Mechanical unloading of the heart with a left ventricular assist device (LVAD) significantly decreases mortality in patients with heart failure. Moreover, it provides a human model to define the critical regulatory genes governing myocardial remodeling in response to significant reductions in wall stress. Statistical analysis of a gene expression library of 19 paired human heart samples harvested at the time of LVAD implant and again at explant revealed a set of 22 genes that were downregulated and 85 genes that were upregulated in response to mechanical unloading with a false discovery rate of less than 1%. The analysis revealed a high percentage of genes involved in the regulation of vascular networks including neuropilin-1 (a VEGF receptor), FGF9, Sprouty1, stromal-derived factor 1, and endomucin. Taken together these findings suggest that mechanical unloading alters the regulation of vascular organization and migration in the heart. In addition to vascular signaling networks, GATA-4 binding protein, a critical mediator of myocyte hypertrophy, was significantly downregulated following mechanical unloading. In summary, these findings may have important implications for defining the role of mechanical stretch and load on autocrine/paracrine signals directing vascular organization in the failing human heart and the role of GATA-4 in orchestrating reverse myocardial remodeling. This unbiased gene discovery approach in paired human heart samples has the potential to provide critical clues to the next generation of therapeutic treatments aimed at heart failure.
congestive heart failure is newly diagnosed in over 500,000 Americans each year and contributes to more than 250,000 deaths annually (25). Animal models and clinical trials have confirmed the importance of neurohormonal systems including the renin-angiotensin-aldosterone axis, the sympathetic nervous system, and natriuretic peptides in the pathogenesis of the heart failure phenotype (12, 13, 29, 36, 37, 46, 51). However, the underlying alterations and interactions in gene expression that govern the transition to heart failure remain largely undefined. Furthermore, a number of recent failed drug trials have highlighted the potential limitations of animal models in replicating a complex disease such as human heart failure and identifying the key regulators of the disease (11, 38, 43). Collectively, these findings emphasize the need for resources and tools to effectively utilize human cardiac tissue to increase our basic understanding of the molecular determinants governing the transition to failure.
The left ventricular assist device (LVAD) is a mechanical device that replaces the pumping of the cardiac ventricle in patients with severe refractory heart failure and has been shown to significantly decrease mortality (42). The device affords near total chamber decompression and reduction of wall stress and achieves the most rapid and complete reverse remodeling of any model to date (1, 17, 18, 30–33, 42). The objective of this study was to utilize an unbiased gene discovery approach to identify the role of mechanical unloading with a LVAD on gene expression in the human heart. Insertion of the device requires removal of a core of myocardial tissue at the apex of the heart to insert the drainage cannula. This tissue is then paired with tissue obtained from the explanted heart at the time of subsequent transplantation to define changes in gene expression in response to mechanical unloading. These paired samples provide a unique opportunity to define the critical genes regulated by workload and wall stress leading to myocardial remodeling. The paired sample design of this study is especially useful in reducing patient variability in gene expression and isolating those genes specifically regulated by mechanical unloading.
Analysis of a compendium of 19 paired patient samples revealed 22 genes that were significantly downregulated and 85 genes upregulated. Included in this list were a host of genes governing vascular organization, as well as Forkhead family genes, and those in the angiotensin/insulin signaling axis. In addition, mechanical unloading led to a significant reduction in GATA-4 binding protein, a critical mediator of hypertrophy and remodeling of the mouse and rat heart. To our knowledge this is the first evidence demonstrating that mechanical unloading of the heart with a LVAD results in significant alterations in the regulation of vascular signaling genes as well as a decrease in GATA-4.
MATERIALS AND METHODS
Human left ventricular samples were collected during both LVAD implantation and transplantation. All tissue was immediately frozen in liquid nitrogen and stored at −80°C. IRB-approved informed consent was obtained from all participating patients prior to tissue collection.
Tissue was placed in TRIzol reagent and homogenized using a rotor-stator homogenizer. Homogenates were transferred to a Phase Lock Gel (PLG) Heavy tube (Eppendorf), chloroform was added, and the mixture was centrifuged. Supernatants were collected, and isopropanol was added to precipitate the RNA. The RNA pellet was washed with 80% ethanol, then air-dried before dissolving in 100 μl of RNase-free H2O. RNA was further purified using the RNeasy Mini protocol according to the manufacturer’s directions (Qiagen).
Double-stranded cDNA synthesis.
Total RNA of 10 μg was used in the synthesis of double-stranded cDNA according to the manufacturer’s directions (Invitrogen) with a T7-(dT)24 primer and 200 U/μl of SuperScript II RT. Second-strand synthesis was performed according to the manufacturer’s directions. The PLG phenol/chloroform extraction and ethanol precipitation procedures were used for the cleanup of synthesized double-stranded cDNA.
In vitro transcription-synthesis of biotin-labeled cRNA.
Biotin-labeled RNA targets were obtained by using the Enzo BioArray High Yield RNA Transcript Labeling Kit via in vitro transcription from bacteriophage T7 RNA polymerase promoters according to the manufacturer’s directions (Enzo). The cRNA was then fragmented according to the Affymetrix protocol. Quality and fragmentation of cRNA was confirmed with gel electrophoresis.
Target hybridization and probe array wash and stain.
Hybridization cocktail for single standard probe array (HG-U133A, Affymetrix) was made by mixing 15 μg of fragmented cRNA with control oligonucleotide B2, 20× eukaryotic hybridization controls, herring sperm DNA (10 mg/ml), acetylated BSA (50 mg/ml) 2× hybridization buffer, and RNase-free water. Hybridization was performed in our Affymetrix Core Facility.
Real-time quantitative PCR.
Real-time quantitative PCR was used to further confirm and quantify the detected gene expression changes as previously described (10, 50). A table of primers and real-time quantitative PCR conditions is listed in the online Supplementary Material (available at the Physiological Genomics web site).1
Each array was initially subjected to a strict set of quality control criteria prior to inclusion in our data set. This included the following. 1) We checked for uniform hybridization of the B2 oligonucleotide on the border of each array. 2) The entire array was inspected visually for gradients in hybridization. 3) The expression value of the GAPDH probe set corresponding to the 5′ end of the gene was at least half of the expression value of the GAPDH probe set corresponding to the 3′ end of the gene. This test reflected the quality of the RNA as well as sample processing. 4) Raw Q values reflecting the noise from each array were less than 8.0. Q values are calculated for each array and are based on small variations in the digitized signal as the scanner samples the surface of each probe array. 5) Scaling factor for each array was below 10. All arrays that did not meet these criteria were excluded from the analysis.
All arrays were processed through Gene Expressionist (GeneData, Basel, Switzerland) utilizing a hierarchical clustering algorithm to assess similarity. A cutoff of 30% similarity was applied to create groups of similar arrays for detection of abnormal signal behavior and masking. For each group of similar arrays a reference array was computed by a robust feature-wise average. The algorithm for abnormal signal behavior identifies localized areas of light or dark features, “defects.” Abnormal signal behavior is determined by comparison of each experiment to the reference experiment. Prior to the comparison the experiment is normalized in memory with respect to the reference experiment. This normalization applies a nonlinear transformation so that the signal response of the experiment equals that of the reference. This ensures the comparability of the experiment to the reference experiment. The difference signal is computed as the logarithm of the ratio of the experiment and the reference experiment. Defects are recognizable in the difference signal as local areas dominated by either positive differences in the case of bright defects, or negative differences in the case of dark defects. Differentially expressed genes will also lead to positive or negative differences and might be interpreted as defects. This is avoided by using the information of the array layout: feature sets that belong to the same gene are excluded from defect detection if these show a common deviation from the reference experiment. Feature intensities for each array were condensed into single intensity values per gene using the Affymetrix statistical algorithm (MAS 5.0), with τ = 0.015, α1 = 0.04, α2 = 0.06, and a target intensity of 500. Normalized expression values were used with “significance analysis of microarrays” (SAM) and paired Student’s t-tests.
The SAM program (Stanford University) was used to identify significant differential gene expression using the paired response analysis of logged expression (48). SAM calculates a significance score for each gene based upon its gene expression change relative to the standard deviation of repeated values. Genes with a score greater than the adjustable threshold are considered potentially significant. From the user-defined threshold, SAM calculates a false discovery rate (FDR) which is the number of false positives divided by the number of significant genes. An FDR of 0.76% was chosen, and the fold change cutoff was 1.2. We performed 500 permutations on the data set. The fold change in Table 2 is the paired post-LVAD/pre-LVAD fold change averaged from all 19 patients.
All 38 expression data files have been uploaded for access on the National Center for Biotechnology Information Gene Expression Omnibus (GEO) site with the following accession numbers: GSM14936 to GSM14973.
The objective of this study was to utilize an unbiased gene discovery approach to identify genes whose expression was significantly altered by mechanical unloading of the failing heart with an LVAD. Gene expression profiles from 19 paired human heart samples taken at the time of LVAD implant and explantation were compiled from the HG-U133 Affymetrix GeneChip containing over 22,000 probe sets. All patients included in this analysis received a transplanted heart following explant of the LVAD. Patient characteristics are summarized in Table 1. Average age of the 19 patients was 51 ± 2 yr, mean time on support was 159 ± 40 days, and average ejection fraction was 19 ± 2% (mean ± SE). Five of the 19 patients had clinical evidence of coronary artery disease and were classified as ischemic, 8 patients were classified as nonischemic (no evidence of coronary artery disease), and 6 patients presented with an acute myocardial infarction within 10 days of LVAD implant. At the time of LVAD implantation the majority of patients were on inotropic agents, intravenous vasodilators, digoxin, angiotensin converting enzyme inhibitor (ACE-I) or angiotensin receptor blocker, beta-blockers, and diuretics (Table 1).
Analysis of the compendium of the 19 paired human heart samples with SAM revealed 22 genes that were significantly downregulated and 85 genes that were upregulated with a fold-change cutoff set at 1.2 (20% difference). The delta was set at 0.58, yielding an FDR of less than 1%. A complete list of these genes is in Table 2. In comparison, a paired Student’s t-test with a more liberal P < 0.05 cutoff identified 81 genes that were significantly downregulated and 124 genes that were upregulated (Supplemental Table S1). Eighty-three of the 85 upregulated genes identified with SAM were included in the 124 upregulated genes identified with the paired t-test. The two genes not included in the paired t-test were SON DNA binding protein (A1936458) and FLJ20288 protein (NM_024668.1). Twenty of the 22 downregulated genes identified by SAM analysis were included in the paired t-test. The two genes not included were Homo sapiens cDNA FLJ25677 fis, clone TST04054 (AI143879), and integrin cytoplasmic domain-associated protein 1 (NM_004763.1). We have reconfirmed changes in gene expression of over 30 genes to date (included in this analysis as well as in earlier publications; Ref. 10). A list of genes reconfirmed in this analysis but not included in the discussion, as well as the respective primer sets, are listed in the Supplementary Material (Supplemental Table S2; Supplemental Figs. S1 and S2). These genes include pyruvate dehydrogenase kinase 4, frizzled 7, thioredoxin interacting protein, and brain natriuretic peptide (BNP).
The highest ranked downregulated gene in the SAM analysis was neuropilin-1, a VEGF receptor that mediates mitogenic signals for endothelial cells. The other vascular-related genes included in the list included Sprouty1, stromal-derived factor 1, FGF9, endomucin, and protocadherin 9. The discovery of Sprouty1, a gene originally defined in the Drosophila trachea, in the heart is a novel finding. Real-time quantitative PCR analysis was used to confirm the upregulation of Sprouty1 and downregulation of neuropilin-1 (Fig. 1).
Next, we performed two-dimensional hierarchical clustering to identify genes that correlate with these statistically significant genes involved in vascular organization. The cluster of genes surrounding neuropilin-1 are listed in Table 3 and include ced-12 (ELMO2) which regulates cell migration, and the rho/rac guanine nucleotide exchange factor (GEF) 2 that is involved in signal transduction. Interestingly, a family of ATP-binding cassette members also grouped closely. The relevance of this is not known. Figure 2 shows the two-dimensional hierarchical cluster of genes surrounding stromal-derived factor 1, which include angiopoietin-like 2, ephrin-B2, transmembrane 4 superfamily, and cadherin 5. Earlier work in our lab reconfirmed the downregulation of the transmembrane 4 superfamily gene (10). These findings further suggest a role for mechanical unloading in the regulation of vascular organization.
Included in the list of significantly downregulated genes from the paired Student’s t-test was GATA-4 binding protein (P < 0.002) (See Fig. 3). Work in mouse and rat models has characterized GATA-4 binding protein as a critical gene regulating myocardial hypertrophy (19, 21, 35). A decrease in the expression of GATA-4 binding protein in the heart on mechanical support which undergoes “reverse remodeling” and a significant decrease in myocyte size (1, 8, 30, 40) is a compelling finding. The downregulation of GATA-4 binding protein was reconfirmed by real-time quantitative PCR (Fig. 3).
Other genes of interest identified from the analysis include the family of Forkhead box transcription factors, FOX03A and FOX01A. Earlier work from our group identified a significant upregulation of FOX03A in seven of the nonischemic patients included in this analysis. It is of interest that this gene remains statistically significant as we build more patients into the analysis, including patients with ischemic cardiomyopathy. To begin to define signaling networks through which FOX03A may be involved, we used a distance metric to define genes following an identical pattern to FOX03A. This approach revealed a striking similarity between FOX03A and the angiotensin II type 1 receptor in the nonischemic cohort (Fig. 4) which is also significantly upregulated in patients following ventricular unloading.
Our goal was to identify genes whose expression was significantly altered in response to mechanical unloading in the failing human heart. We analyzed a compendium of microarray data from 19 paired human heart samples harvested pre- and postimplantation of a LVAD. A list of 107 genes were deemed statistically significant utilizing SAM with an FDR of less than 1%. Two novel discoveries from this analysis were the significant changes in genes governing vascular organization and networks and the decrease in GATA-4 expression in response to mechanical unloading of the heart.
We identified significant regulation of a group of novel genes governing vascular organization and migration in response to mechanical unloading. These finding suggest that ventricular wall stress and strain is an important mediator of vascular organization. Recent work in skeletal muscle and heart demonstrates that exercise, stretch-induced overload, chronic electrical stimulation, and increases in muscle blood flow significantly increase VEGF gene expression and capillary growth (7, 22, 26, 34, 41). Our data would suggest that unloading of the heart results in changes in gene expression congruent with a microenvironment that would inhibit capillary growth, new vessel formation, or the migration of circulating bone marrow or progenitor cells. In summary, our findings support this earlier work demonstrating that stretch and workload are important signals governing vascular networks. Ongoing work in the laboratory is focusing in more detail on the role of mechanical unloading on vascular gene expression and vascular reorganization, given its potential implications in determining the extent of myocardial recovery following LVAD implantation.
The downregulation of GATA-4 binding protein was a novel finding and may serve as a specific diagnostic marker of myocyte remodeling in response to a decrease in mechanical load. BNP harbors a GATA-4 consensus sequence in its promoter. The fact that BNP gene expression does not parallel GATA-4 expression in all cases suggests a likely role for other cofactors in the regulation of BNP expression, including AP-1 (21, 45). In fact, BNP expression was not always reduced in patients following mechanical unloading. The reason for this finding may be the ongoing presence of right ventricular failure in a subset of patients. The advantage of an unbiased gene discovery approach is the potential to discover and define the critical clues and basic mechanisms underlying the regulation and relationship of GATA-4 and BNP. For example, our data would suggest that a decrease in GATA-4 binding protein expression is not sufficient to induce changes in BNP expression. One could begin to define other transcriptional factors regulating BNP expression, upstream factors, and feedback loops, governing GATA-4 expression by focusing on the significant gene expression data or applying pathways software to the data. The gene expression data also provides a short list of potential GATA-4-dependent genes that have yet to be discovered.
Our analysis also identified significant alterations in the Forkhead family of transcription factors. Mechanical unloading resulted in a significant increase in FOX03A and FOX01A. A role for mechanical strain on the regulation of Forkhead family members was recently identified in the vasculature by Sedding et al. (44). Specifically, vascular injury, which induces significant stress and strain on the vessel wall, leads to a significant decrease in transcriptional activation of downstream Forkhead-like family members (44). Utilizing a distance metric designed to identify genes whose expression pattern most closely positively correlated with Forkhead, we identified the gene encoding the angiotensin II type I receptor. This is the first report demonstrating a potential correlation between Forkhead and angiotensin. These findings may provide clues as to genes regulating Forkhead family members and/or signaling pathways affected by Forkhead. The stimulus for the increase in the AT1a receptor is unclear. Herzig et al. (21) demonstrate a direct upregulation in AT1 with increased angiotensin II. However, James et al. (24) reported a significant lowering of both renin and angiotensin II levels in the plasma of patients following LVAD placement. Less than 5% of the patients were receiving an angiotensin receptor blocker following LVAD implant; thus it does not appear to be mediated through pharmacological blockade of the receptor. Chen et al. (9) recently demonstrated a significant upregulation of the apelin-angiotensin receptor-like 1 gene in response to unloading with the LVAD. Interestingly, upregulation of AT1 receptors decreases myocardial microvessel density (14), which would be consistent with our gene expression data reflecting an anti-angiogenic environment.
A review of the literature emphasizes our limited understanding of the genomic and morphological changes that occur in the heart in response to mechanical unloading. Studies to date looking at gene and protein expression in response to ventricular support have focused on ANP/BNP, Ras/ERK, G-protein-coupled signaling, calcium handling proteins, metabolic genes, apoptosis mediators, TNF, and matrix metalloproteinase expression (2–4, 10, 15, 16, 18, 20, 28, 39, 47, 49). We have reconfirmed earlier work demonstrating significant upregulation in TIMP3 expression (27), as well as the downregulation of phosphoERK protein expression reported by Flesch et al. (18) in response to mechanical unloading in a separate study (23). Interestingly, Sprouty1, a gene significantly upregulated in response to mechanical unloading, is an intrinsic negative inhibitor of ERK.
The paired design of the study and the stringent approach utilized in this analysis leading to an FDR of less than 1% most likely decreases the potential variability induced by age, sex, length of time on support, or pharmacological-dependent differences that clearly play a role in the transcriptome pattern (6). The mean age of the patients is 51 ± 2. The 19 pairs included 4 females and 15 males. We recently utilized a penalized least squares approach to define the role of length of time on support on gene expression (21a). As expected, we identified a subset of genes whose expression was influenced by the length of time of mechanical support. The focus of this paper was to identify those genes that were affected by mechanical unloading irrespective of the length of time on support. To date we have reconfirmed expression of 19 genes by real-time quantitative PCR (10, 23). Our analysis also confirmed significance of a set of genes identified and reconfirmed by real-time quantitative PCR in an earlier analysis of a subset of seven nonischemic pairs included in this analysis; these genes included FOX03A, metallothionein IH, GADD45, connexin 43, and transmembrane 4 superfamily 1 (10). A head-to-head comparison of the genes shared in common between our analysis of eight nonischemic patients (Supplemental Table S3) and an earlier report by Chen et al. (10) of seven nonischemic pairs revealed a list of 25 genes that notably also included neuropilin-1, SDF-1, CD163 antigen, TNF superfamily member 10, and metallothioneins 1X, 1L, and 2A.
Earlier work by Blaxall et al. (5) suggested that the underlying etiology of the patients receiving mechanical support played an important role in gene expression patterns. To address this issue, we redid the analysis within each separate cohort. The age, length of time on support, left ventricular ejection fraction, and medications were not significantly different between etiology cohorts. As expected, the gene list of statistically significant genes within each cohort included a subset of different genes, reconfirming earlier work by Blaxall et al. (5). The lists of statistically significant genes within the nonischemic, ischemic and acute MI cohorts are listed in the Supplemental Tables S3–S5. A Venn diagram revealed that when analyzed as separate groups with a strict cutoff of P < 0.005, only one gene with an unidentified function was found in common to all three etiologically distinct cohorts: DKFZP434J214 protein (AL556438). Although it is clear that the underlying etiology is an important variable, this analysis is weakened by the loss of power within each cohort due to the decreased sample size compared with the overall n of 19. Thus, at the current time, we feel more confident with our overall SAM analysis of all patient samples utilizing a strict FDR and have thus chosen to focus on these genes. The fact that we were able to reconfirm the significant changes in gene expression in the overall analysis of 19 paired samples further strengthens our findings. Collaborative studies are currently underway to define a series of genes that best discriminate distinct etiology subgroups.
In summary, we have utilized an unbiased gene discovery approach to define the molecular response of the human heart to mechanical unloading. This analysis led to the discovery that the LVAD induces a significant reduction in GATA-4 expression, which may serve as a specific marker of cardiomyocyte remodeling in response to unloading. Second, we identified a cluster of genes governing vascular organization and migration including FGF9, Sprouty1, sdf-1, and neuropilin-1 whose expression was significantly altered in response to mechanical unloading. These findings may provide important new targets for therapeutic intervention.
This work was supported by the Lillehei Heart Institute at the University of Minnesota (to J. L. Hall) and the Minnesota Medical Foundation (to J. L. Hall and L. W. Miller).
↵1 The Supplementary Material for this article (Supplemental Tables S1–S5, as well as Supplemental Figs. S1 and S2) is available online at http://physiolgenomics.physiology.org/cgi/content/full/00004.2004/DC1.
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).
Address for reprint requests and other correspondence: J. L. Hall, Lillehei Heart Institute, Univ. of Minnesota, 420 Delaware St., Minneapolis, MN 55455 (E-mail:).
- Copyright © 2004 the American Physiological Society