Unique MicroRNA Profile in End-stage Heart Failure Indicates Alterations in Specific Cardiovascular Signaling Networks*

It is well established that gene expression patterns are substantially altered in cardiac hypertrophy and heart failure, but the reasons for such differences are not clear. MicroRNAs (miRNAs) are short noncoding RNAs that provide a novel mechanism for gene regulation. The goal of this study was to comprehensively test for alterations in miRNA expression using human heart failure samples with an aim to build signaling pathway networks using predicted targets for the miRNAs and to identify nodal molecules that control these networks. Genome-wide profiling of miRNAs was performed using custom-designed miRNA microarray followed by validation on an independent set of samples. Eight miRNAs are significantly altered in heart failure of which we have identified two novel miRNAs that are yet to be implicated in cardiac pathophysiology. To gain an unbiased global perspective on regulation by altered miRNAs, predicted targets of eight miRNAs were analyzed using the Ingenuity Pathways Analysis network algorithm to build signaling networks and identify nodal molecules. The majority of nodal molecules identified in our analysis are targets of altered miRNAs and are known regulators of cardiovascular signaling. A heart failure gene expression data base was used to analyze changes in expression patterns for these target nodal molecules. Indeed, expression of nodal molecules was altered in heart failure and inversely correlated to miRNA changes validating our analysis. Importantly, using network analysis we have identified a limited number of key functional targets that may regulate expression of the myriad proteins in heart failure and could be potential therapeutic targets.

Heart failure has been classified as an epidemic of the 21st century and is now the major cause of morbidity in the elderly in the United States. End-stage heart failure is characterized by significantly perturbed neurohormonal and mechanical (hemodynamic) stimuli to the heart. The altered pathological signaling leads to remodeling of the heart with adaptive to maladaptive hypertrophy transitioning into dilated cardiomyopathy (DCM). 3 DCM is the most common and well documented outcome of various deleterious stimuli the heart perceives (1). DCM is characterized clinically by left ventricular dilatation, ventricular wall thinning, and homogeneous myocardial dysfunction leading to congestive heart failure (1). The myocytes under the continuously changing conditions of biomechanical stress during this transition undergo a remodeling process through the activation of intracellular signaling pathways and transcriptional mediators (2). The pathological end-stage DCM is a result of the concomitant cross-talk between various deleterious and compensatory signaling pathways. The balance between these two dynamic pathways ultimately determines the progression of the pathology. Despite significant advances in identification of genes and signaling pathways, the overall complexity of hypertrophic remodeling suggests the involvement of additional global regulatory mechanisms modulating signaling networks. A growing body of evidence suggests that genome-encoded regulatory RNA molecules such as micro-RNAs (miRNAs) regulate many processes like cell proliferation, cell death, metabolism, and neuronal patterning (3).
MicroRNAs (miRNAs) are 18 -25-nucleotide noncoding RNAs that are known to regulate gene expression in a sequence-specific manner (3)(4)(5). MicroRNAs regulate gene expression by binding to mRNAs with the consequence of mRNA degradation or translational inhibition of targeted transcripts (3)(4)(5). Initial studies of miRNAs in animal models of hypertrophy and heart failure showed specific signature patterns suggesting that they could be used as valuable biomarkers for disease (6), but the high sequence conservation of miRNAs across the metazoan species suggests strong evolutionary pressure for potential regulatory roles in complex biologic processes (3,7). Indeed, recent studies have documented the regulatory role of miRNAs in DCM and cardiac development in mice (8,9). Studies by Srivastava and co-workers (10) using miRNA 1 knock-out mice have shown that miRNA 1 plays a critical role in cardiac development. Importantly, studies from Olson and co-workers (11) and Condorelli and co-workers (12) have shown in mice that miRNAs play a critical role in development and/or progression toward heart failure.
The targeting sequence of miRNA theoretically allows for binding of miRNAs to many gene products ultimately regulating the expression of target genes. Therefore, miRNAs could potentially play a significant role in global regulation of signaling networks during remodeling and transition to heart failure. Despite the potential role of miRNAs in targeting many genes, previous studies on miRNA targets in the cardiovascular system have been limited to analyzing specific gene products. Importantly, the appreciation that miRNAs could play a global role in regulating cardiac function and heart failure was recently elucidated by cardiac specific knock-out of dicer, an endonuclease critical for maturation of miRNAs (13). Absence of dicer in the heart leads to loss of mature miRNAs leading to significant DCM and heart failure suggesting a pivotal role for miRNA in globally regulating cardiac function. Considering the global regulatory role of miRNAs, previous cardiac miRNA studies are potentially limited as many of the targets could be concomitantly regulated in vivo. Furthermore, a single gene product may be a target for multiple miRNAs adding complexity in the regulatory process that cannot be appreciated by focusing on the effects of specific miRNA on specific genes.
This study was designed to elucidate the global regulation of the signaling networks by a unique miRNA pattern in end-stage human heart failure. Therefore, the goals of our study were as follows: (a) to measure the changes in expression of 288 human miRNAs using a microarray hybridization platform; (b) use an independent set of human heart failure samples to validate this data set; (c) through an in silico strategy of bioinformatics delineate canonical/functional pathways regulated by genes that are potential targets of miRNAs in heart failure to provide a global picture of the molecular network and nodal molecules regulated by these miRNA; (d) analyze the expression pattern of target nodal molecules using the available gene expression data bases from end-stage human heart failure; and (e) validate by immunoblotting for critical nodal molecules from the same set of patient samples to demonstrate inverse correlation between protein expression and miRNA level. By using this systems biology approach, multiple genes and pathways involved in the complex pathology of heart failure have been visualized simultaneously leading to identification of potential novel therapeutic targets.

EXPERIMENTAL PROCEDURES
Patient Samples-Tissue from the left ventricular free wall was obtained from explanted hearts of transplant recipients at the Cleveland Clinic with a diagnosis of DCM. The nonfailing control hearts were obtained from unmatched donors whose hearts were not suitable for transplantation despite normal ventricular structure and function as measured by echocardiography. The hearts were arrested and transported in ice-cold, oxygenated cardioplegic solution (14). Once in the laboratory the tissue was flash-frozen in liquid N 2 and stored at Ϫ80°C. All protocols for tissue procurement and procedures carried out on the tissues were in performed in compliance with institutional guidelines for human research and approved by the Cleveland Clinic Institutional Review Board.
RNA Isolation-RNA was isolated as described previously (14). Briefly, 100 mg of left ventricular tissue was homogenized using TRIzol reagent (Invitrogen), and the homogenized samples were incubated at room temperature for 5 min. Chloroform was added to the samples, vigorously mixed, and incubated at room temperature for 5 min. Following incubation, the samples were centrifuged at 12,000 ϫ g for 15 min at 4°C. RNA was precipitated from the aqueous phase by addition of isopropyl alcohol to a fresh tube containing the supernatant aqueous phase. The integrity of the RNA was tested by spectroscopic analysis and by resolving the isolated samples on a denaturing formaldehyde gel.
Target Preparation and Array Hybridization-Target preparation and array hybridization were carried out as described previously (15). Briefly, 5 g of total RNA was added to biotinylated oligonucleotide primer. Following incubation, the first strand was synthesized using Superscript II RNase H Ϫ reverse transcriptase. After synthesis of the first strand, the reaction was incubated at 65°C to denature the RNA/DNA hybrids and degrade RNA templates. The labeled targets were then used for chip hybridization. Hybridization was carried out on the miRNA microarray (Ohio State Comprehensive Cancer Centre, version 3.0) containing 627 probes for mature miRNA corresponding to 288 different human miRNAs spotted in quadruplicate. Often, more than one probe set is present for a given mature miRNA, and there are quadruplicate probes corresponding to most precursor miRNAs. The detection of biotincontaining transcripts was carried out by streptavidin-Alexa Fluor 647 conjugation, and scanned images (Axon 4000B) were quantified using the GenePix 6.0 software (Axon Instruments).
Computational Analysis of miRNA Microarray Data and miRNA Target Prediction-Microarray images were analyzed by GenePix Pro. Average values of the replicate spots for each miRNA were background-subtracted, normalized, and subjected to further analysis. Global median normalization and Lowess normalization of the heart microarray data were carried out using BRB ArrayTools (15). The probes with Ͼ70% missing data were excluded from further analysis. Differentially expressed miRNAs between control and dilated cardiomyopathic samples were identified by using t test procedure within significance analysis of microarrays. Furthermore, Lowess normalization was carried out to analyze differentially expressed miRNAs in control versus diseased state. Following the identification of differentially expressing miRNAs, the predicted targets for these differentially expressed miRNAs were identified using TargetScan 3.1 and PictTar data bases (16 -18). Use of the recently released TargetScan 4.2 did not significantly alter the nodal molecules identified in this study. The need for analysis using a variety of data bases was based on the need to encompass all the potential targets as they are built using slightly different algorithms. We have used results of predicated targets from TargetScan data base to carry out pathways and network analysis.
Network Analysis-A data set containing genes and the corresponding expression values was uploaded into the Ingenuity Pathways Analysis Network TM application. The dataset mole-cules of our interest (predicted targets of altered miRNA), which interact with other molecules in the Ingenuity knowledge base, are identified as network-eligible molecules. Network-eligible molecules serve as "seeds" for generating networks. Network-eligible molecules are combined into networks that maximize their connectivity in the Ingenuity knowledge base. A defined network is limited to a maximum of 35 molecules, and the additional molecules from the Ingenuity knowledge base are used to connect networks resulting in large merged networks. Predicated targets for the altered miRNA (see under "Computational Analysis of miRNA Microarray Data") gene were mapped to the corresponding gene object in the Ingenuity knowledge base. A specific value for cutoff of 1.5 was set to identify genes whose expression was significantly differentially regulated. These genes, called focus genes, were overlaid into a global molecular network developed from the information contained in the Ingenuity Pathways Knowledge Base. Networks of these focus genes were then algorithmically generated based on their connectivity. A network pathway is a graphical representation of the molecular relationships between genes/gene products. Genes or gene products are represented as nodes, and the biological relationship between two nodes is represented as an edge (line). All edges are supported by at least one reference from literature, textbook, or from canonical information stored in the Ingenuity Pathways Knowledge Base. Nodes are displayed using various shapes that represent the functional class of the gene product. Edges are displayed with various labels that describe the nature of the relationship between the nodes (i.e. P for phosphorylation, T for transcription, etc.).
Canonical Pathway and Functional Analysis-Canonical pathway analysis was carried out using the Ingenuity Pathways Analysis (IPA) library of canonical pathways by uploading the data set of predicted targets for the significantly altered mi-RNAs in DCM to the IPA server. These target genes (focus genes) were analyzed for over-represented canonical pathways in control and diseased human samples. Significance of association between these genes and the canonical pathway was measured in two ways as follows: (a) ratio of the number of genes from the data set that map to the pathway divided by total number of genes that map to the canonical pathway is displayed, and (b) Fischer's exact test was used to calculate a p value determining the probability that the association between the genes and the canonical pathway is explained by chance alone. Functional analysis of a network identified the biological functions that were most significant to the genes in the network. The network gene-associated biological function/disease state in the Ingenuity knowledge base was considered for analysis. The network score is based on the hyper-geometric distribution, is calculated with the right-tailed Fisher's Exact Test, and is represented as a negative log of this p value. For example, a network of 35 molecules has a Fisher's Exact Test p value of 1 ϫ 10 Ϫ6 , the network score ϭ Ϫlog(p value) ϭ 6 (19).
Cardiac Microarray Expression Data Base-To test for changes in expression in the target gene sets, microarray data from cardio-genomics data base for normal and idiopathic heart failure were used. The raw data were reprocessed using the GCRMA algorithm, which is a three-step function imple-mented in the GCRMA package (version 2.8.1) of the Bio-conductor open source library (version 2.5.0). Steps include correction of perfect match probe set expression signals for optical noise and nonspecific binding using probe sequence information, followed by quantile normalization to smooth individual probes intensities. Finally, expression values were summarized by the robust multichip model fit using median polish. The summarized probe set expression values were subsequently fit to a linear correlation analysis model.
Western Immunoblotting on Patient Samples-End-stage human heart failure samples (100 mg) were homogenized using Polytron homogenizer in 1.5 ml of lysis buffer (1% Nonidet P-40, 10% glycerol, 137 mM NaCl, 20 mM Tris-Cl (pH 7.4), 1 mM phenylmethylsulfonyl fluoride, 20 mM NaF, 1 mM sodium pyrophosphate, 1 mM sodium orthovanadate, and 2 g/ml each of aprotinin and leupeptin) and centrifuged at 38,000 ϫ g for 25 min. Transverse Aortic Constriction (TAC)-The TAC was carried out on mice as described previously (20). Briefly, mice were anesthetized with a mixture of ketamine and xylazine. After endotracheal intubation, mice were connected to a rodent ventilator. Using microsurgical procedures, the chest cavity was entered in the second intercostal space, and the transverse aorta between the right (proximal) and left (distal) carotid arteries was isolated. TAC was performed by tying a 7-0 nylon suture ligature against a 27-gauge needle, the latter being promptly removed to induce pressure overload cardiac hypertrophy. After aortic constriction, the chest was closed; the pneumothorax was evacuated, and the mice were extubated and allowed to recover from the anesthesia. Sham-operated animals underwent the same operation without aortic constriction. After 0, 1-4, 7, and 12 days of aortic constriction, mice were anesthetized, and hearts were rapidly excised. The individual chambers were separated and frozen in liquid N 2 for miRNA and biochemical analysis.

RESULTS
Patient Population-To get a global expression pattern for the miRNAs in end-stage heart failure, mRNA was isolated from 70 patient samples (nonfailing (20) and end-stage heart failure (50) with the diagnosis of DCM). The patient characteristics are summarized in Table 1. All failing hearts had left ventricular ejection fractions Ͻ15%, and all the nonfailing had ejection fraction Ͼ61% (Table 1). Mean age for the patients with DCM was 51 Ϯ 2 years and for nonfailing was 54 Ϯ 1 years. The average age of the patients was 52.5 Ϯ 3 years and was gender-and race-independent. They all had relatively normal ventricular function as measured by echocardiography with no associated visible characteristic dysfunction as measured by echocardiography. Furthermore, a majority of the patients were receiving some form of inotropic and/or vasopressor support (Table 1).
End-stage Dilated Cardiomyopathy Has a Specific miRNA Signature-To test whether end-stage DCM is associated with a distinctive miRNA signature, we followed the experimental plan described in Fig. 1. RNA from 10 nonfailing and 30 endstage DCM patient samples were used to hybridize with the miRNA microarray. RNA from each patient was used as an independent sample, and therefore the miRNA microarray data set obtained is a unique expression profile for each individual patient sample. We have used a custom microarray platform containing 5760 miRNA probes (including 627 probes for 288 human miRNAs) that is well established and validated by previous studies (15) to evaluate miRNA expression profiles in end-stage dilated cardiomyopathy. To identify miRNAs differentially expressed in the failing versus the nonfailing hearts, the data from quantitative expression was normalized using global median and Lowess normalization methods (see under "Experimental Procedures"). Following these iterative processes of normalization, expression of 9 miRNAs of 288 different mi-RNAs was found to be significantly different between nonfailing and DCM samples. The miRNAs hsa-mir-001 (p Ͻ 0.00005), hsa-mir-29b (p Ͻ 0.0087), hsa-mir-007 (p Ͻ 0.0086), and hsa-mir-378 (p Ͻ 0.0055) were significantly down-regulated in the DCM samples compared with nonfailing controls. In contrast, miRNAs hsa-mir-214 (p Ͻ 0.0001), hsa-mir-342 (p Ͻ 0.0004), hsa-mir-145 (p Ͻ 0.009), hsa-mir-125b (p Ͻ 0.078), and hsa-mir-181b (p Ͻ 0.0047) were significantly upregulated in DCM compared with nonfailing controls. The expression values of these nine miRNA probes for all individual patient samples are shown in the heat map (Fig. 2). Despite the individual patient variability in each of the miRNA probes, the The abbreviations used for nonfailing hearts are as follows: F, female; M, male; W, white; B, black; EF, left ventricular ejection fraction measured prior to explant; CVA, cerebrovascular accident; MVA, motor vehicle accident; GSW, gunshot wound. Drug therapy acute indicates treatment in the emergency room or intensive care unit prior to brain death: NE, norepinephrine (n ϭ 9); DA, dopamine (n ϭ 12); other, epinephrine, pitressin, phenylephrine, labetolol, lisinopril (n ϭ 1 or 2). Drug therapy chronic, indicates drugs taken by patients prior to admission, as reported by family members (n ϭ 6). The abbreviations used for failing hearts are as follows: F, female; M, male; W, white; B, black; EF, left ventricular ejection fraction measured prior to explant; DCM, dilated cardiomyopathy (pre-transplant diagnosis). Drug therapy lists those drugs taken by over 25% of patients in the group as follows: DIG, digoxin; DOB, dobutamine; AMIO, amiodarone; ACEI, angiotensin-converting enzyme inhibitor (usually lisinopril); BB, ␤ -adrenergic blocker (metoprolol or carvedilol).
heat map clearly demonstrates a distinctive pattern for specific miRNA expression associated with DCM.
To test whether the same set of miRNAs can be validated, we used a new independent set of patient samples (10 nonfailing and 20 DCM). To validate these miRNAs, RNA from a new set of samples was isolated and subjected to RT-PCR using specific primers for the set of miRNAs. As internal control we used both 18 S and U6. Our studies found that U6 gave more consistent results compared with 18 S, and therefore the miRNA data has been normalized to U6 values. RT-PCR from the new set of human patient samples showed that the differential expression of all the identified miRNAs could be validated except hsa-mir-145. To further confirm these results, we also carried out RT-PCR on the RNA from the samples that were first used for the miRNA microarray. Consistently, these studies also showed that differential expression of miRNAs could be validated with the exception of hsa-mir-145. The RT-PCR data from the two independent sets of human samples consistently showed the same results, and therefore it was appropriate to pool the data (20 nonfailing and 50 dilated cardiomyopathy human patient samples) from the two analyses (Fig. 3). The results from the pooled data show significant down-regulation of hsa-miRNA 1 (p Ͻ 0.00001), 29b (p Ͻ 0.0002), 7 (p Ͻ 0.00007), 378 (p Ͻ 0.001), and up-regulation of hsa-miRNA 214 (p Ͻ 0.0005), 342 (p Ͻ 0.003), 125b (p Ͻ 0.01), and 181b (p Ͻ 0.005) validating the microarray studies (Fig. 3). Critically, hsa-miRNA 145 showed a trend toward higher expression in DCM but did reach a level of significance compared with nonfailing samples (Fig.  3g). Taken together, our data demonstrate that end-stage DCM has a specific miRNA signature that can be consistently revalidated in a large number of samples. Importantly, our studies have identified novel miRNAs, hsa-miRNA 7 and hsa-miRNA 378 which are significantly down-regulated in DCM, that may have critical roles in the pathophysiology of heart failure.
Altered miRNAs Are Significantly Associated with Specific Canonical and Functional Pathways-Since alterations in these miRNAs occur simultaneously and modulate their respective targets, the global effect would be a sum total of effects coordinated by individual miRNAs. A bioinformatic view of the global effects potentially mediated by the eight altered miRNAs on signaling in dilated cardiomyopathy is shown in supplemental Fig. 1. Analysis shows that a total of 1785 genes are predicted targets for the eight differentially expressed miRNAs in DCM. To evaluate potential functional consequences based on combinatorial effects of the predicted targets of the miRNAs, an unbiased computational approach (see "Experimental Procedures") was taken using Ingenuity Pathway Knowledge base. Cardiovascular system development and function was the top canonical functional network with the highest level of significance (score 39). In comparison, significance score for nearest functional networks, cell development (score 35), cellular assembly (score 35), and cell death (score 33) was lower. Identification of this network function as potentially affected is entirely consistent with a role of the predicated targets of the altered miRNAs in cardiovascular disease. Cellular and molecular functions influenced by these network-eligible molecules, cell signaling (368 molecules, p Ͻ 5.91e-146), gene  expression (254 molecules, p Ͻ 1.67e-126), and cell death (245 molecules, p Ͻ 4.55e-93), are highly significant and consistent with the documented dysregulation of these events and pathways in DCM.
Predicted Targets of Altered miRNAs Associate with Diverse Signaling Networks-Of a total of 1785 predicated targets, 1716 could be mapped to signaling networks (see under "Experimental Procedures" for definition of network) in the IPA TM , and 995 predicated targets were found to be network-eligible (see "Experimental Procedures" for definition of eligibility). The 995 network-eligible candidates mapped to 43 networks that are predicted to be involved in the cross-talk with the peripheral molecules bridging different networks (supplemental Fig. 1) resulting in the phenotype.
A representative network with NFB, a known mediator in cardiac dysfunction (21) as a central node, is shown in Fig. 4 wherein the members that network with NFB are targets for the miRNAs 1, 29b, 125b, 181b 214, 342, and 378. As individual miRNA acts on each target, the net effect on the node would be the collective influence of all the members connected to the central node, NFB (Fig. 4). Based on this consideration, we predict that the complete NFB regulatory signaling network (Fig. 4) would be significantly down-regulated in DCM since molecules in this network are predicted targets for up-regulated miRNAs 125b, 181b, 214, and 243. Although Fig. 4 predicts one specific network with NFB as a central node for various miRNA, each network does not affect a physiological process in isolation, and global regulation would involve integrative cross-talk among the networks to mediate the disease phenotype. To test for such a cross-talk between networks, we have used Ingenuity Pathways Analysis algorithm for overlaying and merging networking (supplemental Fig. 1). The analysis of networks involved in top molecular and cellular functions associated with DCM showed that out of the 75 networks, only 43 are predicted to be involved in the crosstalk with the peripheral molecules bridging different networks (supplemental Fig. 1). A representation of the merged networks (supplemental Fig. 2) predicts that the pathways are specific for DCM; 32 networks (supplemental Fig. 1) were not incorporated into the network merge, suggesting they are not involved in cross-talk. Taken together this iterative analysis indicates that a specific set of pathways are operational, which are associated with integrative connecting networks resulting in the manifestation of DCM phenotype.
We next analyzed whether nodes (see "Experimental Procedures") of the various networks, which are interconnected and over-represented in DCM, are predicted targets for the differentially expressed miRNAs in end-stage heart failure. IPA-predicted nodal molecules on the various merged networks were analyzed for miRNA targets using target prediction algorithms (see "Experimental Procedures"). This analysis shows that a significant number of nodal molecules are predicted targets of miRNA that are altered in DCM (Table 2). Indeed, several of the FIGURE 3. Semi-quantitative real time RT-PCR validation of the altered miRNAs identified by miRNA microarray hybridization. a, cDNA synthesized from nonfailing (NF) and DCM samples were subject to real time PCR using miRNA 1-specific primers along with U6 as internal control. Data are presented as fold over normalized internal U6 control. Similar method as described above was used with specific primers for miRNA 29b (b), 007 (c), 372 (d), 214 (e), 342 (f), 145 (g), 125b (h), and 181b (i), and data are presented following normalization with U6 control. *, p Ͻ 0.01. predicted molecules are potential targets for two or three mi-RNAs, and some are targets for none. We analyzed whether these nodal molecules are altered in end-stage human dilated cardiomyopathy using the cardio-genomics expression data base (see "Experimental Procedures"). Many of the nodal molecules are significantly altered in end-stage heart failure (Table  3) consistent with prediction, thus relating the regulation of predicted targets by altered miRNAs in dilated cardiomyopathy. We have mapped the changes in the expression pattern of the nodal molecules to changes in miRNAs (Table 3, green and red represent down-and up-regulation, respectively). Indeed, the data show reverse complementary alterations in the levels of many of the predicted targets compared with our validated miRNA expression profile, e.g. MLL and STAT3 (Table 3). Interestingly, we also see parallel alterations for some nodal molecules and miRNAs, e.g. MMPs and TIMP2 (Table 3), and in some cases no effect on the nodal molecules, e.g. RB1 and EZH2 ( Table 3), suggesting that the miRNA is a component of the complex regulation of signaling during pathophysiology of heart failure.
Analysis and Validation of the miRNA Targets and Network-To directly evaluate whether nodal molecules are potential targets for altered miRNAs, we carried out Western immunoblotting studies in our collection of end-stage human heart failure samples. If the nodal molecules are targets for miRNAs, we would expect an inverse correlation between expression levels of nodal molecule to altered miRNA. Indeed Western analysis of some of the nodal molecules well known to have a role in heart failure shows an inverse co-relationship with the level of respective miRNAs (Fig. 5) and is very consistent with the data presented in Table 3. Western immunoblotting shows that ERBB2, HDAC4, COL1, MMP2, and TIMP2 are significantly up-regulated in human DCM (Fig. 5) and are inversely correlated to the down-regulation of their respective miRNAs. In contrast, STAT3 and E2F3 are down-regulated, consistent with the observation of up-regulation of their respective miRNAs suggesting that expression of these molecules may be regulated by miRNAs. Interestingly, we did not observe changes in expression levels of RB1 or EZH2 (Fig. 5) despite being predicted targets for miRNAs consistent with our data in Table 3.
An important question arising from these studies is whether these miRNAs are the "cause" or an "effect" of heart failure. Since it is not possible to identify alterations in miRNA profile with initiation of cardiac dysfunction in humans, we have used a well known surgical mouse model of transverse aortic constriction (TAC) (20) to analyze alterations in eight miRNAs (identified in end-stage heart failure) with initiation of cardiac dysfunction. Mice (C57Bl/6) underwent TAC surgery, and the experiments were terminated on 0, 1-4, 7, and 12 days post-TAC. RNA isolated from the mouse hearts was subjected to RT-PCR to evaluate the alteration in the eight identified mi-RNAs. RT-PCR analysis showed no specific trends within 7 days of TAC except for miRNA 7, which consistently showed  down-regulation and reached significance by 7 days (Fig. 6a). Importantly, 12 days post-TAC (a time point known from our previous studies to be characterized by initial cardiac dysfunction associated with cardiac hypertrophy (20)), significant alteration in a subset of four miRNAs (miRNA 7, 378, 214, and 181b) is observed (Fig. 6a) out of the eight altered miRNAs identified in end-stage heart failure (Fig. 3). miRNA 7 and 378 are significantly down-regulated, and miRNA 214 and 181b are significantly up-regulated (Fig. 6a). These studies demonstrated that with initiation of cardiac dysfunction, a subset of four miRNAs (out of eight miRNAs identified at end-stage heart failure) are altered and may potentially dysregulate signaling networks, contributing to pathology.
To dynamically perturb miRNAs to validate predicated targets and signaling networks, we generated constructs for expressing miRNA or its respective antagomir in cell systems. We characterized the expression of miRNA 7 and validated some of its targets as verification for miRNA expressing constructs. hsa-miRNA 7 expression constructs were transfected into HEK cells, and expression was confirmed by confocal microscopy (Fig. 6b) and Northern blot analysis (Fig. 6c) followed by Western immunoblotting for ERBB2, COL1, and RB1, which are the nodal molecules (Table 3) and predicted targets for miRNA 7. Western blot analysis shows that overexpression of miRNA 7 results in a significant decrease in the expression of ERBB2 and COL1 (Fig. 6d). In contrast, no appreciable change was observed in RB1 levels following miRNA 7 expression (Fig.  6d) suggesting that RB1 may not be an actual target. These data are very consistent with our findings in Table 3 depicting no changes in human RB1 levels despite significant changes in miRNA levels. Taken together these data provide evidence that miRNAs may play an integral role in regulation of nodal molecules in human DCM, and dysregulation of miRNAs would be an important component in cardiac dysfunction.
To validate whether a combination of miRNAs/antagomirs will regulate the NFB signaling network (depicted in Fig. 4), we assessed the predicted targets of the four altered miRNAs (Fig.  6a) in the NFB signaling network as alteration of the targets due to miRNAs may contribute toward imbalance in the signaling network with cardiac stress. NFB signaling network has 35 molecules, including NFB. Analysis of the predicted targets indicated that NFB signaling network has 12 molecules that are targets for miRNA 378, 214, and 181b. Expression analysis assessed using the cardio-genomic data base for dilated cardiomyopathy showed that 4 of the 12 predicted targets were altered in end-stage heart failure. HDGF and BCL2 are predicted targets for miRNA 214, and CLCF1 and SLC2A are predicted targets for miRNA 378 indicating that these molecules may feed into the network with initiation of cardiac stress resulting in modulation of NFB, which is not a target of either miRNA 214 or 378. Although CLCF1 is a predicted target of miRNA 378, its expression pattern did not inversely correlate with miRNA expression profile indicating that it may be regulated by mechanism(s) independent of miRNA 378. Furthermore, none of the predicated targets for miRNA 181b were altered in the NFB network.
To test whether NFB is altered due to cross-talk in the signaling network, we transfected HEK cells with constructs expressing either miRNA 214 and 378 or miRNA 214 and antagomir of miRNA 378. Seventy two hours following transfection, cells were lysed, and Western immunoblotting was carried out. Immunoblotting showed an appreciable reduction of NFB levels in cells expressing miRNA 214 along with antagomir for 378 compared with cells expressing miRNA 214 and 378 (Fig. 7a). To further identify which of the molecules feed into the alteration in NFB, immunoblotting was carried out for HDGF, BCL2, and SLC2A, which are predicted targets for miRNA 214 and 378. Immunoblotting for HDGF and BCL2 did not show any appreciable changes in cells expressing miRNA 214 and 378 or miRNA 214 and antagomir 378 compared with the cells transfected with scrambled control (Fig.  7a). Interestingly, marked up-regulation of SLC2A was FIGURE 5. a, Western immunoblotting analysis was carried out on nonfailing human hearts (CON, controls; n ϭ 8), and hearts from patients diagnosed with DCM (n ϭ 8). 170 g of myocardial lysate was resolved with SDS-polyacrylamide gel and immunoblotted (IB) with respective antibodies. The blots were stripped and re-probed multiple times with various antibodies, including ␤-actin antibody that was used to ensure equal loading. b, densitometric analysis in the DCM samples is represented as fold over nonfailing controls. *, p Ͻ 0.001 control versus DCM.
observed in cells expressing miRNA 214 and antagomir 378 compared with cells expressing miRNA 214 and 378 (Fig. 7a) suggesting that NFB may be regulated by input in from SLC2A following perturbation of miRNA 378. To test whether such regulation exists in the NFB network in hearts, we used 12-day-old TAC mice, which are characterized by up-regulation of miRNA 214 and down-regulation of miRNA 378. The ventricular lysates from 0-day (sham) and 12-day-old TAC were resolved on SDS-polyacrylamide gel and immunoblotted for NFB. Marked reduction in NFB was observed in the 12-day-old TAC mice compared with the control sham (Fig.  7b). There were no changes in BCL2 levels ( Fig. 7b) but significant up-regulation in HDGF was observed following 12 days of TAC. The up-regulation of HDGF in TAC seems to be independent of miRNA 214 because HDGF is a predicted target for miRNA 214 and miRNA 214 is up-regulated post-TAC (Fig.   6a). Importantly, SLC2A is significantly up-regulated in TAC mice compared with the controls consistent with our results of down-regulation in miRNA 378 (Fig. 6a) following TAC. Taken together, these studies demonstrate that miRNAs do mediate regulation of signaling network, and more specifically the NFB as shown in our studies may be modulated by single or multiple target proteins. In our studies, both in the heart as well as in transfected cells, miRNA 378 perturbation seems to modulate the nodal molecule, NFB, by directly altering peripheral molecules that feed into the NFB network as predicted by IPA network analysis.

DISCUSSION
The most important finding from our studies is identification of a unique miRNA fingerprint in human dilated cardiomyopathy empowering us to define the global molecular signaling FIGURE 6. a, C57BL6 mice underwent transverse aortic constriction, and the animals were sacrificed on days 0 (sham), 1-4, 7, and 12 post-TAC. This well established model of heart failure is known to result in cardiac dysfunction by 12 days of TAC and is associated with significant cardiac hypertrophy. RNA was isolated from the excised hearts and subjected to quantitative RT-PCR for the altered miRNAs identified in our study. b, phase contrast and confocal microscopy showing expression of miRNA7 transfected in HEK 293 cells. miRNA 7 encoding plasmid also co-expresses green fluorescence protein to visualize transfection. c, to directly test for expression of miRNA7, Northern blot analysis was carried out. Con represents vector-transfected cells; miRNA-7 represents cells expressing miRNA 7. Lower panel shows methylene blue staining of the nylon membrane depicting equal loading of RNA for each of the samples for Northern analysis. d, miRNA 7 expressing cells were lysed, and Western immunoblotting was carried out for nodal molecules (shown in Table 3). Interestingly, overexpression of miRNA 7 results in significant down-regulation of ERBB2 (epidermal growth factor receptor 2) and Col1 (collagen 1). In contrast, we do not see appreciable changes in RB1 protein levels. ␤-Actin immunoblotting (IB) was used as loading control. Tnsf, transfection. *, p Ͻ 0.01. network regulated by miRNAs in cardiac pathology. Our comprehensive custom microarray platform along with the use of a relatively large set of human samples lead to identification of eight differentially expressed miRNAs in dilated cardiomyopathy compared with nonfailing human hearts. We report here for the first time two novel miRNAs that are significantly downregulated in dilated cardiomyopathy and have never been implicated in cardiac pathophysiology. The signaling networks and functional pathways formed by the predicted targets of these altered miRNAs enabled unbiased prediction of molecular cardiovascular disease bio-function in understanding heart failure biology within the Ingenuity TM knowledge compendium. The unique miRNA signature leads us to identify nodal molecules on the global signaling networks that are potential targets for miRNAs laying the foundation for us to propose for the first time the global regulatory role of miRNAs in modulating molecular networks. Furthermore, we have validated the expression pattern of nodal molecules through a meta-analysis of heart failure gene expression fingerprint from the publicly available cardio-genomics data base. Nodal molecules occupy key positions in the signaling networks, and alterations of these by combinatorial action of miRNAs could account for the observed global changes associated with cardiac dysfunction and failure. We have further used TAC mouse model of heart failure and transfected cells to demonstrate the regulation of NFB nodal molecules and the inputs into this signaling network.
Our miRNA microarray and RT-PCR validation on independent sets of human cardiac samples consistently identified eight miRNAs to be differentially expressed in human dilated cardiomyopathy. Importantly, along with identification of miRNAs that have been previously reported in mouse studies and human tissues (6,22,23), this study has identified two novel miRNAs (miRNA 7 and 378), which are differentially expressed in dilated cardiomyopathy, that until now have never been implicated in cardiac diseases. Identification of novel miRNAs has been possible in our study due to the use of a comprehensive set of miRNA probes for microarray hybridization. We have also identified two other miRNAs (miRNA 29b down-regulated and 181b up-regulated in dilated cardiomyopathy) to be differentially expressed significantly compared with the recent study by Ikeda et al. (22). Although the trends of differential expression for miRNA 29b and 181b were similar to our study, it did not reach the level of significance probably due to the relatively small sample size in the study (21). A critical question that remained following identification of the eight unique altered miRNAs in end-stage human heart failure was whether miRNAs are a cause or an effect of heart failure as the heart is known to remodel to stress. To address this question, we used the well established TAC mouse model to test for alteration in miRNAs with initiation of cardiac dysfunction. Our studies show that only a subset of four miRNAs (miRNAs 7, 378, 214, and 181b) are altered with initiation of cardiac dysfunction out of the eight miRNAs identified at the end-stage heart failure suggesting that other miRNAs identified in the study may be an effect of heart failure. In-depth studies are required to determine the role of remodeling in altering the miRNA profiles.
The two novel miRNAs (Ϫ7 and Ϫ378) identified in this study are down-regulated in end-stage dilated cardiomyopathy indicating that their respective target molecules would be upregulated in disease states. Consistently, Western blot analysis from the DCM samples shows significant up-regulation of specific nodal molecules like ERBB2 and Col1A without changes in RB1 suggesting that ERBB2 (epidermal growth factor receptor 2) and COL1A (Collagen 1) are targets for miRNA 7. Indeed, overexpression of miRNA 7 in cells shows significant downregulation of ERBB2 and COL1 without appreciable changes in Top panel shows immunoblotting for NFB nodal molecule. NFB densitometric analysis is shown below. Lower panel show immunoblotting for network molecules HDGF and BCL2 (predicted targets for miRNA 214) and SLC2A (predicted target for miRNA 378). Actin was used as loading control. b, since miRNA 214 is up-regulated and miRNA 378 is down-regulated following 12 days of TAC, Western immunoblotting was carried out for nodal molecule (NFB) and other network molecules (HDGF, BCL2, and SLC2A). *, p Ͻ 0.001, vector (control) versus cells expressing miRNA 214 and antagomir 378.
RB1. Importantly, analysis of some of the predicated targets of miRNA 7 like phospholipase C␤, regulator of G-protein signaling, RAF1, and phosphodiesterase 4, which are well studied in heart failure (24 -26), are all up-regulated in cardiac hypertrophy and heart failure (24 -26) suggesting an important role for miRNA 7. Studies on neuronal development have shown that miRNA 7 targets basic helix-loop-helix and Brd box (27,28) containing transcriptional repressor proteins that interpret Notch signaling. Whether such regulation exists in the cardiac system needs further elucidation. Similarly, transforming growth factor ␤, v-SRC, and platelet-derived growth factor receptor are some of the potential targets of miRNA 378. These molecules (24 -26) are also dysregulated in cardiac hypertrophy and heart failure indicating potential regulation by miRNA 378. Recent studies in cancer have shown that expression of miRNA 378 enhances cell survival and promotes tumor growth and angiogenesis (29). What effect the down-regulation of these novel miRNA 7 and 378 will have in the pathology of heart failure is currently an active area of our study.
The miRNA-mediated regulation of gene expression within the myocardium is perhaps the most intriguing and novel mechanism in heart failure biology. Although several studies have begun focusing on miRNA in heart failure, genetic changes and miRNA-mediated critical transitions in progression of heart failure remain undefined. The miRNAs regulate gene expression primarily through repression, and the sets of expressed miRNAs combinatorially delineate precisely the profound system-wide influence on gene expression programs. Thus, combinatorial expression patterns of miRNAs associated with DCM provide the mechanistic basis for reading out cooperative action of miRNAs on gene expression in human heart failure. Traditional molecular profiling analyses have focused on identifying individual genes deregulated during the disease process, but fail to show little overlap between individual genes and processes driving heart failure. Therefore, to move beyond the single miRNA or single gene target approach, we analyzed our target scan signatures using IPA, which allows for identification of enriched networks. The "targets of altered miRNA in heart failure" uncovered the GO process containing cardiovascular development and reprogramming. Further analysis identified networks of 1, 2, 31, and 32 to name a few as potential targets of miRNA regulation. One of these networks, network 1, was the most enriched for genes with defined NFB binding sites in their proximal promoters. These results suggest that a major process affected by miRNA regulation in network 1 associated with heart failure progression is probably through NFB target genes. This is consistent with a growing body of evidence showing that inhibition of NFB significantly ameliorates cardiac dysfunction by reducing pro-inflammatory responses (30,31).
Recent studies have provided compelling evidence of involvement of miRNA in the regulation of hypertrophy and heart failure (6,22,23). Critically, miRNAs target and regulate many gene products by the simple method of sequence hybridization and therefore bring about nonconventional global regulation of signaling networks leading to physiological responses. Indeed, recent studies using cardiac specific deletion of dicer support the idea that miRNAs may have a critical role in globally regulating cardiac signaling and function as deletion of dicer leads to significant reduction in mature miRNAs resulting in DCM and heart failure (13). Since the knowledge of underlying networks and canonical function pathways regulated by the miRNAs in conditions of heart failure is not known, we have examined potential regulation of specific signaling networks by differentially expressed miRNA in heart failure. Analysis of all differentially expressed miRNAs using the Ingenuity Pathways Network algorithm showed that the predicted targets are associated with the networks of cardiovascular development and function. Importantly, this association demonstrates by an independent "in silico" methodology that these unrelated predicted targets for differentially expressed miRNAs are critical signaling components in the cardiovascular system and provide an independent corroboration of our findings. This corroboration is independently supported by Western blotting analysis on human samples showing a tight correlation between the expression patterns of the predicted target nodal molecules to the alterations in their respective miRNAs. These studies suggest that miRNAs could regulate unrelated signal transduction networks. Therefore, it is important to determine the regulation of signaling networks by miRNAs to provide an understanding of the global scale of regulation by miRNAs instead of looking at individual targets. To demonstrate that miRNAs do regulate the signaling networks, we have analyzed the predicted targets of miRNAs altered with initiation of cardiac dysfunction on the NFB signaling network. Our studies for the first time show that perturbation in the miRNAs leads to modulation in the network by a feed-in mechanism. In our studies we show by using transfected cells and mice that nodal molecule NFB, which is not a predicted target for miRNA 378, is altered following changes in miRNA 378. These studies indicate that mi-RNAs will have a prominent role to play in altering global signaling networks and pathways in progression toward cardiac hypertrophy and failure. Interestingly, data for our TAC mice indicates that HDGF is up-regulated and therefore may not be regulated by miRNA 214. In the event that HDGF was regulated by miRNA 214, we would have seen a marked inhibition in HDGF levels inversely complementing the miRNA levels. As this does not occur, we believe that it may not be playing a role in the miRNA-mediated regulation of nodal molecule. Although HDGF may not be regulating NFB by an miRNAdependent mechanism, we cannot exclude the possibility that HDGF regulates NFB independent of miRNA, which is not the focus of this study.
Despite the elucidation of several clinically relevant signal transduction pathways that can lead to heart disease progression, the means by which these pathways are coordinated with respect to the development of cardiac dysfunction remain obscure. Manifestation of the phenotype of dilated cardiomyopathy is the net result of traditional cross-talk between the molecules and the nontraditional regulation by miRNAs. Inclusion of all the predicted targets for these altered miRNAs showed that they represent 75 annotated networks out of which 44 networks could be merged to cross-talk leading to global regulation of signaling. The cross-talk between networks happens via peripheral interconnecting molecules that are represented in between the networks (supplemental Fig. 1). Analysis of this kind provides a global understanding of signaling networks and sheds light on the atypical regulation of molecules by miRNAs resulting in a specific phenotype that cannot be explained by currently accepted signaling mechanisms. One of the major difficulties for functional studies of miRNAs is in determining their specific target genes as available algorithms predict hundreds of target genes for any single miRNA with a likely high fraction of false-positives. Despite these deficiencies, use of all the predicted targets for the differentially expressed miRNAs with the network analysis algorithm shows that these molecules regulate cardiovascular system and cell signaling.
Taken together our studies show that eight miRNAs are significantly altered in DCM compared with nonfailing controls. Importantly, we have identified two novel miRNAs that are down-regulated in end-stage heart failure. Our study in cell systems and mice indicates that miRNAs may play a pivotal role in altering global signaling networks during progression of cardiac pathology. Our analysis in this study provides the following: (a) a better understanding of the global regulation of signaling network pathways by miRNAs; (b) a reduced number of predicted targets, and (c) evidence that a single gene could be targeted by more than one miRNA, ultimately resulting in regulation that is a net effect of relative abundance of independent miRNAs. Importantly, our study lays a foundation for the concept that therapeutic interventions with miRNAs would have profound global effects on signaling networks in a nonconventional manner compared with the conventional feedback mechanism of regulation.