Differential myocardial gene expression in the development and rescue of murine heart failure

Burns C. Blaxall1, Rainer Spang2, Howard A. Rockman3 and Walter J. Koch1

1 Departments of Surgery
2 Statistics
3 Medicine, Duke University Medical Center, Durham, North Carolina 27710


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 DISCLOSURES
 References
 
Numerous murine models of heart failure (HF) have been described, many of which develop progressive deterioration of cardiac function. We have recently demonstrated that several of these can be "rescued" or prevented by transgenic cardiac expression of a peptide inhibitor of the ß-adrenergic receptor kinase (ßARKct). To uncover genomic changes associated with cardiomyopathy and/or its phenotypic rescue by the ßARKct, oligonucleotide microarray analysis of left ventricular (LV) gene expression was performed in a total of 53 samples, including 12 each of Normal, HF, and Rescue. Multiple statistical analyses demonstrated significant differences between all groups and further demonstrated that ßARKct Rescue returned gene expression toward that of Normal. In our statistical analyses, we found that the HF phenotype is blindly predictable based solely on gene expression profile. To investigate the progression of HF, LV gene expression was determined in young mice with mildly diminished cardiac function and in older mice with severely impaired cardiac function. Interestingly, mild and advanced HF mice shared similar gene expression profiles, and importantly, the mild HF mice were predicted as having a HF phenotype when blindly subjected to our predictive model described above. These data not only validate our predictive model but further demonstrate that, in these mice, the HF gene expression profile appears to already be set in the early stages of HF progression. Thus we have identified methodologies that have the potential to be used for predictive genomic profiling of cardiac phenotype, including cardiovascular disease.

genomics; microarrays


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 DISCLOSURES
 References
 
SEVERAL GENETIC MOUSE MODELS of heart failure (HF) have recently been described involving mutation, ablation, or cardiac overexpression of different genes. In particular, ablation of the muscle LIM protein gene (MLP-/-), a muscle-restricted cytoarchitectural protein (2), or myocardial-targeted transgenic overexpression of calsequestrin (CSQ), a high-capacity sarcoplasmic reticulum Ca2+ binding protein (6, 15), results in genetic mouse models of HF. Both models develop cardiomyopathy and significant left ventricular (LV) dysfunction, although the CSQ model is a much more aggressive phenotype that results in premature death by 9–16 wk of age (6, 15). The MLP-/- mice develop dilated cardiomyopathy between 3–6 mo of age (2). Both mouse models of HF also exhibit several molecular manifestations present in the failing human heart, including numerous biochemical abnormalities in the ß-adrenergic receptor (ß-AR) system, such as functional receptor uncoupling, enhanced expression and activity of the ß-AR kinase (ßARK1), and loss of ß-AR inotropic reserve (25).

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{gamma}-subunits of heterotrimeric G proteins (Gß{gamma}) (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
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 DISCLOSURES
 References
 

Tissue sampling and RNA extraction.
LV myocardial tissue was dissected out of hearts freshly harvested from 53 individual mice, weighed, and immediately frozen in liquid nitrogen, then stored at -80°C. All animal procedures and tissue harvesting were performed in concordance with NIH guidelines and approved Duke University animal procedures.

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.


View this table:
[in this window]
[in a new window]
 
Table 1. Top 25 genes significantly different between Normal and HF

 

View this table:
[in this window]
[in a new window]
 
Table 2. Top 25 genes significantly different between Normal, HF, Rescue, and Tg

 

View this table:
[in this window]
[in a new window]
 
Table 3. The 10 genes found to be significantly different in the progression from early to advanced-stage HF by normalized t-test

 
Bayesian regression statistics.
Predictive Bayesian regression modeling using permutation-tested t-statistic determined priors for classifying the mouse models based on gene expression profiles can be rigorously evaluated by cross validation. We use the regularized linear regression model as described (28, 33). To avoid bias toward only the highest expressed genes (linear models) or lowest expressed genes (logarithms), we normalized the data using the variance stabilization technique(13). Normalization is followed by a predictive binary classification (i.e., 0 = Normal, 1 = HF), based on a Bayesian regression model designed for binary class prediction from gene expression data, initialized using the 100 genes with the highest t-score from a permutation-tested t-statistic. Importantly, the predictive power of this analysis was evaluated by "leave-one-out" cross-validation, where the subset of 100 predictive genes are reselected at every round of cross-validation. To analyze the 11 samples from the investigation of the progression of HF, data from the 12 Normal and 12 HF samples were used as a training data set, and the novel 11 samples of mild and end-stage HF were added blindly to the analysis, then independently cross-validated (i.e., full cross-validation, not leave-one-out cross-validation), providing a clear prognostic measure of the probability of belonging to a particular group (i.e., Normal or HF) (for further description of the Bayesian regression model, see Ref. 33 and associated experimental procedures available online at http://www.pnas.org). The same normalization was used for identifying a list of differentially expressed genes with low false discovery rate. Based on the t-score, which tends to favor genes with relatively high fold change between groups but low variance within each group, the method of Tusher et al. (30) was used to identify a list of potentially induced genes and calculate the false discovery rate (i.e., the expected percentage of unchanged genes in this list). Note, that this method considers the list of genes as a single entity, making no claims regarding the significance of individual genes within the list.

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
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 DISCLOSURES
 References
 
Our current study was performed based on the following phenotypic groups of mice that have previously been extensively characterized (2, 6, 11, 15, 16, 24, 25): 1) 12 Normal mice, comprising 6 adult wild types of MLP-/- background (6–8 mo old) and 6 adult nontransgenic littermate controls (NLCs) of CSQ (3–4 mo old); 2) 12 HF mice, comprising 6 adult MLP-/- (6–8 mo old) and 6 adult CSQ (3–4 mo old); 3) 12 Rescue mice, comprising 6 adult MLP-/-/ßARKct (6–8 mo old) and 6 adult CSQ/ßARKct (3–4 mo old); and 4) 6 Tg mice, comprising all adult ßARKct transgenic littermates of CSQ/ßARKct (3–4 mo old). First, we found no clear distinction in LV gene expression between the wild-type and NLC mice of the MLP-/- and CSQ lines, respectively, by multiple statistical methods (see MATERIALS AND METHODS and Ref. 33; also, see Supplemental Fig. 1 and Supplemental Table 1, available at the Physiological Genomics web site).1 Thus we determined that it was feasible and reasonable to combine mice from slightly different backgrounds into phenotypic groups for our subsequent analyses as delineated above.

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, AC; also, see Supplemental Fig. 4).



View larger version (17K):
[in this window]
[in a new window]
 
Fig. 1. Distinction and prediction of Normal and HF by gene expression. A: multidimensional scaling (MDS; visualization of high-dimensional data in low-dimensional space, see MATERIALS AND METHODS) of the entire gene expression data set for Normal (green) and HF (red). B: hierarchical clustering of Normal and HF, based on the entire gene expression data set. Normal 1–6 are NLC(MLP), Normal 7–12 are NLC(CSQ), HF 1–6 are MLP-/-, and HF 7–12 are CSQ. C: out-of-sample cross-validated prediction of Normal (dark blue, left 12) and HF (light blue, right 12). Values on the vertical axis are estimated classification probabilities with posterior means and corresponding 90% classification probability confidence intervals marked as dashed lines. If predicted accurately, then Normal should equal 0, and HF should equal 1, with 0.5 being the misclassification threshold. Prediction is established by posterior distributions of the probability that the left-out sample fall in the Normal class. The analysis and prediction for each sample is initialized based on the screened subset of 100 discriminatory genes, which are then reselected at every round of cross-validation (see MATERIALS AND METHODS). NLC, nontransgenic littermate control; MLP, muscle LIM protein; CSQ, calsequestrin.

 
Specific genes associated with cardiomyopathy were determined by a corrected, unpaired t-test comparing the Normal and HF groups, which included a number of novel genes not previously associated with HF. Importantly, this list included several known marker genes of HF from several functional categories, such as atrial natriuretic factor (ANF), brain natriuretic peptide (BNP), ß-myosin heavy chain (ß-MHC), skeletal {alpha}-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).



View larger version (29K):
[in this window]
[in a new window]
 
Fig. 2. Cardiac function of early stage and advanced stage HF mice. Percent fractional shortening (%FS) for normal (wild type or NLC E&A), early stage (E), and advanced stage (A) HF mice. Early stage MLP-/- or wild-type mice were 3 mo old, and advanced stage were 6 mo old. Early stage CSQ or NLC mice were 7 wk old, and advanced stage mice were 14 wk (3 mo) old.

 


View larger version (17K):
[in this window]
[in a new window]
 
Fig. 3. Classification and prediction of both early and end-stage HF as end-stage HF phenotype by gene expression. A: MDS image demonstrating the prior Normal (green) and HF (red) distinction, along with the novel HF classification of both the early stage HF MLP-/- (purple) and CSQ (light blue), as well as the end-stage HF MLP-/- (dark blue) and CSQ (yellow). B: full cross-validation (i.e., blind) prediction of both early stage and advanced stage HF as end-stage HF phenotype based on the Normal vs. HF data set; a, end-stage HF; Y, early stage HF; Normal, dark blue; HF, light blue; early stage HF, red and yellow; end-stage HF, magenta and rust.

 
The above data validate our predictive model and suggest, using these mouse models of HF, that the gene expression profile of HF is set at a very early stage in the pathogenesis of this disease. As anticipated from this predictive analysis, a majority of the top 20 genes significantly associated (i.e., estimated 0% false discovery rate) with both early and advanced stage HF were shared in common with those previously determined in the analysis of Normal vs. HF (Supplemental Table 5). There was no correlation of chromosomal location associated with any of the lists of differential gene expression described herein (data not shown).

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 G{alpha}o, 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).



View larger version (41K):
[in this window]
[in a new window]
 
Fig. 4. Correlation between Affymetrix and Northern blot determination of gene expression. Representative comparisons of gene expression determined by either Northern blot or Affymetrix GeneChip were performed using 10 µg total RNA aliquoted from the same original total RNA isolation procedure. A: Northern blot and raw Affymetrix hybridization intensity values for atrial natriuretic factor (ANF). B: Northern blot and raw Affymetrix hybridization intensity values for brain natriuretic peptide (BNP).

 


View larger version (12K):
[in this window]
[in a new window]
 
Fig. 5. Correlation between RNA (Affymetrix) and protein expression. Immunoblot for the G protein G{alpha}o demonstrating correlation of protein expression in three Normal and three HF samples with RNA expression data obtained by Affymetrix. Upregulation of G{alpha}o in HF was one of the top 20 overall predictors in our study.

 

    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 DISCLOSURES
 References
 
The current study provides substantial insight into the development, progression, and rescue of murine HF at the level of differential myocardial gene expression. We have described and validated methods that can blindly predict HF phenotype at both early and advanced stages based on gene expression data, which suggests that the gene expression profile of HF is set very early in the disease progression. Our study provides several salient findings, including: 1) investigation of the development, progression, and rescue of murine HF using myocardial gene expression profiling from a substantial and statistically testable sample size; 2) use of and concordance between multiple statistical methods; 3) development and validation of a blindly predictive model for HF phenotype; 4) confident identification of genes significantly associated with and predictive of the phenotypes studied herein; and 5) early stage HF gene expression is highly similar to and predictive of advanced stage HF.

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-{alpha}-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 {alpha}-actin and cardiac {alpha}-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
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 DISCLOSURES
 References
 
This work was supported by National Heart, Lung, and Blood Institute Grants R01-HL-61690 (to W. J. Koch) and HL-65184 (to H. A. Rockman) and by an American Heart Association (Mid-Atlantic Affiliate) Postdoctoral Fellowship (to B. C. Blaxall).


    ACKNOWLEDGMENTS
 
We thank Dr. Robert J. Lefkowitz for helpful insights and discussion, Dr. Holly Dressman for astute help with Affymetrix array processing, and Dr. Lan Mao for invaluable expert assistance with murine cardiovascular physiology.

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
 
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).

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. Back


    References
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 DISCLOSURES
 References
 

  1. Akhter SA, Milano CA, Shotwell KF, Cho MC, Rockman HA, Lefkowitz RJ, and Koch WJ. Transgenic mice with cardiac overexpression of {alpha}1B-adrenergic receptors. In vivo {alpha}1-adrenergic receptor-mediated regulation of ß-adrenergic signaling. J Biol Chem 272: 21253–21259, 1997.[Abstract/Free Full Text]
  2. Arber S, Hunter JJ, Ross J Jr, Hongo M, Sansig G, Borg J, Perriard JC, Chien KR, and Caroni P. MLP-deficient mice exhibit a disruption of cardiac cytoarchitectural organization, dilated cardiomyopathy, and heart failure. Cell 88: 393–403, 1997.[ISI][Medline]
  3. Aronow BJ, Toyokawa T, Canning A, Haghighi K, Delling U, Kranias E, Molkentin JD, and Dorn GW II. Divergent transcriptional responses to independent genetic causes of cardiac hypertrophy. Physiol Genomics 6: 19–28, 2001.[Abstract/Free Full Text]
  4. Barrans JD, Allen PD, Stamatiou D, Dzau VJ, and Liew CC. Global gene expression profiling of end-stage dilated cardiomyopathy using a human cardiovascular-based cDNA microarray. Am J Pathol 160: 2035–2043, 2002.[Abstract/Free Full Text]
  5. Blaxall BC, Tschannen-Moran BM, Milano CA, and Koch WJ. Differential gene expression and genomic patient stratification following left ventricular assist device support. J Am Coll Cardiol 41: 1096–1106, 2003.[ISI][Medline]
  6. Cho MC, Rapacciuolo A, Koch WJ, Kobayashi Y, Jones LR, and Rockman HA. Defective ß-adrenergic receptor signaling precedes the development of dilated cardiomyopathy in transgenic mice with calsequestrin overexpression. J Biol Chem 274: 22251–22256, 1999.[Abstract/Free Full Text]
  7. Chu PH, Ruiz-Lozano P, Zhou Q, Cai C, and Chen J. Expression patterns of FHL/SLIM family members suggest important functional roles in skeletal muscle and cardiovascular system. Mech Dev 95: 259–265, 2000.[ISI][Medline]
  8. Cook SA, Matsui T, Li L, and Rosenzweig A. Transcriptional effects of chronic Akt activation in the heart. J Biol Chem 277: 22528–22533, 2002.[Abstract/Free Full Text]
  9. Cox TF and Cox MAA. Multidimensional Scaling. New York: Chapman and Hall, 1994.
  10. Friddle CJ, Koga T, Rubin EM, and Bristow J. Expression profiling reveals distinct sets of genes altered during induction and regression of cardiac hypertrophy. Proc Natl Acad Sci USA 97: 6745–6750, 2000.[Abstract/Free Full Text]
  11. Harding VB, Jones LR, Lefkowitz RJ, Koch WJ, and Rockman HA. Cardiac ßARK1 inhibition prolongs survival and augments beta blocker therapy in a mouse model of severe heart failure. Proc Natl Acad Sci USA 98: 5809–5814, 2001.[Abstract/Free Full Text]
  12. Hoshijima M and Chien KR. Mixed signals in heart failure: cancer rules. J Clin Invest 109: 849–855, 2002.[Free Full Text]
  13. Huber W, Von Heydebreck A, Sultmann H, Poustka A, and Vingron M. Variance stabilization applied to microarray data calibration and to the quantification of differential expression. Bioinformatics 18, Suppl 1: S96–S104, 2002.[Abstract/Free Full Text]
  14. Hwang JJ, Allen PD, Tseng GC, Lam CW, Fananapazir L, Dzau VJ, and Liew CC. Microarray gene expression profiles in dilated and hypertrophic cardiomyopathic end-stage heart failure. Physiol Genomics 10: 31–44, 2002. First published April 30, 2002; 10.1152/physiolgenomics.00122.2001.
  15. Jones LR, Suzuki YJ, Wang W, Kobayashi YM, Ramesh V, Franzini-Armstrong C, Cleemann L, and Morad M. Regulation of Ca2+ signaling in transgenic mouse cardiac myocytes overexpressing calsequestrin. J Clin Invest 101: 1385–1393, 1998.[Abstract/Free Full Text]
  16. Koch WJ, Lefkowitz RJ, and Rockman HA. Functional consequences of altering myocardial adrenergic receptor signaling. Annu Rev Physiol 62: 237–260, 2000.[ISI][Medline]
  17. Koch WJ, Rockman HA, Samama P, Hamilton RA, Bond RA, Milano CA, and Lefkowitz RJ. Cardiac function in mice overexpressing the ß-adrenergic receptor kinase or a ßARK inhibitor. Science 268: 1350–1353, 1995.[ISI][Medline]
  18. Lee SB, Huang K, Palmer R, Truong VB, Herzlinger D, Kolquist KA, Wong J, Paulding C, Yoon SK, Gerald W, Oliner JD, and Haber DA. The Wilms tumor suppressor WT1 encodes a transcriptional activator of amphiregulin. Cell 98: 663–673, 1999.[ISI][Medline]
  19. Lelivelt MJ and Culbertson MR. Yeast Upf proteins required for RNA surveillance affect global expression of the yeast transcriptome. Mol Cell Biol 19: 6710–6719, 1999.[Abstract/Free Full Text]
  20. Lim DS, Roberts R, and Marian AJ. Expression profiling of cardiac genes in human hypertrophic cardiomyopathy: insight into the pathogenesis of phenotypes. J Am Coll Cardiol 38: 1175–1180, 2001.[ISI][Medline]
  21. Lowes BD, Gilbert EM, Abraham WT, Minobe WA, Larrabee P, Ferguson D, Wolfel EE, Lindenfeld J, Tsvetkova T, Robertson AD, Quaife RA, and Bristow MR. Myocardial gene expression in dilated cardiomyopathy treated with beta-blocking agents. N Engl J Med 346: 1357–1365, 2002.[Abstract/Free Full Text]
  22. Minamisawa S, Hoshijima M, Chu G, Ward CA, Frank K, Gu Y, Martone ME, Wang Y, Ross J Jr, Kranias EG, Giles WR, and Chien KR. Chronic phospholamban-sarcoplasmic reticulum calcium ATPase interaction is the critical calcium cycling defect in dilated cardiomyopathy. Cell 99: 313–322, 1999.[ISI][Medline]
  23. Redfern CH, Degtyarev MY, Kwa AT, Salomonis N, Cotte N, Nanevicz T, Fidelman N, Desai K, Vranizan K, Lee EK, Coward P, Shah N, Warrington JA, Fishman GI, Bernstein D, Baker AJ, and Conklin BR. Conditional expression of a Gi-coupled receptor causes ventricular conduction delay and a lethal cardiomyopathy. Proc Natl Acad Sci USA 97: 4826–4831, 2000.[Abstract/Free Full Text]
  24. Rockman HA, Chien KR, Choi DJ, Iaccarino G, Hunter JJ, Ross J Jr, Lefkowitz RJ, and Koch WJ. Expression of a ß-adrenergic receptor kinase 1 inhibitor prevents the development of myocardial failure in gene-targeted mice. Proc Natl Acad Sci USA 95: 7000–7005, 1998.[Abstract/Free Full Text]
  25. Rockman HA, Koch WJ, and Lefkowitz RJ. Seven-transmembrane-spanning receptors and heart function. Nature 415: 206–212, 2002.[ISI][Medline]
  26. Sato Y, Kiriazis H, Yatani A, Schmidt AG, Hahn H, Ferguson DG, Sako H, Mitarai S, Honda R, Mesnard-Rouiller L, Frank KF, Beyermann B, Wu G, Fujimori K, Dorn GW II, and Kranias EG. Rescue of contractile parameters and myocyte hypertrophy in calsequestrin overexpressing myocardium by phospholamban ablation. J Biol Chem 276: 9392–9399, 2001.[Abstract/Free Full Text]
  27. Soukas A, Cohen P, Socci ND, and Friedman JM. Leptin-specific patterns of gene expression in white adipose tissue. Genes Dev 14: 963–980, 2000.[Abstract/Free Full Text]
  28. Spang R, Zuzan H, West M, Nevins J, Blanchette C, and Marks JR. Prediction and uncertainty in the analysis of gene expression profiles. In Silico Biol 2: 369–381, 2002.[Medline]
  29. Tan FL, Moravec CS, Li J, Apperson-Hansen C, McCarthy PM, Young JB, and Bond M. The gene expression fingerprint of human heart failure. Proc Natl Acad Sci USA 99: 11387–11392, 2002.[Abstract/Free Full Text]
  30. Tusher VG, Tibshirani R, and Chu G. Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA 98: 5116–5121, 2001.[Abstract/Free Full Text]
  31. Valenzuela D, Han X, Mende U, Fankhauser C, Mashimo H, Huang P, Pfeffer J, Neer EJ, and Fishman MC. G{alpha}o is necessary for muscarinic regulation of Ca2+ channels in mouse heart. Proc Natl Acad Sci USA 94: 1727–1732, 1997.[Abstract/Free Full Text]
  32. Webb GC, Akbar MS, Zhao C, and Steiner DF. Expression profiling of pancreatic ß-cells: glucose regulation of secretory and metabolic pathway genes. Proc Natl Acad Sci USA 97: 5773–5778, 2000.[Abstract/Free Full Text]
  33. West M, Blanchette C, Dressman H, Huang E, Ishida S, Spang R, Zuzan H, Olson JA Jr, Marks JR, and Nevins JR. Predicting the clinical status of human breast cancer by using gene expression profiles. Proc Natl Acad Sci USA 98: 11462–11467, 2001.[Abstract/Free Full Text]