Effects of stretch and shortening on gene expression in intact myocardium

Abstract

Multiple cues have been suggested as the mechanical stimulus for the heart's hypertrophic response. Our work has previously suggested that the amount of cyclic shortening in cardiomyocytes controls myocyte shape and the amount of stretch controls myocyte size. To identify gene expression changes that occur in response to these mechanical perturbations, we used microarray analysis of papillary muscles cultured for 12 h at physiological or reduced levels of cyclic shortening and physiological or reduced mean stretch. Overall, genes related to extracellular matrix (ECM) were surprisingly prominent in our analysis. Connective tissue growth factor was among a small group of genes regulated by the amount of cyclic shortening regardless of the level of mean stretch, and many more ECM genes were regulated by shortening with reduced amounts of stretch. When we compared our results to gene expression data from an in vivo model of pressure overload (PO), which also decreases myocyte shortening, we found the genes that were commonly regulated in PO and our decreased shortening groups were most significantly enriched for ontology terms related to the ECM, followed by genes associated with mechanosensing and the cytoskeleton. The list of genes regulated in PO and our decreased shortening groups also includes genes known to change early in hypertrophy, such as myosin heavy chain 7, brain natriuretic peptide, and myosin binding protein C. We conclude that in intact myocardium, the amount of cyclic shortening may be an important regulator not only of myocyte genes classically associated with hypertrophy but also of ECM genes.

hemodynamic overload in the heart arises in many clinical circumstances and leads to varying degrees and patterns of hypertrophic growth and remodeling. The growth and remodeling response to chronic hemodynamic overload, particularly changes in cardiomyocyte shape, markedly affects both the likelihood and rate of progression through stages of deteriorating function into heart failure. In patients with pressure overload (PO) states such as chronic hypertension, individual cardiomyocytes become wider and the left ventricular wall becomes thicker. These cell shape changes, referred to as concentric hypertrophy, create a relatively stable clinical prognosis. Heart failure develops only after decades of chronic hypertension in elderly patients ( 12). In volume overload (VO) states such as mitral regurgitation, myocytes become longer and the left ventricle dilates, a pattern termed eccentric hypertrophy. When VO occurs, there is a much shorter time until the development of heart failure. For example, 90% of patients demonstrate early signs of heart failure within 10 years of diagnosis of severe mitral regurgitation ( 26). Understanding how cell shape is regulated could therefore enable new therapeutic approaches to control hypertrophy and prevent or reverse heart failure.

Although a host of potential mechanical and biochemical stimuli have been proposed and studied over the past century, it has been experimentally challenging to definitively identify the mechanisms directly responsible for cell shape regulation. In vitro cell stretching experiments allow precise manipulation of mechanical and chemical inputs, yet this experimental system is unphysiological for myocytes: the cells are plated in two dimensions on artificial substrates, often devoid of appropriate connections to the extracellular matrix (ECM), neighboring myocytes, and cardiac fibroblasts, all of which have been shown to modulate cardiac hypertrophy ( 6, 33). In vivo experiments are confounded by neuroendocrine signaling. Hormones such as angiotensin II and norepinephrine are known to induce hypertrophy without mechanical perturbation; efforts to pharmacologically block these responses to isolate mechanical from chemical effects on hypertrophy during in vivo overload have produced few clear insights. In vitro culture of papillary muscles offers a convenient middle ground between cell culture and in vivo preparations, allowing excellent control of mechanical and biochemical inputs while maintaining cell-cell and cell-ECM connections ( 14).

In seeking mechanical signals that might trigger changes in cardiomyocyte shape, we began by comparing changes in a long list of candidate mechanical stimuli during experimental VO and PO ( 17). One clear difference between these two states was their effect on stroke volume, ejection fraction, and related measures such as regional systolic strain and shortening. In PO, stroke volume, ejection fraction, and regional shortening decreased; in VO, they increased. We therefore hypothesized that the amount of cyclic shortening experienced by myocytes controls myocyte shape and demonstrated directly that myocytes from muscles cultured at a reduced level of systolic shortening to simulate PO get shorter and wider over 36 h compared with muscles cultured at physiological levels (15%) of shortening ( 14). The primary goal of the present study was therefore to identify candidate genes that may play a role in regulating myocyte shape by identifying genes whose expression is specifically responsive to changes in myocyte shortening.

In addition to changes in myocyte shape, myocyte size increases rapidly following the imposition of experimental PO or VO ( 17) and regresses rapidly following mechanical unloading with left ventricular assist devices (LVADs) ( 4). Based on the strong correlation between myocyte size and average chamber volume (or related local quantities such as myocyte stretch and sarcomere length) across these different states, we have hypothesized that average sarcomere length or myocyte stretch may trigger changes in myocyte mass. Therefore, in the present study we also sought to identify genes whose expression is specifically responsive to average myocyte stretch.

MATERIALS AND METHODS

Papillary muscle isolation.

All experiments were conducted in accordance with the Guide for the Care and Use of Laboratory Animals (19a) and approved by the University of Virginia's Institutional Animal Care and Use Committee. Thirty-two adult male LBN-F1 rats (Harlan, Indianapolis, IN), 275–300 g, were euthanized by 1 ml intraperitoneal injection of 50 mg/ml pentobarbital sodium (Ovation Pharmaceuticals, Deerfield, IL). Following intracardiac heparinization, the heart was rapidly dissected, transferred to a microdissection dish, and retrograde perfused via the ascending aortic stump with 4°C Krebs-Henseleit (K-H) Buffer Modified containing (in mM) 118 NaCl, 4.7 KCl, 1.2 MgSO ₄, 1.2 KH ₂PO ₄, 20 NaHCO ₃, 11 glucose, 0.25 CaCl ₂, 30 2,3-butanedione monoxime (BDM), and 20 IU/l insulin in equilibrium with 95% O ₂-5% CO ₂. Next, the right ventricle (RV) was opened by cutting along the posterior and superior edges of the free wall, followed by careful dissection of appropriately sized and shaped RV papillary muscles with the following dimensions (mean ± SD): major radius, 318 ± 71 μm; minor radius, 218 ± 47 μm; length (between the two mounting pins), 2.44 ± 0.57 mm. To excise each papillary muscle, we first dissected all chordae tendinae attachments and subsequently cut around the septal insertion of the muscle. This septal block was used to handle each muscle without damaging the tissue.

Muscle culture protocol.

We cultured each RV papillary muscle at 37°C for 12 h in a sterile culture system, modified from that described by Janssen et al. ( 22) by the addition of a programmable servomotor to control muscle length. RV papillary muscles were chosen because they are conveniently dissected and provide more mRNA per sample than endocardial trabeculae yet are thin enough to be adequately oxygenated by diffusion in the muscle culture system; in our previous studies, RV papillary muscles cultured in this system maintained developed force levels for at least 36 h, while size and sarcomere structure were preserved ( 14). Muscles were mounted at slack length, such that a 50 μm stretch yielded a measurable increase in passive force via pins connected to a force transducer at one end and the servomotor at the other. After mounting each muscle in chilled BDM K-H buffer, we applied sterile TiO ₂-PBS (1:1 wt/wt) to each of four corners of the muscle's central region, creating two longitudinal pairs of markers used to track muscle stretch and shortening throughout the culture experiment. After enclosing the baths, initiating superfusion of 95% O ₂-5% CO ₂ through each bath, and increasing the culture temperature to 37°C, we exchanged the BDM K-H buffer for BDM-free K-H buffer (0.25 mM Ca ²⁺) and then exchanged this for Medium 199 culture medium (M199, 1.85 mM Ca ²⁺; Mediatech, Manassas, VA) supplemented with the following: 21.7 mM NaHCO ₃, 2.0 mM l-carnitine, 4.4 mM creatine, 5.0 mM taurine, 2.0 mM l-glutamine, 20 IU/l insulin, 105 IU/ml penicillin, 100 μg/ml streptomycin, and 1 μg/ml amphotericin B. From this medium exchange throughout the duration of the culture period, muscles were electrically stimulated end to end by 5 ms asymmetric pulses at 1 Hz frequency, 30–50% above the contraction-stimulating threshold voltage (2–4 V). Following a 30–60 min equilibration period, each muscle was stretched by 0.05 mm increments every 60–90 s until the central region of the muscle reached a predetermined amount of stretch from slack (5 or 15%), which we confirmed by measuring the increased spread between both TiO ₂ marker pairs during diastole. Following this prestretch, we programmed the servomotors using one of four waveforms to impose a distinct combination of time-averaged stretch and cyclic shortening upon each cultured muscle throughout the 12 h culture period, as described below.

We used a customized data acquisition system (LabVIEW; National Instruments, Natick, MA) to continuously measure and record muscle force, control servomotor position, and periodically acquire video of muscle contraction-relaxation. From this video feed, we measured diastolic stretch (compared with T = 0 slack) and adapted a marker tracking algorithm to calculate time-averaged stretch and cyclic shortening between both longitudinal pairs of TiO ₂ markers. To preserve identical culture conditions for each sample, M199 and 95% O ₂-5% CO ₂ superfusate gas were exchanged only at the midpoint (6 h) of each experiment. Following measurement of force, stretch, and shortening at 12 h, we terminated the electrical stimulus and any length control protocol, relaxed each muscle in cold BDM K-H buffer for 2–3 min while trimming the ends and measuring muscle dimensions, and flash-froze each muscle in liquid nitrogen prior to storage at −80°C until subsequent RNA isolation.

Mechanical input design.

We employed a factorial design ( 31) as the framework for defining our four experimental groups: two mechanical factors (mean stretch and cyclic shortening) at two distinct levels that represent “normal” and “reduced” states (high and low, respectively). In preliminary studies, we observed that too little or too much stretch leads to muscle contracture. To maximize the difference between high and low values for both inputs, while minimizing both intragroup variability and muscle contracture, we defined high and low target values for time-averaged stretch to be 16 and 4% above slack, respectively. We similarly defined the target values for high and low cyclic shortening to be 16 and 4% of slack length. We then cultured six to nine muscles under each mechanical protocol described below ( n = 30 total) and selected the three muscles closest to each of the four target shortening/stretch combinations (16% stretch + 16% shortening, 16% stretch + 4% shortening, etc.) for subsequent analysis. To facilitate discussion, we labeled muscle groups according to the in vivo condition with the most similar mechanics, with an asterisk to clearly identify them as in vitro analogs: high-ΔL/high-L (physiological*) = physiological shortening, physiological stretch; low-ΔL/high-L (PO*) = reduced shortening; high-ΔL/low-L (LVAD*) = reduced stretch; low-ΔL/low-L (unloaded) = reduced shortening and reduced stretch ( Fig. 1 A).

Fig. 1.

Two-way mechanical input study design for 12 h culture of male LBN-F1 rat right ventricle (RV) papillary muscles. Servomotor-driven length-time inputs were varied to maintain physiological levels of mean stretch and shortening ( A, top right corner), to reduce shortening alone (PO*, simulating mechanics of in vivo PO), to reduce stretch alone (LVAD*, simulating mechanics of LVAD support), or to reduce stretch and shortening (unloaded). Target levels of stretch ( B) and shortening ( C) were maintained in each group throughout the 12 h culture period. PO, pressure overload; LVAD, left ventricle assist device.

Measurement and control of muscle mechanics.

We adapted a marker tracking algorithm developed for measuring strain in mechanically tested tissues ( 13) to track the TiO ₂ markers throughout each contractile cycle and quantify mean stretch and cyclic shortening in each muscle's central region. Because the central region of a contracting papillary muscle does not exactly follow its ends, this method allowed us to carefully adjust the amount of prescribed shortening so the muscle's central region experienced the desired amount of cyclic shortening. The unloaded group was prestretched 5% from slack length, while the other three groups were prestretched 16% from slack length. In the case of low-shortening muscles (PO* and unloaded groups), we prescribed up to 5% cyclic lengthening when necessary to counteract substantial shortening in the muscle's central region, especially at the beginning of the culture period; in the case of high-shortening muscles, we prescribed initial shortening slightly above the target mean value to account for typical decreases over the course of the experiment. To achieve different amounts of time-averaged stretch between the high-ΔL groups, the low-stretch (LVAD*) muscles were held at systolic length for a considerably longer duration than the high-stretch (physiological*) muscles (700 vs. 50 ms, respectively). At equal intervals throughout the experiment (0, 3, 6, 9, and 12 h) we measured muscle stretch and cyclic shortening. This allowed us to make frequent adjustments of length control parameters to achieve time-varying length target values.

RNA isolation, preparation, and microarray hybridization.

Three samples from each of the four mechanical input groups were removed from −80°C and partially submerged in a bath of liquid nitrogen within their own 1.5 ml tubes. Each frozen muscle was then pulverized with a liquid nitrogen-cooled pellet pestle (Kontes; Thermo Fisher Scientific, Waltham, MA) within the 1.5 ml tube. We immediately added 1 ml of Trizol (Invitrogen; Life Technologies, Carlsbad, CA) to the sample and carefully removed the pestle from the tube to be washed and rinsed in microfiltered, deionized H ₂O prior to processing the subsequent sample. We drew the Trizol-incubated sample into and out of a 3 ml syringe through a 25 gauge needle (BD, Franklin Lakes, NJ) three or four times to shear any genomic DNA. From this point forward, we followed the standard Trizol protocol, including addition of glycogen during the introduction of isopropyl alcohol. Finally, we resuspended the total RNA pellet in 9 μl of RNase-free H ₂O and warmed the sample for 10 min (60°C) to promote dissolution. We then treated each sample with DNase I (Invitrogen, Life Technologies) according to the manufacturer's protocol, measured sample concentration with a small volume spectrophotometer (NanoDrop; Thermo Fisher Scientific, Waltham, MA), and diluted the samples to a concentration of 33.3 ng/μl. Subsequent processing and microarray hybridization were performed by the Biomolecular Research Facility (School of Medicine, University of Virginia). Prior to microarray target preparation, each selected RNA sample was run on the Agilent Bioanalyzer 2100 with an RNA 6000 Nano Chip (Agilent, Santa Clara, CA) to assess sample quality and confirm total RNA and mRNA concentration. Global gene expression of all samples was assayed using GeneChip Rat Genome 230 2.0 arrays (Affymetrix, Santa Clara, CA). All biotin-labeled, amplified RNA for hybridization was prepared from 100 ng total RNA in 3 μl volume with the standard Affymetrix kit and protocol. Following hybridization, the Biomolecular Research Facility returned the raw expression files to us for bioinformatics preprocessing and analysis.

Microarray data analysis.

We processed the 12 CEL files by loading them and the appropriate Affymetrix CDF library annotation file into R using the Bioconductor package ( http://www.R-project.org, http://www.bioconductor.org) to process the raw data with the robust multichip average (RMA) algorithm ( 8, 20). To avoid the pitfalls associated with analysis of very low-signal probe-sets ( 9, 28), we removed probe-set data from all arrays if the RMA-normalized log ₂ signal of a probe-set was below log ₂(50) in at least 50% of the samples within each group. This signal-based filter reduced the number of probe-sets for subsequent analysis from 31,042 to 16,391. To maximize our ability to detect expression differences between groups, we used the local-pooled-error (LPE) method that has been specifically developed for high-throughput data analysis with a small number of replicates ( 21). Also available as a downloadable R package from Bioconductor, LPE estimates within-probe-set variance by pooling the variances of all probe-sets at similar expression intensities, across all samples within a given experimental group, so it has a shrinkage effect for the variance of each probe-set to a level consistent with its overall expression intensity. We computed LPE-based Z scores for each probe-set comparison to identify the transcriptional effects of reduced cyclic shortening under physiological mean stretch (physiological* vs. PO*) or subphysiological mean stretch (LVAD* vs. unloaded), and the effects of reduced mean stretch under physiological cyclic shortening (physiological* vs. LVAD*) or subphysiological cyclic shortening (PO* vs. unloaded). For each comparison, we used a false discovery rate (FDR) cutoff of 0.3 to identify the most significantly regulated probe-sets by controlling statistical random chances of multiple comparisons due to a large number of candidate genes. We then used DAVID (the Database for Annotation, Visualization, and Integrated Discovery) ( 18, 19) to perform biological annotation and functional analysis of the resulting lists of regulated genes. We used DAVID's “Functional Annotation Clustering” tool to identify statistically significant overrepresentation of groups of gene ontology (GO) terms associated with our significant gene lists.

Comparison to published microarray studies of in vivo PO and LVAD.

To compare transcriptional changes we identified in vitro to analogous in vivo and clinical results, we searched for published microarray studies comprising groups exposed to similar relative alterations in myocardial mechanics. First, we identified a study that assessed gene expression after 7 days of surgically induced transverse aortic constriction (TAC) in transgenic (Gata4 heterozygous) vs. wild-type adult male mice ( 7). We leveraged microarray data from the 10 wild-type mice in this study (6 TAC vs. 4 sham) as an analog to our reduced cyclic shortening comparisons. We also identified a study exploring the global transcriptional effects of LVAD treatment of failing human hearts ( 34). From this study, we utilized pre- and post-LVAD array data from 21 patients with dilated cardiomyopathy for comparison to our reduced mean stretch comparisons. In each case, we used the CEL files from the original studies and performed the same data processing, low-signal filtering, and analysis as for our cultured muscle groups.

PCR validation of microarray analysis.

To assess how well our microarray analysis agreed with an independent method, we conducted qPCR of a panel of 12 genes of interest to our laboratory and compared normalized PCR starting quantities to microarray expression levels. RNA was reverse-transcribed using the iScript cDNA synthesis kit using 150 ng (Bio-Rad, Hercules, CA). Primers and gene-specific probes were designed for each gene and the cardiac housekeeping gene TPT1 ( 31) using BeaconDesigner 2.06 (Premier Biosoft, Palo Alto, CA) and synthesized by Integrated DNA Technologies (Coralville, IA). For each primer set, at least one primer spanned an exon-exon boundary and specificity was confirmed by visualization of a single PCR product by gel electrophoresis. Real-time PCR conditions were optimized for MgCl2 concentration and T _m. We created a standard curve (in triplicate) from six serial dilutions of cDNA derived from LBN-F1 RV free wall in concentrations of 1,000, 500, 250, 125, 67, and 34 ng. Starting quantities were determined in two replicates per muscle using iCycler iQ Real-Time Detection System Software v3.0 (Qiagen, Valencia, CA) and averaged. The corresponding values were normalized to the housekeeping gene. We then assessed agreement between microarray analysis and PCR in two ways. For the three muscles each in the physiological*, PO*, and LVAD groups, we were able to perform qPCR on aliquots from the same mRNA isolation used for microarray hybridization. In those cases, we computed correlation coefficients between the sample-by-sample qPCR starting quantities (normalized to TPT1) and the RMA-normalized expression values for single-transcript probe-sets on the Affymetrix GeneChip Rat Genome 230 2.0 array associated with each of the 12 genes selected for PCR. We also conducted PCR on a separate set of unloaded muscles and used group means from the four muscle groups to compare fold changes computed from microarray expression data to fold changes computed from normalized PCR starting quantities: physiological* vs. PO*, physiological* vs. LVAD*, PO* vs. unloaded, and LVAD* vs. unloaded, a total of 48 fold-change values.

RESULTS

Careful implementation of the prestretch and shortening control protocols described above yielded four distinct combinations of mean stretch and cyclic shortening, as confirmed by our marker-tracking protocol ( Fig. 1). The time-averaged stretch values for each group were calculated to be (means ± SD): physiological*, 1.157 ± 0.012; PO*, 1.165 ± 0.015; LVAD*, 1.033 ± 0.015; unloaded, 1.061 ± 0.008 ( Fig. 1 B). Average cyclic shortening values were (means ± SD): physiological*, 0.165 ± 0.009; PO*, 0.036 ± 0.002; LVAD*, 0.164 ± 0.005; unloaded, 0.022 ± 0.003 ( Fig. 1 C). We recorded mechanical force continuously throughout the culture period and normalized this by muscle cross-sectional area to estimate stresses. Time-averaged diastolic stresses (means ± SD) were not statistically different between any groups by one-way ANOVA. Because of their consistently low active contractile stress, mean systolic stress in unloaded muscles (−0.18 ± 1.08 mN/mm ²) was observed to trend lower than the other three groups (physiological*, 2.28 ± 0.64 mN/mm ²; LVAD*, 2.16 ± 0.66 mN/mm ²; PO*, 5.69 ± 4.10 mN/mm ²); however, this did not yield a statistically significant variance between groups by one-way ANOVA.

To identify genes whose expression responded specifically to the amount of shortening, we analyzed the transcriptional effects of reduced cyclic shortening at high stretch (PO*; vs. physiological*; right oval, Fig. 2 A; gene names are listed in Supplemental Table S1) and at low stretch (unloaded vs. LVAD*; left oval, Fig. 2 A). ¹ Per LPE (described in materials and methods), these two comparisons identified 105 and 10,311 differentially expressed probe-sets, respectively. Of these, 77 probe-sets were significantly regulated within both comparisons, representing 36 annotated genes ( Fig. 2 A, oval overlap). Fourteen of these genes were concordantly regulated in the low-L and high-L comparisons ( Table 1). To sort the significance of each gene's overall differential expression, we calculated the geometric mean of each pair of LPE Z scores resulting from both reduced shortening comparisons. A greater mean Z score reflects consistent shortening-dependent differential expression of a gene regardless of the level of stretch.

Fig. 2.

Venn diagrams visualize overlapping differential gene expression along 1 mechanical input axis, independent of the orthogonal mechanical input axis. A reduction in cyclic shortening resulted in significantly altered expression of 105 probe-sets at high mean stretch and 10,311 probe-sets [false discovery rate (FDR) < 0.3 for both] at low mean stretch ( A). A reduction in mean stretch significantly changed expression of 26 probe-sets under high cyclic shortening and 9,712 probe-sets under low cyclic shortening ( B).

Table 1.

Genes with concordant regulation in response to decreased shortening at both physiological (high) and subphysiological (low) mean stretch

Gene Name	Gene Symbol	Expression Change	Mean Z Score
Prostate transmembrane protein, androgen induced 1	Pmepa1	↑	5.74
Nuclear factor, interleukin 3 regulated	Nfil3	↑	4.04
Connective tissue growth factor	Ctgf	↑	3.84
Activating transcription factor 3	Atf3	↑	3.64
Stearoyl-CoA desaturase (delta-9-desaturase)	Scd	↑	3.35
Myeloid/lymphoid or mixed-lineage leukemia (trithorax homolog, Drosophila); translocated to, 3	Mllt3	↑	3.33
Similar to RNA binding motif, single stranded interacting protein 3 isoform 1	LOC680726	↑	2.65
Heat shock protein 90 kDa alpha (cytosolic), class B member 1	Hsp90ab1	↑	2.61
Cytoplasmic polyadenylation element binding protein 2	Cpeb2	↑	2.45
Solute carrier family 41, member 3	Slc41a3	↓	4.95
Interleukin 6	Il6	↓	4.59
Fin bud initiation factor homolog (zebrafish)	Fibin	↓	4.16
Thymic stromal lymphopoietin	Tslp	↓	3.71
Chemokine (C-C motif) ligand 11	Ccl11	↓	3.27

Conversely, to identify genes whose expression varied with stretch, but not shortening, we used LPE to compare the transcriptional effects of reduced stretch under high shortening (LVAD* vs. physiological*) and under low shortening (unloaded vs. PO*). We identified 26 probe-sets that exhibited differential expression with a reduction in stretch at physiological levels of shortening ( Fig. 2 B, top oval; gene names are listed in Supplemental Table S1) and 9,712 probe-sets differentially expressed with reduced stretch at low levels of shortening ( Fig. 2 B, bottom oval). The overlap between these two comparisons comprises a list of 22 probe-sets representing 17 annotated genes, six with concordant changes in expression ( Table 2).

Table 2.

Genes with concordant regulation in response to decreased mean stretch at both physiological (high) and subphysiological (low) amounts of shortening

Gene Name	Gene Symbol	Expression Change	Mean Z Score
Epstein-Barr virus induced 3	Ebi3	↑	3.55
Interleukin 6	Il6	↑	2.98
Similar to RIKEN cDNA 9530077C05	RGD1561444	↓	3.64
Lymphatic vessel endothelial hyaluronan receptor 1	Lyve1	↓	3.50
Thrombospondin 4	Thbs4	↓	3.33
Collagen, type VI, alpha 1	Col6a1	↓	2.34

As described above, PO induced by TAC results in an acute reduction in cyclic shortening of the myocardium ( 10, 29), a reduction we simulated in our low-shortening (PO* and unloaded) muscle groups in the present study ( Fig. 3 A). An LPE comparison of the 7-day TAC and sham wild-type mice revealed 1,183 significantly regulated probe-sets (FDR < 0.1). With these data, we aimed to identify genes that were regulated in both TAC and at least one of our shortening comparisons. To do this, we first eliminated any probe-sets that were discordantly regulated between our two shortening comparisons and filtered out any probe-sets with FDR ≥ 0.1. The overlap between the remaining 7,958 papillary muscle probe-sets and the 1,183 TAC vs. sham probe-sets yielded 169 genes that were upregulated in both pressure overloaded mouse hearts and reduced cyclic shortening rat papillary muscles, and 20 genes that were significantly downregulated in both datasets (Supplemental Table S2). The GO analysis of these 189 overlapping genes revealed 40 significantly enriched GO clusters (cluster enrichment score > 1.3), each of which consisted of several similar GO terms associated with heavily overlapping sets of genes; we listed one representative GO term for each of the top 10 clusters in Table 3.

Fig. 3.

Venn diagrams illustrate overlapping genes regulated by 1 mechanical input signal. Cyclic shortening was reduced in muscle culture ( left arrow, A) and in 7-day transverse aortic constriction (TAC) operated mice compared with sham operated controls [ right arrow, A ( 7)]. Of the 7,958 probe-sets significantly regulated in the muscle comparisons and the 1,183 probe-sets significantly regulated in the 7-day TAC vs. sham hearts (FDR < 0.1 for both), expression of 189 annotated genes was concordantly regulated. Similarly, mean stretch was reduced both in cultured muscles ( bottom arrow, B) and during LVAD treatment of heart failure patients [ top arrow, B; ( 34)]. Of the 7,241 probe-sets regulated in muscles and the 679 probe-sets regulated between pre- and post-LVAD treatment groups (FDR < 0.1 for both), 102 annotated genes were concordantly regulated in both comparisons.

Table 3.

Representative ontology terms among most-enriched functional clusters of genes with altered expression in response to reduced shortening in vitro and TAC in vivo (Fig. 3A)

Gene Ontology Term	Cluster Score	GO Fold Enriched	GO P Value
Extracellular matrix
COL4A2, COL4A1, LGALS1, COL3A1, COL15A1, MGP, NID1, SPARC, TIMP2, ITGB1, TIMP3, TIMP1, EMILIN1, LGALS3BP, BGN, CD44, CTGF, COL1A2, COL6A2, VCAN, COL1A1, LAMC1	8.73	6.63	1.11E-11
Response to mechanical stimulus
CAV1, ACTA1, BTG2, TGFBR2, MGP, RCAN1, RHOC, COL1A1, TIMP3	5.37	9.12	5.77E-06
Blood vessel development
PLAT, CAV1, SOCS3, MYO1E, COL3A1, TGFBR2, GJA1, MYH9, BGN, CD44, CTGF, CASP8, COL1A2, TGM2, COL1A1	4.93	5.09	1.36E-06
Actin cytoskeleton organization
ACTA1, ARF6, MYH7, TMSB10, MYH9, ITGB1, ACTG1, CDC42, CFL1, TMSB4X, RHOC, CAP1, PLS3	4.78	5.79	2.33E-06
Regulation of actin cytoskeleton
ACTB, MYH9, ITGB1, MYL9, ACTG1, CDC42, ITGA9, ARPC2, CFL1, RRAS, TMSB4X, RHOC, MSN	4.52	4.19	4.48E-05
Chordate embryonic development
ADAM10, SOCS3, MYO1E, TGFBR2, GJA1, MYH7, MYH9, ITGB1, VCAM1, ITGA9, DAB2, LY6E, MEOX1, GRN, CASP8, CFL1, COL1A1	4.52	3.53	2.37E-05
Response to calcium ion
ACTB, ACTG1, CAV1, CCND1, MGP, NPPB, RYR2, SPARC, CALM1	4.28	10.97	1.40E-06
Cell migration
PLAT, GJA1, PF4, MYH9, ITGB1, VCAM1, CD44, CTGF, CFL1, FCER1G, HBEGF, LAMC1, MSN, CAP1	4.08	4.22	2.64E-05
Extracellular region
SCPEP1, AEBP1, MASP1, COL3A1, PF4, TIMP2, TIMP3, C1QC, ITGB1, PCOLCE, TIMP1, VCAM1, LGALS3BP, C1QTNF6, CD44, SERPINE2, CTGF, FNDC1, COL6A2, PLAT, CTSZ, COL4A2, COL4A1, LGALS1, COL15A1, MGP, NID1, SPARC, EMILIN1, C1QA, C1QB, BGN, SERPINF1, PPIA, GRN, CAPG, PECAM1, CFL1, COL1A2, HBEGF, NPPB, TMSB4X, VCAN, LAMC1, COL1A1, CALM1	4.03	2.73	1.97E-10
Cell adhesion
AEBP1, COL3A1, COL15A1, NID1, MYH9, MCAM, ITGB1, VCAM1, CD9, LGALS3BP, CD44, CTGF, PECAM1, DSC2, RHOC, VCAN, LAMC1, MSN	3.84	3.03	8.37E-05
30 additional enriched clusters

TAC, transverse aortic constriction; GO, gene ontology.

Mechanical unloading induced by LVAD treatment of a failing human heart reduces the time-averaged ventricular chamber volume, a three-dimensional analog of mean stretch ( 3); we simulated this reduction in the reduced-stretch (LVAD* and unloaded) muscle groups in our study ( Fig. 3 B). Again using LPE with an FDR cutoff of 0.1, we identified 679 probe-sets with significantly altered expression following LVAD treatment of failing hearts. Using the same approach as described above, we generated a list of 7,241 papillary muscle probe-sets regulated by mean stretch (FDR < 0.1) and subsequently identified 69 genes upregulated in response to reduced mean stretch in both cultured rat papillary muscles and LVAD-treated human hearts and 33 genes downregulated in both datasets (Supplemental Table S3). GO analysis identified 17 significantly enriched functional annotation term clusters; representative GO terms for the top six clusters are listed in Table 4.

Table 4.

Representative ontology terms among most-enriched functional clusters of genes with altered expression in response to reduced mean stretch in vitro and LVAD treatment (Fig. 3B)

Gene Ontology Term	Cluster Score	GO Fold Enriched	GO P Value
Response to endogenous stimulus
TXNIP, IRS2, GJA1, SPARC, NR4A3, AQP1, LPIN1, TIMP3, PTEN, C1QB, CDKN1A, BTG2, JAK2, PIK3R1, NPPA	4.70	3.72	3.62E-05
Negative regulation of cell differentiation
ZFP36, HOPX, NFKBIA, WWTR1, C1QC, TOB2, PIK3R1	2.78	4.70	3.53E-03
Negative regulation of apoptosis
CDKN1A, CEBPB, BTG2, RRN3, HIPK2, BNIP3, NFKBIA, PTEN, PIK3R1	2.58	3.84	2.15E-03
Regulation of cell proliferation
TXNIP, NAMPT, IRS2, NFKBIA, GJA1, SPARC, PTEN, VEGFC, CD47, CDKN1A, GLUL, HIPK2, JAK2, NFIB	2.20	2.99	6.31E-04
Extracellular matrix part
COL4A1, COL3A1, COL15A1, ADAMTS1, SPARC, TIMP3	2.08	9.66	3.55E-04
Collagen
C1QB, COL4A1, COL3A1, C1QC	1.79	13.01	3.36E-03
11 additional enriched clusters

LVAD, left ventricle assist device.

To assess how well our microarray analysis agreed with an independent method, we conducted qPCR of a panel of 12 genes of interest to our laboratory, which mapped to 21 single-transcript probe-sets on the Affymetrix GeneChip Rat Genome 230 2.0 array. In nine muscles for which PCR and microarray hybridization were performed on the same sample (all muscles in the physiological*, PO*, and LVAD* groups), we computed correlation coefficients between RMA-normalized microarray expression values and qPCR starting quantities normalized to the housekeeping gene TPT1 ( Fig. 4 A). We identified six probe-sets with >2.5-fold variation between the highest and lowest microarray expression levels measured; all displayed R-squared values of at least 0.5 (range, 0.50–0.87, P < 0.05 for each correlation). By contrast, among the 15 probe-sets with <2.1-fold differences between the highest and lowest expression values, the highest R-squared value was 0.33 (range, 0.01–0.33). We also compared 48 fold-change values (12 genes × 4 intergroup comparisons per gene: physiological* vs. PO*, physiological* vs. LVAD*, PO* vs. unloaded, and LVAD* vs. unloaded) computed from group mean microarray expression and qPCR starting quantities ( Fig. 4 B). We treated a fold-change value of 1.5 on PCR as a “true” increase or decrease in expression and varied the fold-change threshold used to identify an increase or decrease from the microarray data. As expected, the number of group differences identified decreased as the fold-change threshold increased, with fewer changes identified at more stringent thresholds. Between threshold values of 1.6 and 2.3, the fraction of identified events confirmed by PCR was relatively threshold-independent, ranging from 50 to 60%.

Fig. 4.

Comparison of microarray and qPCR analysis for a panel of 12 genes. In 9 muscles for which PCR and microarray hybridization were performed on the same sample (all muscles in the physiological*, PO*, and LVAD* groups), we computed correlation coefficients between robust multichip average-normalized microarray expression values and qPCR starting quantities normalized to the housekeeping gene TPT1. ( A). All 6 probe-sets with >2.5-fold variation between the highest and lowest microarray expression levels displayed R-squared values of at least 0.5 (range, 0.50–0.87, P < 0.05 for each correlation); by contrast, R-squared values for 15 probe-sets with <2.1-fold variation ranged from 0.01–0.33. We also compared 48 fold-change values (12 genes × 4 intergroup comparisons per gene: physiological* vs. PO*, physiological* vs. LVAD*, PO* vs. unloaded, and LVAD* vs. unloaded) computed from group mean microarray expression and qPCR starting quantities ( B). We varied the fold-change threshold used to identify an increase or decrease from the microarray data and treated a fold-change of 1.5 on PCR as a “true” increase or decrease in expression. As expected, the number of group differences identified decreased as the fold-change threshold increased. Between threshold values of 1.6 and 2.3, the fraction of identified events confirmed by PCR was relatively threshold-independent, ranging from 50 to 60%.

DISCUSSION

We previously hypothesized that the amount of cyclic shortening experienced by myocytes in the heart wall regulates myocyte shape, while the average stretch imposed on those myocytes regulates overall myocyte size. To identify genes specifically responsive to changes in cyclic shortening or time-averaged stretch, we cultured rat RV papillary muscles for 12 h at physiological or reduced levels of shortening and stretch using a 2 × 2 factorial design and then used microarrays to assess changes in gene expression. We also examined the relationship between changes in gene expression in our cultured muscles and those reported during two in vivo interventions that alter shortening and stretch: TAC to induce PO hypertrophy and LVAD support of failing human hearts.

Based on our hypothesis that cyclic shortening regulates myocyte shape, we expected that genes specifically responsive to changes in cyclic shortening would be enriched for those encoding sarcomeric and cytoskeletal proteins. Instead, we found a relatively small set of genes that responded similarly to changes in shortening regardless of the level of underlying mean stretch ( Table 1), including connective tissue growth factor (CTGF), heat shock protein 90 kDa β1 (HSP90AB1), and nuclear factor, interleukin 3 regulated (NFIL3, also known as E4BP4). CTGF expression is induced by transforming growth factor (TGF)-β and is strongly upregulated in heart failure ( 23) and in cultured myocytes exposed to prohypertrophic stimuli ( 15, 27). While its potential to cause cardiac fibrosis is still debated, CTGF was recently shown to be an important mediator of cardiac hypertrophy ( 30). The HSP90 family has been shown to be upregulated in cardiac stress and is involved in protein quality control of crucial cardiac proteins such as actin, tubulin, and TGF-β ( 36). NFIL3 was recently shown to be an important transcription factor for cell survival in myocytes ( 35).

Based on the fact that TAC reduces myocyte shortening and dramatically alters myocyte shape, we also expected that genes encoding sarcomeric and cytoskeletal proteins would dominate the list of genes that were both regulated by reduced shortening in our papillary muscles and altered by TAC in vivo. The list of expression changes shared between these two datasets was enriched for cytoskeletal genes as expected ( Table 3) and also included many known key components of the hypertrophic response such as Nppb, Ankrd1, Mybpc2, Myh7, Ctgf, Myl9, Myocd (see Supplemental Table S2 for full list). However, the degree of overlap in genes associated with ECM and ECM-receptor interaction was equally striking. This finding suggests that by culturing intact myocardium, we were able to detect responses of fibroblasts as well as myocytes to the imposed mechanical signals.

One striking result is the disparity between the numbers of differentially expressed genes within each pair of shortening or stretch comparisons. We saw many more gene expression changes in comparisons involving the low shortening, low stretch (unloaded) group than in other comparisons. This result is particularly interesting, given that the mechanics of this group (low stretch, low shortening) were most similar to those of myocytes cultured on a rigid substrate. Indeed we now know that multiple cell types including fibroblasts and cardiomyocytes behave very differently in two-dimensional tissue culture models that lack appropriate ECM contacts ( 1, 2), cellular, or electrical contacts with other cells ( 11, 24) compared with those within three-dimensional tissue constructs and native organ structures. Our results suggest that mechanically perturbing cultured myocytes from an initial low-stretch, low-shortening background may generate a large number of changes in gene expression that would not be induced by a similar perturbation of intact myocardium cultured at a more physiological initial baseline.

Comparisons to PCR performed on a panel of 12 genes suggested some criteria for interpretation of the microarray results. First, we found that all microarray expression levels for probe-sets that showed at least a 2.5-fold difference between the lowest and highest values measured in various samples correlated well with normalized qPCR starting quantities. This provides some estimate of the amount of “noise” inherent in a set of microarray expression values in this study, suggesting that 2.5-fold variations among the values represent actual variation that can be independently confirmed, while <2-fold variations fall below the level of noise in the array and/or PCR analysis, preventing direct confirmation. Second, we found that the fraction of differences between group mean microarray expression levels that was directly confirmed by PCR ranged from 50 to 60%, depending on the fold-change cutoff used to identify a difference. In view of that finding, we expect that roughly half the changes we identified in various comparisons in this study would ultimately be confirmed by PCR, emphasizing the importance of confirming individual gene expression changes of interest before proceeding with mechanistic studies related to any specific finding.

One potential source of error in our study is that we only examined gene expression at one 12 h time point during our muscle experiments. Although our previous observation that myocyte size changes significantly by 36 h in this model suggested that relevant gene expression changes must be present quite early, this time point is certainly extremely early compared with published in vivo experiments, including those we used for comparison here. We chose an early time point to avoid any confounding effects of gradual degeneration of the cultured muscles at longer times, but that choice restricted the pool of shortening- and stretch-responsive genes we could detect to those that respond rapidly to perturbations. Another source of error in cell and organ culture experiments is the choice of media. Some components in our media are known to alter gene expression; insulin in particular is critical for energetics of contraction but also functions as a prohypertrophic factor ( 5, 16). Because all papillary muscle gene expression comparisons were relative to other papillary muscle groups cultured in identical media, the primary impact of this error was likely that we could have missed a shortening- or stretch-dependent response of genes that are also regulated by insulin. Similarly, we chose to compare muscles cultured in the same media under different mechanical conditions to minimize the impact of differences induced by culture, particularly the fact that muscles were cultured in the absence of the normal hormonal milieu present in vivo. When we compared gene expression in the cultured groups to three freshly dissected muscles that were frozen immediately rather than cultured, we found as expected that thousands of genes differed significantly in expression between each culture group and freshly dissected muscles, with the fewest differences in our physiological* group and the most in our unloaded group. The genes that were most different in each cultured group were shared among all groups and were chemokine-related: IL6, Cxcl1, Cxcl2, Ccl2 were in the top 10 for all groups compared with freshly dissected muscles. This finding suggests that mechanical responses identified in the present study may not be pathophysiologically important if they involve genes whose expression are strongly regulated by basal levels of hormones or chemokines in vivo. Our intersection of muscle culture data with published PO and LVAD studies was intended to overcome this potential limitation by identifying genes that are both responsive to the mechanical signals studied here and relevant to in vivo hypertrophy.

We conclude that careful control of mechanical inputs to cultured papillary muscles allows analysis of early gene expression changes in response to specific mechanical inputs thought to play a role in hypertrophy, such as the amount of cyclic shortening and overall mean stretch. The degree of overlap between gene expression changes in papillary muscles cultured with decreased shortening and those reported in PO suggests that decreased shortening may be important for triggering concentric hypertrophy in vivo. ECM genes were also prominent in the group of genes regulated by decreased shortening in vitro and PO in vivo, suggesting that decreased shortening may also be an important stimulus for fibrosis in response to PO.

GRANTS

This work was funded by National Institutes of Health Grants R01 HL-075639 (J. W. Holmes), T32 GM-007267 (E. G. Ames), and T32 GM-008136 (E. G. Ames).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: C.R.H., E.G.A., and J.W.H. conception and design of research; C.R.H. performed experiments; C.R.H. and E.G.A. analyzed data; C.R.H., E.G.A., J.K.L., and J.W.H. interpreted results of experiments; C.R.H. and E.G.A. prepared figures; C.R.H. and E.G.A. drafted manuscript; C.R.H., E.G.A., J.K.L., and J.W.H. edited and revised manuscript; C.R.H., E.G.A., J.K.L., and J.W.H. approved final version of manuscript.