Cardiovascular Genetic Medicine: Genomic Assessment of Prognosis and Diagnosis in Patients with Cardiomyopathy and Heart Failure

Abstract

In the last half century, epidemiologic studies and basic science investigations revealed that hypertension (Kannel et al., Ann Intern Med 55:33–50, 1961), hyperlipidemia (Dawber et al., Am J Public Health Nations Health 49:1349–1356, 1959), diabetes (Kannel et al., Am J Cardiol 34(1):29–34, 1974), smoking (Dawber et al., Am J Public Health Nations Health 49:1349–1356, 1959), and inflammation (Rossmann et al., Exp Gerontol 43(3):229–237, 2008) posed increased risk for cardiovascular disease. These associations served both as risk factors and offered insight into disease pathophysiology. Currently, it is increasingly appreciated that polygenic factors may also play a role as etiologic or risk factors (Chakravarti and Little, Nature 421(6921):412–414, 2003; Dorn and Molkentin, Circulation 109(2):150–158, 2004). Recent technologic advances in genomic screening make the search for these factors possible, and robust technologies are now available for both entire genome screening for expression or single nucleotide polymorphisms. In this paper, we review the basic principles of gene expression and molecular signature analysis in the context of potential clinical applications of transcriptomics.

Transcriptomics

Transcriptomics is the study of the entire complement of mRNA in a cell or tissue [7]. A variety of techniques including microarrays, standardized real-time reverse transcription polymerase chain reaction (RT-PCR), or Quanti-Gene assays are now widely used and validated approaches. Although concerns were raised regarding the technical fidelity of these approaches at their inception [8, 9], several recent multicenter studies and statistical analyses support the idea that these methodologies are robust and as such contain significant value for gene pathway discovery and biomarker development [10–14].

Expression profiling offers the opportunity to develop biomarkers and to discover new pathological pathways activated or suppressed in disease [15]. In this regard, patterns of gene expression may be useful to identify characteristics that are predictive of specific diagnoses [16], prognoses [16], or response to therapy (personalized medicine) [16]. The important difference between molecular signature analysis (MSA) and gene pathway discovery is that predictive signatures consist of a set of genes, not a single molecule [17]. This important principle should be considered for validation of molecular signatures. Evaluation of the predictive accuracy of a biomarker in an independent test set is much more meaningful than technically reproducing expression levels of selected genes with distinct laboratory methods [16, 18, 19]. Nevertheless, there is still disagreement if split sample validation is sufficient as a solitary validation method of MSA, and some groups continue to perform gene expression confirmation with alternative technologies such as real-time RT-PCR or Northern blotting. In this context, the validity of results obtained by standard real-time RT-PCR should not be overestimated as final proof of an experiment, since significant variations of “housekeeping genes” can also occur between distinct phenotypes [20], thereby obscuring the detection of differential gene expression [20]. A suggested solution to this problem is the use of experiment-specific reference genes, which may be selected from the corresponding microarray data [21].

Molecular Signatures: Development of Biomarkers Based on Phenotype-Specific Gene Clusters

The purpose of MSA is to identify a phenotype-specific gene expression pattern for prognosis, disease etiology, or response to therapy, using the principle of ‘machine learning’ or ‘classification’ [7]. First, samples are divided into classes based on clinical parameters and phenotypes [16]. For each class, a transcriptomic biomarker is created and subsequently used to identify the phenotype of unknown samples [16].

A variety of statistical techniques were developed for classification, such as k-means clustering, principal components analysis, and prediction analysis of microarrays (PAM) [22]. PAM uses the method of nearest shrunken centroids [22] to identify and validate the smallest set of genes necessary to predict the phenotype of samples. This algorithm assigns greater weight to genes of which expression is stable within a class and reduces the detection of insignificant background expression levels or “noise”. Gene expression prediction profiles are fit on the basis of 75–90% of samples and then tested on the remaining set. The molecular signature is chosen as the smallest list of genes that can be used to predict the phenotype of an individual sample with adequate accuracy. While the individual expression value of a single gene may have minor predictive power, it can importantly increase predictive accuracy in the context of a gene expression pattern, as it is used for MSA [16, 22, 23].

A concern that is often discussed in the context of classification using microarray data is the high number of variables (approximately 40,000 transcripts on the most recent microarrays) that are available as classifiers to be linked to a certain phenotype [24]. Logistic regression models have shown that “correct” classification can get inflated, if the variable to sample/patient ratio is high. Therefore, high marker to patient ratio can lead to overestimation of prognostic accuracy [16]. In order to overcome this problem, statistics, such as adjusted P values for multiple comparisons, have been developed [25] and successfully applied in many disease states [18, 23, 26–28]. In addition to that, Dudoit and colleagues [29] demonstrated that if the number of variables that is finally used for classification can be reduced to less than 50, accurate classification may be achieved in microarray data. Furthermore, learning curve modulations suggested that classification algorithms with 20–30 samples in the train set are sufficient for successful MSA [24]. After all, it has to be considered that variables of gene expression data used for classification with MSA are not random, but are the most robust genes that characterize a certain phenotype; they are affected by biology and coexpressed as clusters under specific conditions [16].

Transcriptomic-Based Biomarkers in Cardiovascular Medicine

The earliest developments of transcriptomic biomarker research occurred in neoplastic diseases [19, 22, 23, 26, 30], where molecular signatures were able to enhance diagnostic accuracy relative to standard pathology [22, 28] and successfully predicted prognostic outcome of cancer patients, in particular, early metastatic recurrence [19, 26, 30]. These prognostic chips have rapidly made their way into clinical practice as commercially available kits, covered by third-party payers and endorsed as a standard of care by professional societies.

After this early success in oncology, investigators started to test molecular signatures in cardiovascular medicine. Our group performed a proof-of-concept study that transcriptomic biomarkers are feasible for diagnosis in cardiology [31]. We developed a biomarker that distinguished between patients with the two major forms of cardiomyopathy, ischemic cardiomyopathy (ICM) and nonischemic cardiomyopathy (NICM) by using a molecular signature containing 90 genes. Developed in a training set of 48 myocardial samples obtained at transplantation or left ventricular assist device (LVAD) placement (end-stage; n=25), after LVAD support (n=16), and at presentation (biopsy; n=7), the signature performed with very high accuracy when applied to independent samples (sensitivity=100%, specificity=100% in end-stage samples and sensitivity=33%, specificity=100% in samples after LVAD support or at presentation). Importantly, we showed in a subsequent analysis that the MSA or transcriptomic-based biomarkers (TBB) did not comprise all genes detected to be differentially expressed, highlighting the difference between prediction and the full list of genes with altered expression [32].

The clinical application of TBBs is emerging in the cardiovascular arena. The Allomap test, which is an assay for the detection of cardiac transplant rejection, has recently become available [27]. Interestingly, this assay performs well using peripheral blood mononuclear cells (PBMCs) [27] rather than heart tissue, per se. The rationale for the potential use of blood as surrogate for tissue samples was supported by Liew et al. [33], who documented a high degree of overlap between the gene expression profile of PBMCs and diseased tissue. Nine different types of tissue were compared vs PBMCs, among which heart tissue showed 84.2% correlation with blood cells. Since the introduction of the Allomap test in 2005, this diagnostic test has entered clinical practice.

In addition to the successful application of transcriptomic biomarkers for refined diagnosis and patient monitoring, our group developed a molecular signature to predict prognostic outcome in patients with new onset heart failure [34]. Given that heart failure is a major public health issue with an incidence of about 550,000 patients per year, establishing reliable risk assessment is of particular importance. Prognostic laboratory (e.g., BNP, uric acid) and functional parameters (e.g., pulmonary capillary wedge pressure, ejection fraction) have been evaluated, and complex algorithms, such as artificial neural networks, were created to assign different weights to various cardiovascular risk factors based on the observed impact of those factors on clinical outcomes. The prognostic biomarker developed in our laboratory [34] accurately predicted long-term clinical outcome (74% sensitivity and 90% specificity) [34] and was obtainable from a single endomyocardial biopsy [34]. Figure 1 illustrates the general steps for the development of a transcriptomic biomarker. While this marker has potential to enhance prognostic accuracy in combination with classic cardiovascular risk factors, our data also suggest potential therapeutic target genes [34]. Overexpressed genes in patients with an excellent clinical course included those involved in cellular regeneration and angiogenesis [34]. Functional data from animal experiments will be required to determine which differences in gene expression contributed the most to the recovery of heart function.

Fig. 1.

Path for the development of a transcriptomic biomarker from a single endomyocardial biopsy. In previous studies, we successfully developed transcriptomic biomarkers for prognosis and diagnosis in new onset heart failure derived from a single endomyocardial biopsy. Among five to six collected endomyocardial biopsies, one was used to obtain an individual’s molecular signature, while the remaining samples were used for standard diagnostic techniques. Extracted total RNA was tested for integrity with the Agilent2100 Bioanalyzer and used for microarray hybridization with the Human Genome U133 Plus 2.0 array. The next step was analysis with significance analysis of microarrays (SAM) for phenotype-specific differences in gene expression and prediction analysis of microarrays (PAM) to identify the smallest set of genes necessary to predict the phenotype of unknown samples. Finally, a heat map was created to illustrate the molecular signature and as additional step of validation for robustness of genes (unsupervised clustering by Euclidean distance)

Guidelines for Patient Selection and Clinical Study Design

Patient selection and thoroughly planned study design are crucial for the successful development of TBBs [7, 16]. Accurate initial definition of investigated phenotypes is an essential requirement for objective assessment of the predictive accuracy of a TBB for its final validation [7, 17]. This issue is of critical importance and is particularly challenging for diseases affecting organs such as the heart. Most studies to date have explored transcriptomic analyses on explanted hearts, which represent the end-stage of heart disease; thus, processes operative at disease inception may be obscured by studying tissue at this stage [34].

Additional factors to consider are the optimal sample source and time point for sample collection [16, 34]. Many investigators obtain their preliminary data from animal studies before analyzing patient samples. Possible confounding factors, such as gender, age, strain, and environmental conditions are easy to control in animals and, therefore, reduce the number of required samples for statistical strength, while enabling higher throughput and easy sample collection. On the other hand, there can be important interspecies differences and a mechanism that is found in a certain animal model might not be reproduced in humans. In addition to differences in physiology, possible comorbidities in patients may result in clinical data that diverges from results obtained in animal experiments.

The optimal approach, therefore, is to use a source type, which corresponds entirely with the future clinical application of the TBB [16, 34]. While investigations may be started by obtaining samples via biopsy or fine-needle aspiration from the diseased organ, they may be further expanded in screening blood cells or body liquids for possible correlates of gene expression profiles. This would enable time- and cost-efficient assessment of biomarkers in the clinic, e.g., by venipuncture [16, 34].

Given larger availability of tissue from advanced stages of heart failure, most previous studies investigated heart tissue pre- and post-LVAD placement. While several molecular signatures of recovery have been suggested, there has been significant discrepancy between datasets [35] and unclarity regarding differences in gene expression that were initially present and causative for recovery or bystander effects during compensation for heart failure or volume unloading [35–37].

Many of these issues were relevant in a recent case–control study we performed, using endomyocardial biopsy samples obtained from patients with idiopathic dilated cardiomyopathy at first presentation with heart failure symptoms [34]. We compared the gene expression profile of biopsies from patients with good clinical outcome (survival at least 5 years after presentation) vs patients with poor prognosis (death, requirement for LVAD, or cardiac transplant) and developed a prognostic TBB that was able to predict long-term clinical outcome of independent samples with very high accuracy (Fig. 2). Genes that were overexpressed in patients with excellent clinical outcome involved important regulatory functions in DNA replication, gene transcription, and cell cycle, such as RBMS1 and WDR33 [38, 39]. Furthermore, we discovered the overexpression of HIF3A, discussed in the literature as an inductor of vascular endothelial growth factor and erythropoietin [40, 41]; genes involved in neuromuscular development and heart contractility, such as CUGBP2, LUC7-like, and SEMA3B (http://geneontology.org); and RAD50, a stem cell survival factor with telomerase activity [42, 43]. These findings suggest biological plausibility and are potentially responsible for recovery [34].

Fig. 2.

Molecular signature to predict long-term clinical outcome in patients with new onset heart failure: The illustrated transcriptomic prognostic biomarker consists of a cluster of 45 genes, which was derived from a single endomyocardial biopsy and sufficient to distinguish patients with idiopathic dilated cardiomyopathy and poor prognosis (PP, n=18) vs patients with good prognosis (GP, n=25). Patients with PP experienced an event (death, requirement for cardiac transplant, or LVAD placement) within the first 2 years of presentation with heart failure symptoms, while patients with GP experienced long-term event-free survival for at least 5 years. Each column corresponds to a patient sample and each row represents a gene. Samples classified as having PP form a distinct cluster and are highlighted in a red square. Downregulated genes are depicted with red, whereas upregulated genes are labeled blue. Yellow arrows denote misclassified samples [34]. Published with permission from Circulation [34]

Finally, we compared this list with previously published studies investigating heart tissue from patients pre- and post-LVAD placement [12, 13, 35]. Hall and colleagues [12] observed important changes on the molecular level in patients recovering from heart failure, in particular, an activation of the integrin pathway [12, 44], lamin A/C [44], and sarcomeric proteins, such as β-actin [44], α-tropomyosin [44], α1-actinin, and α-filamin A [44]. While these findings are extremely valuable for a better understanding of involved pathways in improved heart function from advanced heart failure, there was no significant overlap with our suggested set of genes for recovery from new onset heart failure [34]. Also, when we compared our prognostic molecular signature with the results from Margulies et al. [35], we found only correlation of three genes (SNRP 70 kDa, obscuring-like 1, and RNA binding motif) that were reported to be involved in recovery. This may suggest that the majority of differentially expressed genes during recovery from LVAD support are specific for advanced stages of heart failure [34], while genes within the biopsy-derived prognostic TBB rather reflect initial differences in gene expression that may lead to better outcome [34]. Thus, our recent study provides both a potentially valuable diagnostic tool and has offered insights into factors contributing to propensity to recover from an initial cardiomyopathic insult.

Where will Transcriptomic Research Guide us in the Following Years?

The utility of biomarkers for prognosis, diagnosis, and therapy responsiveness is increasingly accepted in cardiovascular medicine. Thus, the quest for TBBs for cardiovascular disorders will continue. An advance of particular importance will be the use of the transcriptome of PBMCs as a surrogate for affected tissue, making the application of this approach more feasible.

Transcriptomic analysis may also serve as a valuable tool for guidance in the clinical development of stem cell therapy. Previous studies demonstrated the value of microarray technology as a comprehensive approach for characterization of different stem cell lines [45–49], providing essential information, both for later therapeutic application or manipulation [50, 51]. In addition, it may be used to monitor patient responsiveness to treatment [7] or even select the optimal type of stem cell for the individual patients, so-called personalized medicine [7].

An additional recent application of microarrays of great interest relates to the evaluation of microRNAs (miRNAs), recently described as molecules with important regulatory functions in cardiac remodeling and heart failure [52]. These single-stranded RNA transcripts of 21–23 nucleotides in length either degrade bound mRNA or directly inhibit the translation of mRNA by pairing with their target mRNAs in a sequence-specific manner. Thus, the function of miRNAs is a direct negative regulatory function on gene expression. miRNA of an intron of the MHCα gene has been shown to be responsible for stress-induced cardiomyocyte growth [52]. Since miRNA were discovered only very recently, we will encounter similar problems as during the early phase of microarray technology. Arrays will have to be improved with respect to validity and reproducibility by increasing internal controls within chips, in order to correct for nonspecific binding. Furthermore, there is no consensus on optimal methods for normalization or standard “housekeeping miRNAs”. Guidelines for statistical analysis of miRNAs will be needed (similar to MIAME guidelines) to enhance reproducibility of data. A possible environment would be on public databases in which data can be shared and information of target genes for miRNAs is available (http://microrna.sanger.ac.uk/, http://www.russell.embl.de/miRNA/, http://mirnamap.mbc.nctu.edu.tw/). Successful advancements of microarray technology over the past years give reason for optimism that miRNA data will soon improve in robustness and facilitate the understanding of gene expression patterns and gene interactions of currently used molecular signatures and biomarkers.

Summary

Microarray technology has emerging applications in cardiovascular medicine that include transcriptomic characterization of cells or tissues, novel gene pathway discovery, and importantly, novel clinical biomarker development. TBBs are evolving into highly useful tools for diagnosis and prognosis in a wide variety of disorders including heart diseases. The high precision of transcriptomic analysis to reflect the gene–environment interactions influencing a particular organ system may also be reflected in surrogate tissues such as circulating mononuclear cells, a finding that could make TBB approaches in cardiovascular disease more practical. In addition to mRNA, transcriptomic technology may also offer novel insights into miRNAs that may provide us with understanding of how previously developed transcriptomic signatures regulate physiology or pathophysiological processes in an organism. Together, the developments in transcriptomic technologies over the past decade are leading to a number of highly innovative biomedical applications.