Genomic Analysis of Left Ventricular Remodeling

Abstract

HomeCirculationVol. 120, No. 5Genomic Analysis of Left Ventricular Remodeling Free AccessResearch ArticlePDF/EPUBAboutView PDFView EPUBSections ToolsAdd to favoritesDownload citationsTrack citationsPermissionsDownload Articles + Supplements ShareShare onFacebookTwitterLinked InMendeleyReddit Jump toSupplemental MaterialFree AccessResearch ArticlePDF/EPUBGenomic Analysis of Left Ventricular Remodeling Rizwan Sarwar, MRCP and Stuart A. Cook, PhD, MRCP Rizwan SarwarRizwan Sarwar From the Medical Research Council Clinical Sciences Centre (R.S., S.A.C.) and National Heart and Lung Institute (S.A.C.), Imperial College, London, United Kingdom. Search for more papers by this author and Stuart A. CookStuart A. Cook From the Medical Research Council Clinical Sciences Centre (R.S., S.A.C.) and National Heart and Lung Institute (S.A.C.), Imperial College, London, United Kingdom. Search for more papers by this author Originally published4 Aug 2009https://doi.org/10.1161/CIRCULATIONAHA.108.797225Circulation. 2009;120:437–444The left ventricle (LV) of the heart remodels in response to hemodynamic load through the processes of physiological or pathophysiological hypertrophy.1–3 Despite extensive study of the molecular basis of cardiac hypertrophy and LV remodeling,1,4 the mainstay of medical treatment of maladaptive remodeling remains based on therapies initially devised for treating hypertension. The limited progress in the development of specific therapies for LV remodeling has led some to refer to “the impossible task of developing a new treatment for heart failure” and is the reason why, in part, device therapies have received such recent attention.5–8 To gain a better understanding of the biology of maladaptive LV remodeling, with a view to identifying therapeutic targets, new paradigms and experimental approaches are needed, and genomics provides one such methodology.Genomics is the study of genome-scale data sets, at the DNA or RNA level, and when combined with studies of physiological traits or disease phenotypes, genomics can be used to infer molecular insights. Unlike single-gene studies, which are the favored hypothesis-driven design,4 genomic approaches examine variation in up to gigabases of sequences to find statistical associations between transcripts and traits in a hypothesis-free system. These hypothesis-free studies have often been criticized as unfocused, but following the successes of genome-wide association studies (GWAS) and the recent achievements of integrated genetic and genomic analyses, a new era of genomic experimentation presents itself.9–14 It is notable that GWAS and integrated genomic studies in humans often identify regions associated with disease rather than specific genes and that effect sizes are small; nevertheless, these data remain highly informative.15–17 In this article, we review the genomic analysis of LV remodeling in the context of the recent past, the state of the art, and the imminent use of next-generation sequencing. A glossary of terms is provided in Table 1. Table 1. Glossary of TermsTermDefinitionAlleleOne version of a pair or series of sequences at a specific genomic positioneQTLExpression quantitative trait locusGeneticsThe study of heredity (inheritance)GenomicsThe study of genome level data (eg, the transcriptome or genome)GenotypeAllele combination present at a specific genomic positionGWASGenome-wide association studyLODLogarithm of the oddsLVMLeft ventricular massQTLQuantitative trait locus; a genomic region associated with a quantitative trait or diseaseNGSNext generation sequencingQTTQuantitative trait transcript; a method used to identify transcripts whose expression correlates with a traitRNA-SeqSequencing of mRNA by next generation sequencingSNPSingle nucleotide polymorphismIntermediate traitA trait (most often at the molecular level) that lies between genetic variation and a whole organ/body phenotypeThe Recent Past: Genes That Go Up or Down in the Remodeled HeartSome 5 years after the earliest microarray-based gene expression studies of cancer,18 the first genomic analyses of cardiac remodeling and heart failure were reported.19,20 Since these initial reports, there have been many studies of gene expression in the context of concentric or eccentric remodeling and overt heart failure, and these data have been reviewed comprehensively elsewhere.21–23 The wealth of information generated from genomic analyses of LV remodeling has undoubtedly advanced our understanding of the transcriptional landscape of the LV. However, issues relating to tissue heterogeneity inherent to the heart, which are less applicable for studies of clonal diseases, may have prevented more informative insights. In addition, most studies of LV remodeling have simply reported lists of genes that go up and/or down (“elevator genomics”) and applied these data to predict prognosis, to derive new molecular insights, and for therapeutic target identification, which are disputed extensions of such studies.24–26The questions relating to the clinical utility of genomic analysis of LV remodeling may be most relevant to the field of molecular diagnostics. Although initial studies suggested that gene expression profiles could discriminate between ischemic and nonischemic cardiomyopathy,27 a subsequent larger study has not supported this finding.28 These potential shortcomings have not diminished the publication of studies that suggest a diagnostic utility of genomic analysis of ventricular biopsies (most often of the right ventricle).29 The practicing cardiologist may point out that a coronary angiogram combined with noninvasive imaging is likely a better discriminator of disease pathogenesis and outcome.A major driving force for genomic studies of LV remodeling is the assumed premise that transcriptional variation infers a mechanistic relationship between a change in gene expression and the remodeling process. However, transcriptional variation in response to a remodeling stimulus is time dependent and over a range of magnitudes for different transcripts.30 Furthermore, a lack of variation in a transcript does not exclude its importance for remodeling, which may be mediated at the protein level. Recent studies provide further reasons in regard to why genomic studies of dichotomized “end phenotypes” of the heart (eg, heart failure versus control) have not been overly informative.10,14,31 These new data show that conserved changes in cardiac gene expression induced by a hypertrophic stimulus are limited across genetic backgrounds and that transcript levels can be better predicted by genetic factors compared with blood pressure status (Figure 1). Furthermore, although elevated blood pressure is irrefutably associated with elevated LV mass (LVM),32 many hemodynamic parameters correlate poorly with LVM in humans3,33,34 and the rat.35–37Download figureDownload PowerPointFigure 1. Genetic regulation of cardiac gene expression is predominant over hemodynamic effects in the rat. A and B, LV gene expression profiles were generated in male rats from 3 hypertensive rat strains and from 3 normotensive control strains, and the relationships between the transcription profiles were then ascertained. A, Schematic representation of how gene expression in the 3 hypertensive strains, which exhibited varying degrees of LV hypertrophy, might be expected to be associated with LV hypertrophy. B, Unsupervised hierarchical clustering of global gene expression profiles. Gene expression profiles were predominantly determined by genetic background as opposed to blood pressure effects. Strains: Sprague-Dawley Hanover (SDH); transgenic for mouse Ren2 (TGR+/−) on SDH background; SHR; Wistar-Kyoto (WKY), from which the SHR was derived; and Lyon (LN) hypertensive rat (LH) and Lyon low blood pressure (LL). Adapted with permission from Cerutti et al.31 Copyright © 2006, the American Physiological Society.Although limited, there are compelling examples of genes that have been identified through genomic analyses of remodeling that are now targets for therapeutic intervention, and we provide 2 examples of this. The first example is that of periostin, which was shown to be markedly upregulated in the first microarray study of post–myocardial infarction (MI) remodeling.19 Subsequent studies revealed that periostin is important for post-MI remodeling and repair of the LV,38 which has led to the idea that exogenous periostin may be useful for limiting LV remodeling.39 The second example is Nix, a proapoptotic mitochondrial protein that is upregulated in genetic and pressure overload–induced models of cardiac hypertrophy in the mouse.40 Overexpression of Nix results in cardiomyopathy, whereas cardiac-specific deletion of Nix protects the heart from adverse remodeling, prevents apoptotic cell death, and preserves LV function.40,41State of the Art: Combined Genetic and Correlation Analyses of Gene ExpressionRare monogenic genetic diseases that exhibit the extreme phenotype of a common disease provide invaluable insights into pathophysiological pathways and also help to direct strategies for drug discovery. The validity of this method is exemplified by the work of Brown and Goldstein,42 with the consequent development of statins. However, beyond genes for mendelian forms of hypertrophic and dilated cardiomyopathy, which currently appear to play a limited role in common forms of cardiac hypertrophy,43 our understanding of the genetic control of LV remodeling is incomplete. In addition, although GWAS have identified and replicated common variants that predispose to many common human diseases, including MI and atrial fibrillation,9 only a single GWAS of baseline LV phenotypes has been published thus far.44In the quest to identify new genes that regulate LV remodeling, alternative strategies are needed. One such approach is to study modifier genes in the context of a mendelian trait, which can reveal genomic regions that alter disease phenotypes in response to a single gene mutation. A good example of this is given in a recent study of genetic modifiers of hypertrophy in an extended family with the InsG791 MYBPC3 hypertrophic cardiomyopathy–causing mutation.45 In this study, 3 genomic regions regulating cardiac hypertrophy were identified, but the causative genes remain to be identified. An alternative approach, which we have applied,10,14 combines genetic mapping and gene expression for gene discovery. The success of this approach depends on mapping a physiological trait to the genome, determining the genetic component of gene expression, and integrating these data, which we discuss in detail.Genetic Mapping of Phenotypes to Genomic Regions in RodentsThe essence of mapping a phenotype to the genome is to measure the phenotype in a population, correlate the phenotype with genetic markers across the genome, and identify genomic regions that are significantly associated with the phenotype.46 Broadly speaking, phenotypes/traits fall into 1 of 2 categories: qualitative or quantitative. When a specific genomic region is statistically associated with a quantitative trait, then that genomic region can be labeled as a quantitative trait locus (QTL) in which a regulatory gene(s) is most likely located (Figure 2). The strength of an association of a genetic marker with a quantitative trait identified through linkage analysis is scored by a logarithm of the odds statistic.47 In segregating populations, QTLs are genetically mapped with the use of linkage analysis compared with association testing, which is used in studies of unrelated individuals, such as in GWAS. Although the first attempts to map a QTL were made almost a century ago,48 the first QTLs were only successfully mapped in the late 1980s in maize49 and subsequently in humans.50Download figureDownload PowerPointFigure 2. Schematic example of mapping a QTL for LVM in the rat. A and B, Biallelic markers (A and B) are shown at 3 distinct genomic regions in the rat genome and are tested for association with quantitative variation in LVM in a theoretical population. B, In the case of the graph on the left, there is a statistically significant association on chromosome 2, whereas there is no evidence for genetic control of LVM at the 2 other genomic loci tested (chromosome 9, middle graph; chromosome 18, right graph). C, Schematic multipoint logarithm of the odds (LOD) plot generated with a genetic map and linkage analysis shows evidence for a LVM QTL on chromosome 2; ticks on the x axis represent beginnings and ends of chromosomes. Genome-wide significance (LOD >3.3)47 is depicted as a dotted, horizontal line.When linkage analysis is performed, the density of the genetic markers determines the resolution of the QTL. In simple terms, the greater the number of informative individuals studied, and the more genetic markers there are, the smaller the QTL region is, and the greater are the chances of finding the causative gene in the QTL region.51 We point out that this simplified explanation does not address the complexities of the genetic architecture underlying complex traits and that polygenic and epistatic effects within a QTL can be significant.52 Early genotyping methods, such as analyses of restriction fragment length polymorphisms53 and microsatellites,54 were labor intensive and featured low throughput. More recently, with the development of microarray technologies and the sequencing of the mouse, rat, and human genomes, commercially available genotyping platforms can now simultaneously screen hundreds of thousands of single nucleotide polymorphism markers in addition to structural variants.Linkage analysis, although useful for studies of human families, is especially suited to studies of segregating rodent populations and was originally developed for the study of model organisms.55 In rodent studies, 2 progenitor strains that differ in a trait of interest (eg, LVM) can be bred together and the progeny genotyped and phenotyped to determine which parental genetic markers segregate with the trait. The 3 most common pedigrees used in rodent genetic studies are a backcross, an F2 intercross, and recombinant inbred (RI) strains (Figure 3).56 The backcross and F2 intercross require 2 rounds of breeding, whereas RI strains can take a decade to establish. RI strains can be maintained, in theory, perpetually and are a cumulative resource. In the rat, there are 2 large sets of RI strains, and there are 4 in the mouse (Table I in the online-only Data Supplement); these resources have generated many hundreds of QTLs. It remains the case that the genes underlying most rodent QTLs remain to be identified, but once discovered, gene effects are often robust across species and can translate to humans.57–59 The particular utility of the rat for the study of LV phenotypes is emphasized by the 62 LVM and cardiac mass QTLs already mapped in the rat60 compared with only 9 in the mouse.61Download figureDownload PowerPointFigure 3. Generation of a panel of RI rat strains bred from BN and SHR parents and application of the integrated genetics and genomics study design. RI strains were generated from mating the BN rat with the SHR, which generated the first filial (F1) generation that was then intercrossed (brother-sister mated) to produce an F2 generation. To create the RI strains, the progeny of separate F2 crosses were brother-sister mated, and this was repeated (>F20). For an integrated genetic and genomic study, quantitative gene expression levels of all genes are tested for association with all genetic markers (gray arrows) by linkage analysis. The graphs on the right show 3 possible and distinct outcomes of expression levels for 3 transcripts that are differentially expressed in the progenitor strains: in the 2 progenitor strains; in all RI strains irrespective of genetic information [RI (all)]; and in the RI strains separated by genotype at a single genetic marker [RI (BN), BN allele; RI (SHR), SHR allele]. Each circle represents the mean expression in 1 strain; 10 RI strains are shown for clarity. ***Highly significant; *significant.Genetic Mapping, Expression Profiling, and Correlation Analysis for Disease Gene IdentificationIn a review of the implications of microarrays for cardiovascular medicine, we posed the following question: “To what extent is the expression profile genetically ‘hard wired’?”62 Around this time, this question was being addressed in elegant studies in the mouse,63 fly,64 killifish,65 and human lymphoblastoid cell lines.66 These studies consistently showed that, regardless of species or cell type, there was a significant component of genetic “hard wiring” of gene expression. After these influential studies, the idea that genetic control of gene expression might be used to help to identify the genes underlying complex traits was developed. The methodology that evolved was first termed genetical genomics67 and was subsequently referred to as integrative genomics68 and expression genetics.69 The approach treats gene expression as a phenotype like any other, but given the quantitative nature of the phenotype and utility of gene expression as an intermediate trait, there are significant advantages to this methodology for disease gene discovery.To summarize the integrated genetic mapping and expression analysis approach, if a transcript’s expression level can be genetically mapped to the genome, it is termed an expression QTL (eQTL), and if this eQTL is coincident with the physical location of the eQTL gene in the genome, it is termed a cis-acting eQTL. As with physiological QTLs, the resolution with which an eQTL can be mapped depends on the number of informative individuals genotyped and the genetic marker density. Regions may then be further refined by haplotype analysis or single nucleotide polymorphism marker association fine mapping.51 To use the eQTL approach to find disease genes, it is necessary to identify cis-eQTLs that colocalize to genomic regions known to regulate disease. An eQTL that maps to regions other than its own physical location in the genome is termed a trans-acting eQTL, and these tend to occur in “hot spots.”70 In some instances, trans-eQTLs mapping together to a single genomic location describe a biological pathway, and this can lead to the identification of a major regulator of that pathway.71Microarray expression data can also be used for quantitative trait transcript (QTT) analysis, which correlates gene expression with quantitative traits and selects for those genes with the highest correlation as candidates for the trait.72 Two recent articles, published back-to-back in Nature, applied a QTT approach to the study of obesity.11,12 It has been suggested that QTT analysis may be used in personalized medicine,73 although we suggest that incorporation of genetic information would be needed to realize this aim. We point out that eQTL and QTT studies require applied statistics and bioinformatics, which are beyond the scope of this review and are discussed elsewhere.74Examples of eQTL and QTT StudiesThe very first genome-wide eQTL study was performed in yeast, like many pioneering genomic studies,75 and further proof-of-principle eQTL studies followed in maize, mouse liver, and human lymphoblastoid cell lines76 and then in rat kidney and adipose tissue13 and also in mouse hematopoietic stem cells77 and brain.78 More recently, 2 studies successfully used the eQTL approach to implicated cis-acting eQTLs as candidates for childhood asthma and high-density lipoprotein cholesterol.79,80 In studies of obesity, the power of combined eQTL and QTT studies has been clearly demonstrated and revealed a network of transcripts enriched in resident macrophages in adipose tissue.11,12To date, there have been only a few studies of cardiac phenotypes that have taken advantage of the eQTL and QTT study design. We recently applied these methodologies to identify genes for LVM and susceptibility to heart failure (Table II in the online-only Data Supplement).10,14 The study of LVM used the power of genetic mapping in rat RI strains generated from the Brown Norway (BN) and spontaneously hypertensive rat (SHR) strains and identified osteoglycin (Ogn) as the main candidate for a blood pressure–independent LVM QTL.81 Complementation was demonstrated in the Ogn knockout mouse. We also performed translational QTT studies in human myocardial biopsies from patients with aortic stenosis, in which, of ≈22 000 transcripts analyzed, human osteoglycin (OGN) had the highest correlation with echocardiographically derived LVM. Although Ogn was first discovered in the bovine bone,82 it is known to regulate collagen fibrillogenesis and may play a wider role in tissue remodeling/growth.83–85 In the heart failure study, an F2 cross was performed, and LV ejection fraction, contractility, and LV end-diastolic volume were mapped to rat chromosome 15 as QTLs.14 Expression profiling identified Ephx2, which is involved with a eicosanoid metabolism,86 as the only cis-acting eQTL in the QTL region. Complementation studies showed that the Ephx2 knockout mouse was resistant to heart failure, and EPHX2 was downregulated in human heart failure.In summary, over the last year, eQTL and QTT studies have revealed new genes for obesity, asthma, high-density lipoprotein cholesterol, LVM, and heart failure. Although eQTL studies of the mouse heart have yet to be published, we anticipate that these studies will soon be performed. These will be of particular interest in the AXB/BXA mouse RI strain panel, which has a single QTL for LVM in both male and female mice in a region syntenic with the rat LVM QTL described in our study.10,87 It is also notable that in this mouse panel, LVM QTLs are independent of cardiac myocyte cross-sectional area, length, or volume.87 Given these data and the fact that half of all cells in the heart are nonmyocytes88 and our finding that an Ogn controls LVM, we should examine in more detail the role(s) of cells other than cardiac myocytes in the regulation of LVM.The Imminent Future: Next Generation Sequencing of the TranscriptomeOver the last decade, oligonucleotide microarrays have provided an invaluable tool for analyzing the cardiac transcriptome and for the study of LV remodeling. However, there are significant limitations to microarray technologies that rely on hybridization of samples to predesigned sequence synthesized on a variety of support matrices.62 This methodology affords interrogation of only a fraction of a transcript, does not provide sequence information, and is semiquantitative, with limited dynamic range. In addition, only known genes and previously identified expressed sequence tags can be assessed, the number and content of which differ between platforms.Recently, next generation sequencing (NGS) technologies, which were primarily designed for DNA sequencing (Table 289) and generate up to gigabases of sequence data (Table 3), have been used for the analysis of the transcriptome (Figure 4). These platforms apply a number of sequencing technologies, including sequencing by synthesis and ligation approaches, which can be strand specific in nature.90 A major difference between some of the more commonly used platforms relates to the length of sequence generated (read length). This has important implications because short-read platforms (read length <50 bp) are perhaps ideally suited for quantitative digital gene expression studies (DGE; also known as Deep SAGE [DSAGE]). In contrast, longer-read platforms (read length >200 bp) may be faster, more accurate, and less bioinformatically demanding for de novo DNA sequencing and downstream genome assembly.90 For transcriptome analysis, NGS affords an unbiased analysis of the transcriptional landscape and can reveal quantitative gene expression, splice variation, allele-specific gene expression, and new coding and noncoding genes and enable annotation of 3′ and 5′ untranslated regions (UTRs) of genes to within 50 bp of Rapid amplification of cDNA ends libraries (Figure 5).90–94Table 2. Changing Logistics of Sequencing the Human GenomeThe genomes compared in the table are those produced by the Human Genome Project and 2 other published sequenced human genomes, namely, that of Craig Venter and that of James Watson. Both the Human Genome Project and the Venter genome were sequenced with the use of the same technology, whereas the Watson genome was sequenced by the use of NGS. Reproduced from Waldman89 by permission from Macmillan Publishers Ltd. Copyright © 2008, the Nature Publishing Group.Genome sequenced (publication year)HGP (2003)Venter (2007)Watson (2008)Time taken, y≈13≈4≈.5No. of scientists>28003127Cost of sequencing, $2.7 billion100 million<1.5 millionCoverage8–10×7.5×7.4×No. of institutes1652No. of countries631Table 3. NGS Platforms for Transcriptome ProfilingComparisons of the sequencing capacity and run times between NGS platforms are shown. Data per run estimates are given as approximate values and differ on a single platform depending on the application and the read length and are increasing rapidly with technological advances. Gb indicates gigabases.Company454 Life SciencesDover SystemsIlluminaApplied BiosystemsCurrent platformGS FLX TitaniumPolonator G.007Genome Analyzer IISOLiD 2.0First publication of NGS transcriptomics2006200720082008Read length, bpUp to 500≈2535–5035–50Data per run, Gb≈0.5≈8≈3≈6Run time, h4489696Download figureDownload PowerPointFigure 4. Schematic representation of the amount of sequence data generated from different transcriptomic platforms over time. Each data point represents the amount of sequence interrogated or generated in the first report by each platform (y axis log10 scale) and is shown for spotted arrays, serial analysis of gene expression (SAGE), and commercially available oligonucleotide arrays (Affymetrix, Agilent, and NimbleGen). Data for NGS are shown for a short-read (MegaClone, Lynx Therapeutics; Genome Analyzer, Solexa/Illumina; SOLiD, Applied Biosystems) and also for a long-read NGS (GS 20, 454 Life Sciences). Details and full references are provided (Table III in the online-only Data Supplement).Download figureDownload PowerPointFigure 5. Schematic representations of applications of NGS for the study of the transcriptome. A reference genome with annotated exons (boxes) is shown, and NGS reads have been aligned (arrows below genome) to the reference genome with the use of bioinformatic approaches. Left, By mapping 3′ UTR tags generated by NGS, quantitative digital gene expression or “molecule counting” is possible. Middle, Coverage of the whole transcript with RNA-Seq provides information on alternative splice variants and 5′UTR and 3′UTR usage. Right, Given sufficient coverage, RNA sequencing can also be used to detect coding sequence variation, allele-specific expression, and somatic mutation.To date, a number of NGS studies have been performed in maize, alfalfa, yeast, and the fly,91–97 and the utility of NGS RNA-Seq for translational studies has been shown in a study of human mesothelioma.98 With refinements of the technology, >100× sequence coverage of more than half of expressed genes may be achieved for transcriptome studies.98 With the marked advantages of NGS platforms over more traditional technologies and the precipitous fall in costs associated with NGS, RNA-Seq may truly be a “revolutionary tool for transcriptomics.”90ConclusionsNumerous genomic studies of LV remodeling have been published over the last decade, and although the critic might suggest that many of these studies are observational in nature, there have been some undoubted successes. More recently, integrated eQTL and QTT approaches have revealed new genes for LV remodeling, and we anticipate further insights from this study design, GWAS, deep resequencing, and modifier gene studies over the next few years. There are some obvious “low-hanging fruit” for the next cohort of genomic studies of the heart, such as defining the cardiac transcriptome at the single-molecule resolution with the use of NGS. Although these experiments will reveal a new level of genomic information, it is important to move on from the study designs that have defined the microarray era and apply and develop cross-disciplinary approaches that combine genetics, bioinformatics, computer science, statistics, and NGS. We predict that the application of these integrated methodologies will reveal many more genes underlying cardiac QTLs and lead to the identification of critical pathways and networks for LV remodeling.The online-only Data Supplement is available with this article at http://circ.ahajournals.org/cgi/content/full/CIRCULATIONAHA.108.797225/DC1.Sources of FundingThis article was supported by the Medical Research Council, UK; British Heart Foundation; and Fondation Leducq.DisclosuresDr Cook has a research collaboration with Applied Biosystems. Dr Sarwar reports no conflicts of interest.FootnotesCorrespondence to Dr Stuart Cook, Medical Research Council Clinical Sciences Centre, Faculty of Medicine, Imperial College, Hammersmith Hospital Campus, Du Cane Rd, London, W12 0NN, UK. E-mail [email protected] References 1 Hill JA, Olson EN. Cardiac plasticity. N Engl J Med. 2008; 358: 1370–1380.CrossrefMedlineGoogle Scholar2 Dorn GW II, Robbins J, Sugden PH. Phenotyping hypertrophy: eschew obfuscation. Circ Res. 2003; 92: 1171–1175.LinkGoogle Scholar3 Devereux RB, Savage DD, Sachs I, Laragh JH. Relation of hemodynamic load to left ventricular hypertrophy and performance in hypertension. Am J Cardiol. 1983; 51: 171–176.CrossrefMedlineGoogle Scholar4 Cook SA, Clerk A, Sugden PH. Are transgenic mice the ‘alkahest’ to understanding myocardial hypertrophy and failure? J Mol Cell Cardiol. 2009; 46: 118–129.CrossrefMedlineGoogle Scholar5 Packer M. The impossible task of develo