) family, targets and phosphorylates agonist-occupied GPCRs, including myocardial ß-ARs, via binding to dissociated ß
-subunits of heterotrimeric G proteins (Gß) (16). This process leads to homologous receptor desensitization, and has been implicated as a key component in the pathogenesis of HF (16, 25). Recent studies in our laboratory have demonstrated that cardiac-targeted transgenic expression of a carboxy-terminal peptide of ßARK1 (ßARKct) can inhibit the activity of this GRK in vivo, and that it can ameliorate or prevent HF when expressed in hearts of various mouse models of HF, including the MLP-/- (MLP-/-/ßARKct) and CSQ (CSQ/ßARKct) mice (11, 24). This also includes enhanced survival and cardiac function, adding further evidence of the critical role of ßARK1 in the development of HF via its regulation not only of ß-ARs but potentially other cardiac GPCRs or signaling effectors (25).
These relatively well-characterized mouse models of HF and their ßARKct-mediated phenotypic rescue provide powerful models with which to investigate the development, progression, and rescue of the HF phenotype at the level of ventricular gene expression. Recently, there have been reports of differential gene expression in various phenotypes of both rodent and human HF (3, 4, 8, 10, 14, 20, 23, 29) as well as in humans before and after salutary mechanical circulatory support (5). Our goal for the current study was to employ both substantial sample size and robust statistical methods to establish statistically significant results and predictive capability, not only for comparison of the nonfailing and failing heart, but also for the progression and rescue of HF.
To address these issues, we performed oligonucleotide microarray analysis of LV gene expression in a total of 53 samples, including the following: 12 Normal (adult wild-type or littermate control); 12 HF (advanced stage HF adult MLP-/- or CSQ); 12 Rescue (adult HF + ßARKct); and 6 Tg (adult ßARKct transgenic), as well as an additional 5 young (mild HF) and 6 adult (advanced stage HF) HF mice. Subsequently, multiple statistical methods were used to significantly distinguish and/or predict distinct Normal, HF, Rescue, and Tg mice by gene expression profile and also identify specific genes significantly associated with each phenotype. Furthermore, we investigated changes in LV gene expression that may be associated with the progression of HF using both young mice with limited ventricular dysfunction and adult mice with advanced HF, and we further validated our predictive models with these mice.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONDISCLOSURESReferences |
|---|
Mice were genotyped, and cardiac function/phenotype was determined (by echocardiography) according to previously published protocols (11, 24) prior to inclusion in the microarray experiments. Total LV RNA was extracted using a tissue disruptor and Ultraspec RNA reagent according to the manufacturer’s protocol (Biotecx, Houston, TX). Total RNA quality was first assessed by 1.2% agarose-formaldehyde gel electrophoresis, followed by Agilent RNA 6000 LabChips coupled with the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA), where 500 ng were used to determine the 28S:18S rRNA ratio. All RNA used in the current study met the minimum requirements set of a 1.8 ratio for 28S:18S rRNA. RNA aliquots were stored at -80°C prior to use.
Northern blot.
Northern blot analysis of gene expression was performed using 10 µg total RNA as described previously (1).
Microarray analysis.
Expression analysis was performed using Affymetrix Murine 11K GeneChips, which represent 
11,000 known genes or novel clones (Affymetrix, Santa Clara, CA). Target cRNA was prepared according to the manufacturer’s protocol. Arrays were hybridized with the targets at 45°C for 16 h, then washed and stained using the GeneChip Fluidics Station. Arrays were visualized with the GeneChip scanner, and scanning results were processed by the GeneChip Expression Analysis Algorithm (version 3.2), with a target intensity of 500 (Affymetrix). GeneChips were processed using the same lot of reagents, equipment, fluidics station, and scanner, to minimize experimental microarray-to-microarray variability. Scaling factors for each GeneChip were within 10% of the mean of all GeneChips.
Statistical analyses.
Average difference values (perfect match to mismatch, or PM:MM ratios) provided by the Affymetrix analysis from each GeneChip were the starting point for all statistical analyses. ANOVA, t-test and multidimensional scaling (MDS) of total gene expression profiles were performed using Partek software (Partek, St. Charles, MO; http://www.partek.com/). For two-sample unequal variance t-test and ANOVA, corrected for multiple comparisons, all data for each microarray were individually linearized (individual data points for the entire data set for each microarray reduced on a linear scale into a range of 0–1) and log2 transformed, then this subset of results was compared with a list of genes with low false discovery rate determined to be significant by t-statistic score and class permutation tests (see below). MDS is a data structure visualization and analysis tool, which, in the simplest of terms, is a mapping of high-dimensional data (11,000 genes per animal) into low-dimensional space, such that the distances between points mapped in low-dimensional Euclidean space represent the similarities, dissimilarities, or distances of the original data. In this case, distance between points in the images approximates dissimilarity between data in high-dimensional space (DR). Herein, metric MDS was performed based on Pearson dissimilarity [defined as (1 - r)/2, where r is the Pearson correlation] and mapped into two-dimensional space (initialized with points drawn randomly from a uniform distribution in the range of -1 to 1) (see Ref. 9). Figures are generated by MDS to minimize a measure of badness of fit, called stress, which depends on the point-to-point dissimilarities in the high-dimensional data compared with the distances in low-dimensional space; all images presented herein had stress values from 0.003 to 0.041. In these images, axes x and y represent the two Euclidean dimensions into which the Pearson dissimilarities have been mapped. There was not a gene selection step prior to MDS calculation of dissimilarities. Hierarchical clustering, another method of data structure visualization, of animals according to total or statistically filtered gene expression was performed using Cluster and TreeView made available from Michael Eisen’s laboratory (http://rana.lbl.gov/EisenSoftware.htm), using default settings, being hierarchical clustering based on uncentered correlations with average linkage clustering. For all tests, where applicable, a corrected P value of <0.05 was considered significant. Because of the multiple testing setting and dependent gene interactions, even corrected P values should be interpreted with some care. Fold change calculations were performed using methods previously published (23). Clustering of data according to biological function and discovery of alternative identities for genomic entities from a simultaneous search of multiple databases was performed using the CELL suite of software from Incellico (Durham, NC; http://www.incellico.com/), and alternative identities are represented in italics in Tables 1 –3.
|
|
|
Database submission of microarray data.
The microarray data was prepared according to "minimum information about a microarray experiment" (MIAME) recommendations, has been deposited in the Gene Expression Omnibus (GEO) database, and can be accessed at http://www.ncbi.nlm.nih.gov/geo/. The samples can be retrieved with GEO accession numbers GSM-10178 through GSM-10283.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONDISCLOSURESReferences |
|---|
To determine the genomic profile of failing and nonfailing murine myocardium, LV gene expression data from Normal and HF mice were compared and analyzed by multiple statistical methods. A clear separation of these two groups was found by data visualization using both MDS (Fig. 1A) and hierarchical clustering (Fig. 1B). More rigorous analysis of the dissimilarity matrix from the MDS by a single exact test and randomization experiment demonstrated that the HF phenotype indeed significantly affected gene expression, in that of the possible 1.3 x 106 randomly assigned sample combinations, the most significantly divergent was that of the actual Normal and HF groups (P = 7.1 x 10-7, see Supplemental Fig. 2). Notably, hierarchical clustering also demonstrated a clear separation not only of Normal and HF, but also a reasonable segregation of MLP-/- and CSQ mice (Fig. 1B). Although one of these mice from the CSQ line appeared to be a slight outlier (HF 10), it was still classified as a HF mouse by all analyses (Fig. 1, A–C; also, see Supplemental Fig. 4).
|
-actin, FHL1, HB-EGF, desmin, osteoblast-specific factor 2, myosin light chain 1, laminin, collagen, and plasminogen activator inhibitor (3, 7, 12, 20) (see Table 1 and Supplemental Table 2). Importantly, a clear prediction of either Normal or HF phenotype was achieved based solely on the LV gene expression profile, using robust statistical methodology (a Bayesian regression model designed for gene expression data analysis; see MATERIALS AND METHODS and Ref. 33) (Fig. 1C). Verification of prediction by cross-validation was initially performed by one-at-a-time removal of each individual sample, fitting of the regression model using the remaining 23 samples, followed by class prediction (i.e., Normal or HF) of the single, removed sample. A majority of the genes found to be predictive of the HF phenotype by this method were shared in common with the HF-associated genes found by a corrected t-test (and Supplemental Tables 2 and 3). Therefore, we are able to detect by multiple statistical methods that there is an overall LV gene expression profile that cannot only distinguish, but also accurately predict, the HF phenotype independent of genetic background or etiology.
Previously, myocardial-targeted expression of ßARKct in mice (Tg) has been demonstrated to result in increased cardiac function and enhanced ß-AR responsiveness (17). However, in the current study, there was no significantly detectable difference at the level of gene expression between Normal and Tg mice using all methods described above (data not shown). Thus, minimal changes in gene expression appear to accompany the ßARKct cardiac phenotype in non-HF mice.
We were particularly interested in discovering changes in the LV gene expression profile following rescue of HF via cardiac expression of the ßARKct, especially in light of the minimally altered gene expression in the Tg hearts. Numerous genes associated with the HF phenotype (Table 1; also, see Supplemental Table 2) were significantly regulated toward Normal expression levels in the Rescue mice, as determined by a corrected ANOVA analysis of LV gene expression (Table 2; also, see Supplemental Table 4). Among the several genes significantly associated with both the development and rescue of cardiomyopathy were common marker genes of HF, such as upregulation of ANF, BNP, and ß-MHC in HF, and downregulation of these genes in Rescue mice. These would be anticipated to change in this manner given the phenotype of the Rescue mice. Beyond these and several other genes already correlated with HF, numerous others were identified having no previous association with HF or its phenotypic rescue (Table 2; see Supplemental Table 4). Cluster analysis of all genes significantly altered in both HF and Rescue revealed that the vast majority of these genes exhibited upregulated expression in HF and a reversion toward Normal expression levels in Rescue (Supplemental Fig. 3). Visualization of the entire data set by both MDS and cluster analysis demonstrated similar trends in reversion of the Rescue mice toward a Normal gene expression profile (Supplemental Fig. 4).
The MLP-/- and CSQ mice both develop a progressive HF phenotype. Thus we sought to determine the changes in gene expression that may accompany the progression from early to advanced stage HF. We determined LV gene expression in five mice with early stage HF (i.e., limited ventricular dysfunction) and in six mice with advanced stage HF (mice from both MLP-/- and CSQ lines), as determined by echocardiography (Fig. 2). A strikingly small number of genes were found to be differentially expressed as determined by a corrected t-test of normalized data (Table 3), including few that have been previously characterized in the HF phenotype. Surprisingly, both early and advanced stage HF mice were classified as HF by MDS (Fig. 3A), and there was no predictive difference between the early and advanced stage HF groups based on their gene expression profile. Importantly, both early and advanced stage HF mice were blindly predicted to be of a HF phenotype/gene expression profile using the predictive model described above followed by blind (i.e., not preclassified) addition of the data from the early and advanced stage samples and more rigorous full cross-validation, demonstrating the applicability of this methodology to blindly predict phenotype based solely on gene expression (Fig. 3B).
|
|
Secondary validation of oligonucleotide microarray expression data was carried out by Northern blot analysis for several genes, including the HF marker genes ANF and BNP. Data obtained by Northern blot were consistently qualitatively concordant with that found by the oligonucleotide microarrays (Fig. 4), similar to qualitative concordance found previously by others (8, 18, 19, 27, 33). These results demonstrate similar qualitative regulation of LV gene expression between phenotypic groups from both genetic models of HF. For further secondary validation, and to test the correlation between mRNA and protein expression, immunoblot analysis was performed. We chose the G protein Go, as its substantial upregulation was one of our top 20 overall predictors of the HF phenotype (Supplemental Table 5) and has also previously been associated with muscarinic regulation of L-type calcium channels (31). Indeed, there was good correlation between upregulation of both mRNA and protein expression in HF compared with Normal mice (Fig. 5).
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONDISCLOSURESReferences |
|---|
The HF phenotype was clearly distinguishable from Normal by gene expression analyses using multiple statistical methods, with close concordance between these methods for both distinction between phenotypic groups and identification of specific genes significantly associated with the distinction. Importantly, our predictive model demonstrated that the HF phenotype could be blindly predicted based solely on gene expression. Furthermore, the future applicability of this predictive model was validated with subsequent analysis of samples from advanced stage HF in our investigation of gene expression during the progression of cardiomyopathy.
Surprisingly, there appears to be little change in gene expression during progression of the HF phenotype in these mouse models, as determined both by our predictive model as well as other statistical methods. Although initially perplexing, these data strongly suggest very early "activation" of the HF gene expression profile in the progression of the disease. Although this finding remains to be tested in humans or other animal models of HF, it bears important implications regarding the potential of early HF diagnosis and treatment.
There have been several recent reports of differential gene expression associated with various HF phenotypes, both in rodents and humans (3, 4, 8, 10, 14, 20, 23, 29) (reviewed by Hoshijima and Chien, Ref. 12), as well as in humans before and after salutary mechanical circulatory support (5). Overall, we found good concordance between genes that were significantly associated with the development and/or rescue of HF with previous reports of HF-associated genes, including ANF, BNP, ß-MHC, skeletal--actin, FHL1, HB-EGF, desmin, osteoblast-specific factor 2, myosin light chain 1, laminin, collagen, and plasminogen activator inhibitor, as well as a substantial regulation of genes involved in metabolic processes and the extracellular matrix. In our analyses, numerous novel genes as well as known genes not previously associated with HF were found to be significantly altered in the HF and Rescue groups. Discussion of the relevance and significance of each gene identified in our analyses would be beyond the scope of this manuscript; however, all of these significantly altered genes are listed in the accompanying Supplemental Material, available from the Physiological Genomics web site.
The statistical detection of differential gene expression of numerous HF-associated genes provides a twofold validation of our approach. First, it validates these mouse models as appropriate genetic models of HF in which to further elucidate the molecular mechanisms of HF and test novel therapeutic targets. Second, it verifies that our approach using oligonucleotide microarrays analyzed with robust statistical methods indeed identifies genes significantly associated with and predictive of phenotype. Overall, we believe these data provide a statistically valid benchmark for future comparisons of murine cardiac phenotype with gene expression profile.
Our approach of using multiple statistical methods, including blind and cross-validated prediction of phenotype, enhances the validity of the data, in that each method generally reaches similar conclusions. Although there is some divergence in specific genes identified by each statistical method, genes that are common to multiple analytical methods are quickly recognized as the most favorable targets for future investigation (see Supplemental Tables 2–5). We have identified gene expression profiles germane to each phenotype, many of which are targets of current and future investigation in our laboratory.
An important finding of our study was that the phenotypic rescue of HF by ßARKct expression in the Rescue mice is significantly associated with a return toward Normal gene expression, which we believe further validates ßARK1 inhibition as a therapeutic target. Interestingly, in the ßARKct Tg mice, which demonstrate enhanced cardiac function (17), nonsignificant but qualitative salutary expression of a few marker genes of HF was detected compared with Normal, such as decreased skeletal -actin and cardiac -actin (data not shown). Our data suggest that even in normal myocardium, the ßARKct is both enhancing cardiac functional response and producing subtle yet potentially beneficial effects on gene expression. This beneficial effect on gene expression was more profoundly manifested in the analysis of Rescue mice, where the ßARKct not only rescued the functional phenotype (11, 24) but also resulted in a significant shift of the HF gene expression profile toward Normal, which although not blindly predictive, was detected by all of our other statistical methodologies. Of particular interest, changes in gene expression accompanying the (ßARKct) Rescue of HF, particularly decreased ANF and ß-MHC, were similar to those seen in human HF patients that responded to ß-blocker therapy (21). These data suggest that improved contractile function may be the stimulus for these significant changes in gene expression, since the ßARKct alone (Tg) did not significantly change the LV gene expression profile. Future studies will test this hypothesis by using our methods on other non-ßARKct rescued murine HF phenotypes (22, 26).
Overall, our current study considered two separate mouse models of HF and Rescue as a single group, to allow investigation of the gene expression profile germane to Normal, HF, and Rescue, along with the development of HF, independent of genetic cause or HF etiology. Although beyond the scope of the this report, it will be important to further investigate the similarities and differences between each of these models of HF and their Rescue, as well as to investigate the differences between etiologies of human HF. In our current investigation, we have identified methodologies, including novel statistical and predictive analyses, that demonstrate the potential for using multifactoral gene expression profiling coupled with all other available patient data as a predictive tool for cardiovascular diseases such as HF.
|
DISCLOSURES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONDISCLOSURESReferences |
|---|
|
ACKNOWLEDGMENTS |
|---|
Present address of B. C. Blaxall: Center for Cellular and Molecular Cardiology, Department of Medicine, University of Rochester, Rochester, NY 14642.
Present address of R. Spang: Max-Planck Institute for Molecular Genetics, Ihnestr. 73 Berlin, Germany.
|
FOOTNOTES |
|---|
Address for reprint requests and other correspondence: W. J. Koch, Center for Translational Medicine, Jefferson Medical College, 1025 Walnut Street, Room 822, Philadelphia, PA 19107 (E-mail: walter.koch{at}jefferson.edu).
10.1152/physiolgenomics.00087.2003.
1 The Supplementary Material for this article (four figures and five tables) is available online at http://physiolgenomics.physiology.org/cgi/content/full/00087.2003/DC1. 
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONDISCLOSURESReferences |
|---|
1B-adrenergic receptors. In vivo 1-adrenergic receptor-mediated regulation of ß-adrenergic signaling. J Biol Chem 272: 21253–21259, 1997.o is necessary for muscarinic regulation of Ca2+ channels in mouse heart. Proc Natl Acad Sci USA 94: 1727–1732, 1997.This article has been cited by other articles:
|
|
 |
|
  K. Pandya, K. Porter, H. A. Rockman, and O. Smithies Decreased beta-adrenergic responsiveness following hypertrophy occurs only in cardiomyocytes that also re-express beta-myosin heavy chain Eur J Heart Fail, July 1, 2009; 11(7): 648 - 652. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N. S. Rajasekaran, M. A. Firpo, B. A. Milash, R. B. Weiss, and I. J. Benjamin Global expression profiling identifies a novel biosignature for protein aggregation R120GCryAB cardiomyopathy in mice Physiol Genomics, October 8, 2008; 35(2): 165 - 172. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Zhu, A. A. Gach, G. Liu, X. Xu, C. C. Lim, J. X. Zhang, L. Mao, K. Chuprun, W. J. Koch, R. Liao, et al. Enhanced calcium cycling and contractile function in transgenic hearts expressing constitutively active G{alpha}o* protein Am J Physiol Heart Circ Physiol, March 1, 2008; 294(3): H1335 - H1347. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. A. Bullard, T. L. Protack, F. Aguilar, S. Bagwe, H. T. Massey, and B. C. Blaxall Identification of Nogo as a novel indicator of heart failure Physiol Genomics, January 17, 2008; 32(2): 182 - 189. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. Pawlinski, M. Tencati, C. R. Hampton, T. Shishido, T. A. Bullard, L. M. Casey, P. Andrade-Gordon, M. Kotzsch, D. Spring, T. Luther, et al. Protease-Activated Receptor-1 Contributes to Cardiac Remodeling and Hypertrophy Circulation, November 13, 2007; 116(20): 2298 - 2306. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. K. Dunnick, K. A. Thayer, and G. S. Travlos Inclusion of Biomarkers for Detecting Perturbations in the Heart and Lung and Lipid/Carbohydrate Metabolism in National Toxicology Program Studies Toxicol. Sci., November 1, 2007; 100(1): 29 - 35. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Mirotsou, V. J. Dzau, R. E. Pratt, and E. O. Weinberg Physiological genomics of cardiac disease: quantitative relationships between gene expression and left ventricular hypertrophy Physiol Genomics, January 12, 2007; 27(1): 86 - 94. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Rajan, S. S. Williams, G. Jagatheesan, R. P. H. Ahmed, G. Fuller-Bicer, A. Schwartz, B. J. Aronow, and D. F. Wieczorek Microarray analysis of gene expression during early stages of mild and severe cardiac hypertrophy Physiol Genomics, November 21, 2006; 27(3): 309 - 317. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. Pandya, H.-S. Kim, and O. Smithies Fibrosis, not cell size, delineates beta-myosin heavy chain reexpression during cardiac hypertrophy and normal aging in vivo PNAS, November 7, 2006; 103(45): 16864 - 16869. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Schiekofer, I. Shiojima, K. Sato, G. Galasso, Y. Oshima, and K. Walsh Microarray analysis of Akt1 activation in transgenic mouse hearts reveals transcript expression profiles associated with compensatory hypertrophy and failure Physiol Genomics, October 11, 2006; 27(2): 156 - 170. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Liang and B. Ventura Physiological genomics in PG and beyond: July to September 2005 Physiol Genomics, October 17, 2005; 23(2): 119 - 124. [Full Text] [PDF]
|
|
|
|
|
|
S. W. Kong, N. Bodyak, P. Yue, Z. Liu, J. Brown, S. Izumo, and P. M. Kang Genetic expression profiles during physiological and pathological cardiac hypertrophy and heart failure in rats Physiol Genomics, March 21, 2005; 21(1): 34 - 42. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
F. Schwartz, A. Duka, I. Duka, J. Cui, and H. Gavras Novel targets of ANG II regulation in mouse heart identified by serial analysis of gene expression Am J Physiol Heart Circ Physiol, November 1, 2004; 287(5): H1957 - H1966. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
