Physiological stimuli such as chronic exercise lead to compensatory growth and remodeling of the heart associated with preserved or improved cardiac function. Recently, class IA phosphoinositide 3-kinase (PI3K) and Akt1 have emerged as important regulators of physiological adaptation [32
] but the broader signaling cascades associated with physiological LVH remain poorly understood. In this study we show that network analysis has the potential to infer genome-wide biological mechanisms related to physiological LVH phenotype. Importantly, we report on the network topology and functional properties of the physiological LVH networks, the first such analysis in a mammalian cardiovascular system.
Gene expression profiles were used to identify conserved gene co-expression patterns in PI3K, Akt1, and Swimming models of physiological LVH and to obtain a global overview of biological functions involved in physiological cardiac remodeling. Previous reports have explored gene co-expression networks derived from heterogeneous microarray platforms [14
] and confirm that observing a conserved gene co-expression suggests a biological relevance [9
]. The consensus gene co-expression model, referred to as the Conserved network, consisted of 2128 genes and 4144 links (Additional files 1
). It was confirmed to be scale-free, highly structured, and non-random, suggesting the presence of a small number of critical hub genes that may be biologically relevant. Additionally, the Conserved network had only a trivial intersection with the Normal interactome (88 genes, 57 links), suggesting that our consensus model may present a reliable physiological LVH signature. Topological features were consistent with the general behavior of biological networks [36
] and topologies detected in protein-protein interaction collections such as STRING [37
]. At PCC≥0.70, 31% of all genes in the Conserved network were identified in the KEGG pathways database. This coverage increased exponentially with PCC threshold, approaching 80% at PCC = 0.88 (Figure ). These results are comparable to previous studies of co-expression networks [20
] and suggest that an increase in PCC stringency produces a marked positive effect on network precision.
Due to a large number of co-expression links (3 million), it is possible that some of these links are artifacts or byproducts of systematic error. Thus, evaluation of conserved co-expression links across three physiological LVH networks has a number of strengths compared to conventional statistical approaches. First, reproducible co-expressions are less likely to be false-positives and may reflect biologically relevant links, thus presenting a reliable interactome for further experimental validation [9
]. For example, in a recent meta-analysis of >300 tissue samples of gastric cancer, this hypothesis helped to identify a functional link between prognostic marker PLA2G2A and the EphB2 receptor [34
]. Second, network intersections account for putative platform- or experiment-dependent variability (e.g. number of transcripts) between multiple microarray datasets [14
]. Third, due to the heterogeneous (molecular and physiological) nature of physiological LVH models, conserved co-expressions provide an overview of common regulatory mechanisms.
These assumptions were confirmed using automated PubMed queries, whereby each gene in the Conserved network was searched in the context of 'hypertrophy', 'heart', or 'heart failure'. Indeed, 933 out of 2128 (44%) genes in the Conserved network had at least one abstract per search term while 50 of those have at least one hundred abstracts for all terms, suggesting that the Conserved network provides an acceptable coverage of current molecular knowledge of cardiac biology (Additional file 4
The Conserved network may be used to describe the regulatory mechanisms underpinning the cardiac remodeling response to physiological stress. 'Oxidative phosphorylation' was noted as one of the most abundant KEGG pathways (n
= 27 genes). The most over-represented members of this pathway were genes encoding subunits of mitochondrial cytochrome c oxidase (COX) (n
= 6 genes). COX is localized to the inner membrane of mitochondria and is the last component of respiratory chain. To sustain respiration, this enzyme catalyzes the transfer of electrons from cytochrome c to molecular oxygen and facilitates the aerobic production of ATP by ATPsynthase (n
= 2 genes in the Conserved network) [39
]. To maintain efficient cardiac contractility under increased energetic demand, the regulation of COX function must be preserved. In post-myocardial infarction this mechanism is disrupted by the generation of reactive oxygen species (ROS) such as superoxide, leading to a marked loss of COX activity [40
]. These results are consistent with the well-established concept that suppression of mitochondrial energy metabolism can lead to depression of cardiac contractile function [41
The Conserved network was useful in the delineation of the cardiac response to increased protein synthesis and energy deprivation through activation of autophagy. This is a highly conserved cellular pro-survival mechanism for bulk lysosomal degradation of cytoplasmic components that mobilizes energy resources in response to starvation or hypoxia [42
]. Autophagy also has a protein quality-control housekeeping function. The Conserved network identified two key genes related to autophagic processes, Atg5 (Autophagy-related protein 5) and Becn1 (Beclin-1). Both of these genes were topologically central to the Conserved network (betweenness centrality of 53356.0 and 12262.3 respectively), implicating them in critical mediation of network information flow. Recent studies in mice with temporally controlled cardiac-specific deficiency of Atg5 demonstrated that Atg5 was essential for normal physiological growth and function of the heart. However, Atg5-deficient animals developed contractile dysfunction and heart failure accompanied by increased levels of ubiquitinated proteins. Furthermore, Atg5-deficient hearts showed disorganized sarcomere structure and mitochondrial misalignment and aggregation [43
]. These abnormalities were suggested, at least in part, to be due to loss of the protein quality control function of autophagy. Becn1 is part of a PI3K complex that plays an important role during the initiation of autophagosome formation [42
]. Interestingly, mice with heterozygous disruption of Becn1 (Becn1+/-
) exhibited reduced levels of autophagy during reperfusion but had decreased apoptosis and reduced infarct size compared to wild type mice [45
], suggesting that in this case autophagy was detrimental. However, Becn1 is an important point of crosstalk with apoptotic pathways through its interaction with anti-apoptotic proteins such as Bcl-2 [46
]. Disruption of Becn1 could therefore have pro- or anti-survival effects [42
]. Of note, in the Conserved network, Becn1 localized to the same MCL cluster as Bcl-2, which is known to inhibit Becn1-depended autophagy [46
]. Thus, in physiological LVH, autophagy compatible with cell survival, rather than cell death, may be regulated by coordinated changes in Atg5, Becn1 and Bcl-2. Indeed, autophagy- and proteolysis-related genes localized to the same cluster as genes involved in cell cycle regulation, providing further support for this hypothesis.
To explore if key regulatory mechanisms may be encoded by topologically significant nodes, the Conserved network was studied using concepts of betweenness centrality and node degree. These approaches are known to detect essential hubs in interaction networks [47
] and previous studies have demonstrated that betweenness is a good indicator of biological essentiality [48
]. Interestingly, when the top 200 hub genes were systematically removed from the Conserved network, average network betweenness remained mostly constant and high, while characteristic path length increased dramatically, to a threshold beyond which the network collapsed. This may suggest a presence of a large number of well-connected genes that preserve network information flow, possibly an indicator of maintained functional cardiac integrity during physiological remodeling. Additionally, topologically-central genes (core genes) localized to KEGG pathways including 'Oxidative phosphorylation' (n
= 19 genes), 'MAPK signaling pathway' (n
= 18 genes), and 'Focal adhesion' (n
= 17 genes) (Figure ).
Several genes associated with the mammalian target of rapamycin (mTOR) pathway (Cab39, Hif1a, Tsc2) were also identified. The mTOR pathway controls changes in cell size following activation of the PI3K/Akt system. Akt phosphorylates the Tsc2 gene product tuberin, and thereby reduces its ability to stimulate GTP hydrolysis on the Ras-like G protein Rheb, leading to increased protein synthesis via ribosome biogenesis - a key feature of cardiac hypertrophy - and cell growth [49
]. Recently, inhibition of the mTOR pathway by rapamycin was demonstrated to alleviate load-induced cardiac hypertrophy in mice, making it a potential therapeutic target [50
]. Indeed, Tsc2 had a very large betweenness centrality value (174802.9, top 1%), confirming that it is one of the key constituents of the Conserved network. Core genes present in the 'MAPK signaling pathway' included Map4k3, Map3k7, Rap1a, Mapkapk2, Cacng2, and Ppm1b. Of these, Ppm1b (protein phosphatase 1B) had the greatest node degree (32) and betweenness centrality (73822.0) values, supporting its biological importance. These findings are reinforced by demonstration of direct inhibition of Map3k7 by Ppm1b [51
], thus providing further evidence that Map3k7 activity is reduced in physiological hypertrophy protecting the heart from interstitial fibrosis, severe myocardial dysfunction, and apoptosis [52
Similarly, the core Conserved network suggests that the genes involved in KEGG 'Calcium signaling pathway' may be involved in physiological LVH. There were 13 genes (e.g. Ppp3ca, Egfr, Vdac3, Slc25a4, Tnnc1) allocated to 'Calcium signaling pathway', of which Ppp3ca (calcineurin A alpha) had the largest betweenness centrality value (71043.2). Ppp3ca has been shown to be a key regulator of cardiac hypertrophy through activation of the transcription factor NFAT (nuclear factor of activated T-cells) which promotes the expression of pro-hypertrophic genes in concert with other transcription factors such as GATA4 and MEF2 [53
]. It can also inhibit Map3k7 signaling [54
]. The Conserved network also provides further evidence that calcineurin activity is highly regulated under physiological conditions by elucidation of the Rcn2 gene, which is known to inhibit calcineurin signaling [55
The use of MCL in the core network (Figure ) identified enriched clusters of genes participating in similar biological pathways. For example, cluster 1 was enriched for KEGG pathway 'Apoptosis' (n
= 5 genes: Birc2, Irak1, Pik3ca, Prkaca, Ppp3ca). Birc2 (baculoviral IAP repeat-containing 2, betweenness = 3316.0) encodes a protein that inhibits apoptosis by binding to tumor necrosis factor receptor-associated factors TRAF1 and TRAF2. Although previously not reported in the mammalian heart, Birc2 was confirmed as a critical regulator of vascular integrity and endothelial cell survival in zebrafish [56
]. Null mutants for Birc2 showed severe hemorrhage and vascular regression due to endothelial cell integrity defects and activation of Caspase-8-dependent apoptosis program. Coordinated regulation of angiogenesis is essential for preserved cardiac contractile function [57
] and our results provide further molecular evidence for angiogenic gene programs in physiological LVH that merits further exploration.