Skip Navigation

European Journal of Heart Failure 2006 8(4):381-389; doi:10.1016/j.ejheart.2006.05.004
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (8)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Beisvag, V.
Right arrow Articles by Ellingsen, O.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Beisvag, V.
Right arrow Articles by Ellingsen, O.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

© 2006 European Society of Cardiology

Aetiology-specific patterns in end-stage heart failure patients identified by functional annotation and classification of microarray data

Vidar Beisvaga,*, Per Kristian Lehreb, Herman Midelfartb, Halfdan Aassc, Odd Geirand, Arne Kristian Sandvike,f, Astrid Lægreide, Jan Komorowskib,g and Øyvind Ellingsena,h

a Department of Circulation and Medical Imaging, Norwegian University of Science and Technology Trondheim, Norway
b Department of Information Technology and Computer Science, Norwegian University of Science and Technology Trondheim, Norway
c Department of Cardiology Rikshospitalet, Oslo, Norway
d Department of Thorax Surgery Rikshospitalet, Oslo, Norway
e Department of Cancer Research and Molecular Medicine, Norwegian University of Science and Technology Trondheim, Norway
f Department of Medicine, St. Olav's Hospital Trondheim, Norway
g The Linnaeus Centre for Bioinformatics, Uppsala University Uppsala, Sweden
h Department of Cardiology, St. Olav's Hospital Trondheim, Norway

* Corresponding author. Department of Circulation and Imaging, Faculty of Medicine, NTNU, Medical Research Centre, N-7489 Trondheim, Norway. Tel.: +47 7359 8888. E-mail address: vidar.beisvag{at}ntnu.no


   Abstract
Top
 Abstract
1. Introduction
 2. Materials and methods
 3. Results
 4. Discussion
 5. Conclusion
 Appendix A. Supplementary data
 References
 
Background: The objective of the present study was to use gene expression profiling, functional annotations and classification to identify aetiology-specific biological processes and potential molecular markers for different aetiologies of end-stage heart failure.

Methods and results: Individual left ventricular myocardial samples from eleven coronary artery disease and nine dilated cardiomyopathy transplant patients were co-hybridized with pooled RNA from four non-failing hearts on custom-made arrays of 7000 human genes. Significance analysis identified differential expression of 153 and 147 genes, respectively, in coronary artery disease or dilated cardiomyopathy versus non-failing hearts. Analysis of Gene Ontology biological process annotations indicated aetiology-specific patterns, primarily related to genes involved in catabolism and in regulation of protein kinase activity. Gene expression classifiers were obtained and used for class prediction of random samples of coronary artery diseased and dilated cardiomyopathic hearts. Best classifiers frequently included matrix metalloproteinase 3, fibulin 1, ATP-binding cassette, sub-family B member 1 and iroquois homeobox protein 5.

Conclusion: Combining functional annotation from microarray data and classification analysis constitutes a potent strategy to identify disease-specific biological processes and gene expression markers in e.g. end-stage coronary artery disease and dilated cardiomyopathy.

Key Words: Coronary artery disease • Cardiomyopathy • Gene expression • Functional annotation • Classification

Received December 6, 2005; Revised March 7, 2006; Accepted May 9, 2006


   1. Introduction
Top
 Abstract
 1. Introduction
2. Materials and methods
 3. Results
 4. Discussion
 5. Conclusion
 Appendix A. Supplementary data
 References
 
Heart failure is a multifactorial disease and expression of multiple gene clusters that is altered. Microarray analysis has been used extensively to assess global gene expression of coronary artery disease (CAD) and dilated cardiomyopathy (DCM) in humans [1-4], and in rodent models of heart failure [5-7]. These and other studies show changes in expression levels of genes encoding cytoskeletal and extracellular matrix (ECM) proteins, as well as proteins involved in apoptosis and intracellular signalling, energy and lipid metabolism, cell communication and calcium-handling pathways. These results have extended our knowledge of gene regulation in heart failure. However, few attempts have been made to characterize differences in genome-wide gene expression patterns between CAD and DCM hearts.

A frequently used method to analyze microarray data is class discovery (unsupervised learning), which clusters samples and genes with similar patterns of expression. The most commonly used algorithm is based on hierarchical clustering. Such analyses have been useful in revealing potentially meaningful relationships among genes and among samples, but cannot explain the underlying biological mechanisms [8]. Recently, some attempts have been made to characterize small groups of patients with end-stage heart failure based on hierarchical clustering of gene expression profiles [2,9,10]. An alternative approach is to use machine learning and class prediction (supervised learning). Supervised methods use gene expression data linked to patient information like disease-state, aetiology etc., as training examples to build predictive models that describe relationships between gene expression and aetiology. So far, these methods have mostly been used in cancer research [11-14].

The present study identifies genes that are differentially expressed in end-stage CAD and DCM hearts, and aetiology-specific features of differentially expressed genes by annotation to biological processes according to Gene Ontology (GO). A specific aim was to use class prediction and machine learning procedures to generate classifiers based on gene expression to distinguish between end-stage heart failure of different aetiologies. Classifier genes identified with high predictive value, are likely to be molecular markers of heart disease and point to important biological processes that distinguish between CAD and DCM.


   2. Materials and methods
Top
 Abstract
 1. Introduction
 2. Materials and methods
3. Results
 4. Discussion
 5. Conclusion
 Appendix A. Supplementary data
 References
 
2.1. Patient characteristics
Myocardial samples were obtained from eleven CAD and nine DCM patients undergoing heart transplantation because of end-stage heart failure, and from four normal hearts. All CAD and DCM patients had end-stage heart failure, and comparable clinical and haemodynamic characteristics (Table 1). The investigation conforms to the principles outlined in the Declaration of Helsinki.


View this table:
[in this window]
[in a new window]

  		Table 1 Clinical and haemodynamic characteristics

 
2.2. Heart tissue and RNA isolation
Samples were taken from the left ventricle free wall immediately after explantation, snap frozen in liquid nitrogen and stored at –80 °C. Frozen tissue was homogenized using an Ultra-Turrax rotating knife homogenizer and total RNA extracted using a TRIzole reagent (GIBCO BRL Life Technologies, New York, NY, USA). Quantitation, purity and RNA integrity were evaluated by absorbance at 260 and 280 nm (Gene Quant, AmershamPharmacia, Buckinghamshire, UK), and agarose gel electrophoresis. High quality RNA with A260/A280 ratio above 1.8 and intact ribosomal 28S and 18S RNA bands was used for microarray.

2.3. Microarrays, labelling, hybridization and scanning
RNA from the normal hearts was pooled, labelled and hybridized against labelled RNA from each patient, resulting in a total of 20 hybridizations. Moreover, RNA from each of the individual four normal hearts was co-hybridized with pooled RNA from the same hearts to assess heterogeneity among the samples.

We used cDNA arrays from the Norwegian Microarray Consortium. Approximately 7.000 spots were printed on the arrays. Among these, 6524 represent genes from the IMAGE consortium human cDNA clone set (Research Genetics, Huntsville, AL, USA). According to UniGene build 180, 6089 clones were associated with a UniGene cluster, which gives a name and possible annotations for that gene. The remaining 476 spots on the array represented controls (e.g. Cot-1 DNA and Salmon Sperm DNA) and some local clones of interest.

Local resequencing of a major portion of the PCR products (IMAGE clones) printed on our microarray showed that at least 80% of the clones were correct. The remaining 20% of the clones presented different problems. Some of the sequencing reactions did not give satisfactory results, possibly due to the quality of the PCR products or sequencing reactions. Other clones gave good sequences but could not be referred to an IMAGE clone, either because they showed only vector sequences or had no similarity with any GenBank sequence. Thus, from this resequencing we concluded that approximately 80% of the sequences are indeed correct and possibly also some of the remaining 20%, but it is impossible to state this with absolute certainty. This is a worldwide and well-known problem with the IMAGE clone sets. We did not exclude any of the genes in our analysis, since we chose to focus on groups of genes and not specific genes. However, we have ensured that all the genes represented in the best classifier had correct sequences.

Five µg total RNA was reverse transcribed and labelled with Cy3 (CAD and DCM hearts) or Cy5 (pooled normal hearts) attached dendrimers, respectively, using the Genisphere 3DNA kit as described in the manufacturer's one step protocol (Genisphere, Montvale, NJ, USA). Hybridization was performed at 60 °C for 15 h using 0.25 M NaPO4, 1 mM EDTA, 4.5% SDS, 1X SSC, 1 µl enhancer reagent (Genisphere) and 0.5 µg human Cot-1 DNA (GIBCO BRL, Life Technologies, Carlsbad, CL, USA) in a total volume of 35 µl. Three 15 min post-hybridization washes were performed; first at 55 °C with 2X SSC and 0.2% SDS, then at room temperature with 2X SSC, and finally at room temperature with 0.2X SSC. Scanning was done with fixed PMT gain (80%) and laser power adjusted to balance Cy3 and Cy5 signal intensities. Images were analyzed using an ArraySuite version 2.0 extension of the IPLab image processing software package (Scanalytics, Fairfax, VA, USA).

2.4. Database submission
The microarray data were prepared according to the "minimum information about microarray experiment" (MIAME) recommendations, and deposited in the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) with accession number GSE2014 [NCBI GEO] .

2.5. Data processing, significance analysis and classifier assignment
The raw data processing corrects for systematic errors using filtering and normalization, and reduces the number of genes considered based on significance analysis. Procedures are similar to those used by Nørsett et al. [12]. Measurements from probe duplicates were averaged. Spots with a signal to background ratio <2.0 in either of the channels were excluded. Remaining spots were normalized by intensity-dependent normalization (lowess). Differentially expressed genes were identified by t-test of the log-ratio differences between classes, using bootstrapping. Complete lists of differentially expressed genes (p≤0.05) are reported as supplementary data (see Supplementary Tables 1 and 2 in electronic version, Appendix A).

Classification was performed with discretisized data, using a machine learning approach from the rough set theory based Rosetta tool kit [15], which implements and tests a wide range of machine learning algorithms. The supervised learning algorithms were used to generate multiple sets of rules to determine assignment of a sample to CAD or DCM. Each rule is in the form of "IF gene X is expressed at a higher/lower level than in normal hearts, THEN the sample belongs to the group CAD/DCM". A set of rules is referred to as a classifier. Several classifiers were obtained and the classification performance assessed by the AUC-measure and validated by leave-one-out cross-validation.

The classification procedure has been described in detail by Midelfart et al. [16]. Briefly, data analysis and classification were repeated using genes with p-values in the range 0.001-0.01 and a number of genes in the range 5-40, and different discrimination and classification algorithms. As an extra control, a randomised data set was simulated for validation of data analysis and classification.

2.6. Functional clustering according to gene ontology annotations
To obtain information about gene functions, we used the eGOn (explore geneontology) web tool (http://www.genetools.no). Lists of differentially expressed genes were submitted into eGOn, which automatically associates Gene Ontology (GO) terms from public databases to the submitted gene reporters. Statistical comparison of the number of differentially expressed genes representing specific GO nodes from DCM and CAD hearts was performed by eGOn. For analysis within a gene list (Master-Target test) and between gene lists (Target-Target test), eGOn uses the Fisher's exact test.

2.7. Validation of microarray results by real-time PCR
Expression of four genes represented in the best classifier (fibulin 1 (FBLN1), ATP-binding cassette, subfamily B, member 1 (ABCB1), matrix metalloproteinase (MPP3) and iroquois homeobox protein 5 (IRX5)), as well as protein kinase C, nu (PRKCN) and beta actin (ACTB), were quantified by RT-PCR using TaqMan probes. Five µg total RNA samples from four hearts from each group (DCM, CAD and Normal) were reverse transcribed using the Eurogenetec reverse transcriptase core kit (Eurogentec SA, Seraing, Belgum) according to the manufacturer's instructions. RT-PCR was done in triplicate with cDNA corresponding to 0.0625 µg total RNA using the Qiagen QuantiTect kit (Qiagen GmbH, Hilden, Germany), according to manufacturer's protocol. Relative quantification was done using the modified {Delta}{Delta}CT equation [17], normalizing the RT-PCR expression ratios for each individual sample by Beta-actin measurements.


   3. Results
Top
 Abstract
 1. Introduction
 2. Materials and methods
 3. Results
4. Discussion
 5. Conclusion
 Appendix A. Supplementary data
 References
 
3.1. Differential gene expression
Compared to normal hearts, we found 153 and 147 genes to be differentially expressed (p≤0.01) in CAD and DCM, respectively. At a significance level of p≤0.05, the corresponding numbers were 307 and 338. The differentially expressed genes at p≤0.001 in either CAD or DCM are shown in Table 2 (see electronic version, Appendix A). Complete lists of differentially expressed genes (p≤0.05) are given as supplementary material (see Supplementary Tables 1 and 2, in electronic version, Appendix A). The variability in the left ventricular gene expression between normal individuals was very low, as no genes were differentially expressed at p<0.01 and only 35 genes at p≤0.05.

3.2. Functional clustering according to the gene ontology (GO) annotations
Genes differentially expressed between normal and diseased hearts, are likely to reflect molecular mechanisms involved in heart failure. We therefore looked for functional clusters among genes differentially expressed (p≤0.05) in either CAD or DCM. Annotations of biological processes according to Gene Ontology (GO) were obtained from the NMC Annotation database (www.genetools.no), which was based on UniGene build nr. 180, LocusLink build nr. 147 and GOA January 2005 at the time of analysis. Of the 307 and 338 differentially expressed genes in CAD or DCM, 256 and 284, respectively, were associated with a UniGene cluster. These genes could be annotated to 647 and 739 GO processes, respectively, and the distribution of annotations across the GO biological processes was investigated using eGOn.

The majority of the differentially expressed genes were related to six main biological process categories: cell communication, signal transduction, metabolism, transcription, response to stimulus and development (Table 3A). This tells which biological processes are associated with most of the differentially expressed genes, but not whether these genes contribute as "enrichment" of genes for each process as compared with the number of genes associated with the same processes on the whole array. We used the Master-Target test in eGOn to assess the relative numbers of GO biological process annotations linked to either CAD or DCM differentially expressed genes, compared to the relative numbers of the same GO biological process annotations linked to all the genes on the microarray (6524 unique probes were associated with 6089 unique genes of which 4148 were linked to one or more GO process annotations). Among the differentially expressed genes we found an over-representation involved in cell communication, response to stress, response to external biotic stimulus, defense response, immune response and inflammatory response (Table 3B).


View this table:
[in this window]
[in a new window]

  		Table 3A Functional annotation of differentially (p≤0.05) expressed genes from the CAD and DCM samples, and selected results from the eGOn test, which shows the number of differentially expressed genes associated with the specific GO terms and p-values calculated

 


View this table:
[in this window]
[in a new window]

  		Table 3B GO categories over-represented among the differentially expressed genes in both CAD and DCM hearts

 
The functional clustering of biological processes indicated that most of the processes represented by many genes were common for both aetiologies. However, some biological processes were over-represented in either CAD or in DCM (Table 3C). To find GO processes with different representations of differentially expressed genes we compared lists from CAD and DCM using the Target-Target test in eGOn. Significantly more genes involved in catabolism were differentially expressed in CAD compared to DCM (Table 3D). Moreover, a separate comparison of up-regulated and down-regulated genes was performed. A significantly higher number of genes involved in catabolism was up-regulated in CAD hearts compared to DCM hearts (13 versus 4 genes, p=0.044) (Table 4, see electronic version, Appendix A).


View this table:
[in this window]
[in a new window]

  		Table 3C GO categories over-represented in either CAD or DCM samples

 


View this table:
[in this window]
[in a new window]

  		Table 3D GO categories with differential representation of gene reporters between the CAD and DCM samples

 
3.3. Assessment of expression-based classifiers
The classification serves to select genes which separate CAD from DCM, assuming that such genes also point to important pathophysiological differences between the two diseases. A simulated control experiment suggested that classification of random data was possible using genes with p>0.005. Most confident discrimination was obtained when the classifier included up to seven genes with the highest t-statistics (p<0.001). Table 5 shows eight of the best classifiers with AUC values≥0.75. The best classifier based on AUC measure (Classifier 1, Table 5), correctly predicted group assignment in 16 of 20 samples. The genes employed by this classifier are shown in Table 6. To assess the validity of the genes in the classifiers, we determined how often each gene was included in a classifier using leave one out cross validation (Table 6). Of seven genes in the best classifier, four genes (MMP3, FBLN1, ATBC1 and IRX5) appeared in a high number of classifiers generated during validations.


View this table:
[in this window]
[in a new window]

  		Table 5 Eight of the best classifiers generated

 


View this table:
[in this window]
[in a new window]

  		Table 6 The genes associated with the best obtained classifier (classifier 1)

 
3.4. Real-time PCR validation
To validate the results obtained by microarray analysis, RT-PCR was performed on five selected genes. These included the four genes most often used in the best classifiers and protein kinase C, nu (PRKCN), which is known as an important signalling component promoting cardiac hypertrophy. RT-PCR measurements corresponded well with the microarray results (Fig. 1, see electronic version, Appendix A). As observed by us and others, microarray ratios tend to be compressed compared to real-time PCR results [18].


   4. Discussion
Top
 Abstract
 1. Introduction
 2. Materials and methods
 3. Results
 4. Discussion
5. Conclusion
 Appendix A. Supplementary data
 References
 
Previous gene profiling studies suggest that a number of primary and secondary mechanisms are involved in heart failure, including energy pathways, protein metabolism, muscle contraction, apoptosis, and cell communication processes like signal transduction [1-5,19]. The present study used differentially expressed genes combined with functional annotation (GO) and classification to identify biological processes that distinguish end-stage CAD from DCM.

Until now, only one study has used gene profiling to compare gene expression patterns in end-stage heart failure caused by either CAD or DCM. Steenman et al. [9] successfully used gene expression data and hierarchical clustering to explore patterns of gene expression related to heart failure patients, but were unable to identify specific clusters for different aetiologies. In contrast, our classifiers selected a small number of genes whose expression levels distinguish between CAD and DCM, thus demonstrating the feasibility of using microarray data and class prediction for studying heart failure pathobiology.

4.1. Functional annotation analysis
Gene Ontology (GO) analysis performed with the eGOn tool showed that the differentially expressed genes in both CAD and DCM were related to six main biological process categories (cell communication, signal transduction, metabolism, transcription, response to stimulus and development) with no significant differences in process annotations between CAD and DCM. This indicates that some main biological processes in end-stage CAD and DCM are similar. Analyzing the differentially expressed genes in relation to the total number of genes explored on our microarrays, we identified an over-representation of genes involved in cell communication, response to stress, response to external biotic stimulus, defense response, immune response and inflammatory response (Table 3B). This indicates that both diseases were characterized by an increase of stress genes. A main objective was to identify aetiology-specific gene expression features in terms of biological processes. Even if most processes were common for the two aetiologies, some were over-represented in either CAD or in DCM (Table 3C). Significantly more genes involved in catabolism were differentially expressed in CAD compared to DCM (Table 4, see electronic version, Appendix A), which is consistent with catabolic/anabolic imbalance in heart failure [20]. Interestingly, we found more genes involved in catabolism were up-regulated in CAD hearts compared to DCM hearts (13 versus 4 genes, Table 4, see electronic version, Appendix A). Among 12 genes exclusively up-regulated in CAD, we found a sub-cluster of four genes involved in lipid catabolism, apolipoprotein C-II, (APOC2), lipoprotein lipase (LPL), phospholipase C, beta 1 (PLCB1) and phospholipase C, gamma 2 (PLCG2). In contrast, lipase, member H (LIPH) and phospholipase C, delta 1 (PLCD1) were down-regulated only in DCM. This indicates that lipid catabolism and lipid signalling is stimulated in end-stage CAD but not in DCM. During development of hypertrophy, phospholipase C (PLC) is increased, and in the failing heart PLC isozymes are decreased [21]. Normal, exercise-induced cardiac growth is mainly regulated by the growth hormone/IGF axis through the PI3K/Akt pathway. However, pathological cardiac growth is triggered by autocrine and paracrine neurohormonal factors that signal through the Gq/phospholipase C pathway, leading to increased cytosolic calcium and PKC activation. We found IGF1 up-regulated in CAD, but not in DCM hearts, suggesting activation of a compensatory effect through the growth hormone/IGF pathway in end-stage CAD. However, this signal seems to be attenuated by an activation (up-regulation in both groups) of the phosphatase and tensin homolog (TPEN), which blocks signalling downstream of this pathway. Hence, our results indicate that the PLC pathway is activated in end-stage CAD, but not in DCM. These results are supported by the functional clustering, which shows that more genes regulating protein kinase activity are differentially expressed in DCM than CAD hearts (Table 3C).

Among the genes regulating protein kinase activity that were differentially expressed in DCM but not in CAD, we found a sub-group of up-regulated genes, which inhibit protein kinase activity or cell division, including protein kinase inhibitor alpha (PKAI), protein kinase membrane associated tyrosine/threonine 1 (PKMYT1), CDC28 protein kinase regulatory subunit 2 (CKS2) and cyclin-dependent kinase inhibitor 2A (CDKN2A). Thus, an alternative to the Gq/phospholipase C pathway for hypertrophy may be active in end-stage DCM.

In addition to the differences between CAD and DCM in the expression of genes involved in catabolism, our eGOn analysis showed a trend that genes involved in cellular morphogenesis had different expression patterns in CAD and DCM hearts (Table 3D). Although not significantly different between CAD and DCM, many of these genes are of interest because of their involvement in heart failure. We found that four and a half LIM domains 1 (FHL1 or SLIM1) and cysteine and glycine-rich protein 3 (cardiac LIM protein) (CSRP3) were down-regulated exclusively in the DCM samples. Inactivation of cardiac LIM proteins (MLP) is a major genetic factor in cardiomyopathy, possibly caused by a disrupted interaction between the cytoskeleton and the mitochondria, interfering with energy sensing and energy transfer [22]. The differential expression of MLP may represent an important contrast in signalling between end-stage CAD and DCM, and be a potential target for therapy.

4.2. Classifier analysis
Despite a similar clinical end-stage, heart failure resulting from CAD and DCM may develop via different remodelling and molecular pathways. In this study, significant information was gained by generating classifiers which consist of small gene sets whose expression patterns can distinguish between CAD and DCM and may identify robust differences between the diseases. Despite the fact that each classifier chooses the genes to be used independently, and that different methods for discretization were used, four genes appeared in all selected classifiers. These were matrix metalloproteinase 3 (MMP3, stromeolysin1), fibulin 1 (FBLN1), ATP-binding cassette, sub-family B, member 1 (ABCB1) and iroquois homeobox protein 5 (IRX5).

Both MMP3 and FBLN1 are related to remodelling and extracellular matrix structure, which correlates with what we found by functional annotation analysis. MMP3 is known to be regulated in myocardial infarction [23] and dilated cardiomyopathy [24]. FBLN1 is involved in heart development [25]. Several other genes related to the cytoskeleton and extracellular matrix, such as catherin receptor 2 (CELSR2), cadherin 3 (CDH3), protocadherin 1 (PCDH1), mucin 5 (MUC5AC) were down-regulated in both groups. Genes like keratin 10 (KRT10), osteomodulin (OMD), matrix metalloproteinase 15 (membrane-inserted) (MMP15), neurofilament 3 (NEF3), osteonectin (SPARC) and elastin (ELN), were down-regulated only in the DCM hearts. Interestingly, ELN seems to be down-regulated by MMP3 [26]. Thus, the results from classification and functional annotation indicate different patterns of adverse extracellular matrix remodelling between end-stage CAD and DCM hearts.

The remaining two genes from the classifiers are involved in transport (ABCB1) and in regulation of transcription (IRX5). The ABCB1 gene product belongs to the multidrug resistance protein family, and can be an important determinant of the effect of drugs influencing cardiac function, like beta-blockers, cardiac glycosides, and doxorubicin. Expression levels were significantly reduced in DCM both in the present and in a previous study [27], and up-regulated in the CAD heart. Since no difference in medication between CAD and DCM patients was detected, the difference in gene expression probably results from disease factors. Iroquois homeobox transcription factors are evolutionarily conserved transcriptional regulators. In the absence of genes controlled by IRX4, ventricular function deteriorates and cardiomyopathy ensues [28]. Little is known about IRX5, but we found an up-regulation in DCM, suggesting that IRX5 may be a factor in hypertrophy of DCM hearts.

Due to the limited availability of heart failure tissue samples, we had to establish the classifiers on the same samples as they were tested. This is not ideal, but the only solution in this case, and may limit the use of our results as classifiers until they have been reassessed in an independent test set. However, in the present paper the classifiers were used to identify specific genes of interest, making the lack of a test set less critical.

4.3. Clinical relevance
The present study demonstrates that microarray gene expression analysis, may become an important complement to current standard methods, by providing precise diagnoses and by sub-grouping existing disease categories into clusters with more predictable prognosis and response to treatment. For heart failure patients, the clinical benefit of expression profiling may not lie within classification at end-stage, but earlier in the course of the disease, where treatment strategies need to be improved. Microarray data may be incomplete and variable across platforms and between laboratories if standardized protocols for hybridization and data acquisition are not used [29]. However, a central question is whether the predictive microarray data set can be used to extract a limited number of relevant genes whose expression levels convey the same information as the whole set. Recently, Lossos et al. [30] evaluated a 6-gene-set QRT-PCR assay against original microarray data which predicted the clinical course of large-B-cell lymphoma and found that this 6-gene set was sufficient to predict survival. Similarly, our results show that classifier analysis of microarray data can be used to identify pathophysiological processes and potential predictive markers of heart disease.


   5. Conclusion
Top
 Abstract
 1. Introduction
 2. Materials and methods
 3. Results
 4. Discussion
 5. Conclusion
Appendix A. Supplementary data
 References
 
In conclusion, the present study demonstrates that combining gene expression data and functional annotations yields a biologically meaningful and statistically robust analysis which sheds light on the pathobiology of end-stage heart disease. It is also suggested that biological representation of the genes is more important than any particular single gene. By looking at process levels in GO instead of single genes, it is possible to make sense of inherent variation, and detect conserved biological processes.


   Appendix A. Supplementary data
Top
 Abstract
 1. Introduction
 2. Materials and methods
 3. Results
 4. Discussion
 5. Conclusion
 Appendix A. Supplementary data
References
 
Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.ejheart.2006.05.004.


   Acknowledgements
 
Vidar Beisvag is the recipient of a Research Fellowship from the National Council on Cardiovascular Diseases. This study was supported by grants from the St. Olavs Hospital, and the Torstein Erbo, Arild and Emilie Bachke, Ingeborg and Anders Solheim, and Randi and Hans Arnet Foundations. The microarray work was done with support from the national technology platform, which is supported by the functional genomic program (FUGE) of the Norwegian Research Council.


   References
Top
 Abstract
 1. Introduction
 2. Materials and methods
 3. Results
 4. Discussion
 5. Conclusion
 Appendix A. Supplementary data
 References
 

  1. Hwang J.J., Allen P.D., Tseng G.C., et al. Microarray gene expression profiles in dilated and hypertrophic cardiomyopathic end-stage heart failure. Physiol Genomics (2002) 10:31–44.[Abstract/Free Full Text]
  2. Barrans J.D., Allen P.D., Stamatiou D., Dzau V.J., Liew C.C. Global gene expression profiling of end-stage dilated cardiomyopathy using a human cardiovascular-based cDNA microarray. Am J Pathol (2002) 60:2035–2043.
  3. Steenman M., Chen Y.W., Le Cunff M., et al. Transcriptomal analysis of failing and nonfailing human hearts. Physiol Genomics (2003) 2:97–112.
  4. Archacki S.R., Angheloiu G., Tian X.L., et al. Identification of new genes differentially expressed in coronary artery disease by expression profiling. Physiol Genomics (2003) 5:65–74.[CrossRef]
  5. Stanton L.W., Garrard L.J., Damm D., et al. Altered patterns of gene expression in response to myocardial infarction. Circ Res (2000) 6:939–945.
  6. Chugh S.S., Whitesel S., Turner M., Roberts C.T. Jr., Nagalla S.R. Genetic basis for chamber-specific ventricular phenotypes in the rat infarct model. Cardiovasc Res (2003) 7:477–485.
  7. Prabhakar R., Petrashevskaya N., Schwartz A., et al. A mouse model of familial hypertrophic cardiomyopathy caused by a alpha-tropomyosin mutation. Mol Cell Biochem (2003) 251:33–42.[CrossRef][Web of Science][Medline]
  8. Shatkay H., Edwards S., Wilbur W.J., Boguski M. Genes, themes and microarrays: using information retrieval for large-scale gene analysis. Proc Int Conf Intell Syst Mol Biol (2000) 8:317–328.[Medline]
  9. Steenman M., Lamirault G., Le M.N., Le C.M., Escande D., Leger J.J. Distinct molecular portraits of human failing hearts identified by dedicated cDNA microarrays. Eur J Heart Fail (2005) 7:157–165.[Abstract/Free Full Text]
  10. Tan F.L., Moravec C.S., Li J., et al. The gene expression fingerprint of human heart failure. Proc Natl Acad Sci U S A (2002) 99:11387–11392.[Abstract/Free Full Text]
  11. Golub T.R., Slonim D.K., Tamayo P., et al. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science (1999) 286:531–537.[Abstract/Free Full Text]
  12. Norsett K.G., Laegreid A., Midelfart H., et al. Gene expression based classification of gastric carcinoma. Cancer Lett (2004) 210:227–237.[CrossRef][Web of Science][Medline]
  13. Hong H., Tong W., Perkins R., Fang H., Xie Q., Shi L. Multiclass Decision Forest—a novel pattern recognition method for multiclass classification in microarray data analysis. DNA Cell Biol (2004) 23:685–694.[Web of Science][Medline]
  14. Adorjan P., Distler J., Lipscher E., et al. Tumour class prediction and discovery by microarray-based DNA methylation analysis. Nucleic Acids Res (2002) 30:e21.[Abstract/Free Full Text]
  15. Komorowski J., Skowron A., Øhrn A. Rosetta Rough Set Software System. In: Handbook of data mining and knowledge discovery—Klosgen W., Zytkow J., eds. (2001) Oxford University Press.
  16. Midelfart H., Nørsett K., Sandvik A.K., Komorowski J., Yadetie F., Lægreid A. Learning rough set classifiers from gene expression and clinical data. Fundam Inform (2002) 52:1–29.
  17. Livak K.J., Schmittgen T.D. Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) method. Methods (2001) 25:402–408.[CrossRef][Web of Science][Medline]
  18. Dallas P.B., Gottardo N.G., Firth M.J., et al. Gene expression levels assessed by oligonucleotide microarray analysis and quantitative real-time RT-PCR — how well do they correlate? BMC Genomics (2005) 6:59.[CrossRef][Medline]
  19. Grzeskowiak R., Witt H., Drungowski M., et al. Expression profiling of human idiopathic dilated cardiomyopathy. Cardiovasc Res (2003) 59:400–411.[Abstract/Free Full Text]
  20. Berry C., Clark A.L. Catabolism in chronic heart failure. Eur Heart J (2000) 21:521–532.[Free Full Text]
  21. Asemu G., Dhalla N.S., Tappia P.S. Inhibition of PLC improves postischemic recovery in isolated rat heart. Am J Physiol Heart Circ Physiol (2004) 287:H2598–H2605.[Abstract/Free Full Text]
  22. Towbin J.A. The role of cytoskeletal proteins in cardiomyopathies. Curr Opin Cell Biol (1998) 10:131–139.[CrossRef][Web of Science][Medline]
  23. Romanic A.M., Burns-Kurtis C.L., Gout B., Berrebi-Bertrand I., Ohlstein E.H. Matrix metalloproteinase expression in cardiac myocytes following myocardial infarction in the rabbit. Life Sci (2001) 68:799–814.[CrossRef][Web of Science][Medline]
  24. Thomas C.V., Coker M.L., Zellner J.L., Handy J.R., Crumbley A.J. III, Spinale F.G. Increased matrix metalloproteinase activity and selective upregulation in LV myocardium from patients with end-stage dilated cardiomyopathy. Circulation (1998) 97:1708–1715.[Abstract/Free Full Text]
  25. Miosge N., Gotz W., Sasaki T., Chu M.L., Timpl R., Herken R. The extracellular matrix proteins fibulin-1 and fibulin-2 in the early human embryo. Histochem J (1996) 28:109–116.[CrossRef][Web of Science][Medline]
  26. Basalyga D.M., Simionescu D.T., Xiong W., Baxter B.T., Starcher B.C., Vyavahare N.R. Elastin degradation and calcification in an abdominal aorta injury model: role of matrix metalloproteinases. Circulation (2004) 110:3480–3487.[Abstract/Free Full Text]
  27. Meissner K., Sperker B., Karsten C., et al. Expression and localization of P-glycoprotein in human heart: effects of cardiomyopathy. J Histochem Cytochem (2002) 50:1351–1356.[Abstract/Free Full Text]
  28. Bruneau B.G., Bao Z.Z., Fatkin D., et al. Cardiomyopathy in Irx4-deficient mice is preceded by abnormal ventricular gene expression. Mol Cell Biol (2001) 21:1730–1736.[Abstract/Free Full Text]
  29. Bammler T., Beyer R.P., Bhattacharya S., et al. Standardizing global gene expression analysis between laboratories and across platforms. Nat Methods (2005) 2:351–356.[CrossRef][Web of Science][Medline]
  30. Lossos I.S., Czerwinski D.K., Alizadeh A.A., et al. Prediction of survival in diffuse large-B-cell lymphoma based on the expression of six genes. N Engl J Med (2004) 350:1828–1837.[Abstract/Free Full Text]

Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


This article has been cited by other articles:


Home page
 J Am Coll CardiolHome page
K. B. Margulies, D. P. Bednarik, and D. L. Dries
Genomics, transcriptional profiling, and heart failure.
J. Am. Coll. Cardiol., May 12, 2009; 53(19): 1752 - 1759.
[Abstract] [Full Text] [PDF]

Home page
 Physiol. GenomicsHome page
G. E. Haddad, L. J. Saunders, S. D. Crosby, M. Carles, F. del Monte, K. King, M. R. Bristow, F. G. Spinale, T. E. Macgillivray, M. J. Semigran, et al.
Human cardiac-specific cDNA array for idiopathic dilated cardiomyopathy: sex-related differences
Physiol Genomics, April 1, 2008; 33(2): 267 - 277.
[Abstract] [Full Text] [PDF]

Home page
 J Am Coll CardiolHome page
H. Ashrafian and H. Watkins
Reviews of Translational Medicine and Genomics in Cardiovascular Disease: New Disease Taxonomy and Therapeutic Implications: Cardiomyopathies: Therapeutics Based on Molecular Phenotype
J. Am. Coll. Cardiol., March 27, 2007; 49(12): 1251 - 1264.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (8)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Beisvag, V.
Right arrow Articles by Ellingsen, O.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Beisvag, V.
Right arrow Articles by Ellingsen, O.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?