Genome-wide Co-expression Network Research Articles

A genome-wide co-expression network analysis revealed ZmNRAMP6-mediated regulatory pathway involved in maize tolerance to lead stress.

A metal transporter ZmNRAMP6 was identified by using a trait-associated co-expression network analysis at a genome-wide level. ZmNRAMP6 confers maize sensitivity to Pb by accumulating it to maize shoots. ZmNRAMP6 knockout detains Pb in roots, activates antioxidant enzymes, and improves Pb tolerance. Lead (Pb) is one of the most toxic heavy metal pollutants, which can penetrate plant cells via root absorption and thus cause irreversible damages to the human body through the food chain. To identify the key gene responsible for Pb tolerance in maize, we performed a trait-associated co-expression network analysis at a genome-wide level, using two maize lines with contrasting Pb tolerances. Finally, ZmNRAMP6 that encodes a metal transporter was identified as the key gene among the Pb tolerance-associated co-expression module. Heterologous expression of ZmNRAMP6 in yeast verified its role in Pb transport. Combined Arabidopsis overexpression and maize mutant analysis suggested that ZmNRAMP6 conferred plant sensitivity to Pb stress by mediating Pb distribution across the roots and shoots. Knockout of ZmNRAMP6 caused Pb retention in the roots and activation of the antioxidant enzyme system, resulting in an increased Pb tolerance in maize. ZmNRAMP6 was likely to transport Pb from the roots to shoots and environment. An integration of yeast one-hybrid and dual-luciferase reporter assay uncovered that ZmNRAMP6 was negatively regulated by a known Pb tolerance-related transcript factor ZmbZIP54. Collectively, knockout of ZmNRAMP6 will aid in the bioremediation of contaminated soil and food safety guarantee of forage and grain corn.

The transcriptional regulatory network of hormones and genes under salt stress in tomato plants (Solanum lycopersicum L.).

Salt stress has become one of the main limiting factors affecting the normal growth and development of tomatoes as well as fruit quality and yields. To further reveal the regulatory relationships between tomato hormones under salt stress, the interaction between hormones and TF and the genome-wide gene interaction network were analyzed and constructed. After salt treatment, the levels of ABA, SA, and JA were significantly increased, the levels of GA were decreased, and IAA and tZ showed a trend of first increasing and then decreasing. The expression patterns of hormone biosynthesis and signal transduction related genes were analyzed based on RNA-seq analysis, the co-expression network of hormones and genome-wide co-expression networks were constructed using weighted gene co-expression network analysis (WGCNA). The expression patterns of specific transcription factors under salt stress were also systematically analyzed and identified 20 hormone-related candidate genes associated with salt stress. In conclusion, we first revealed the relationship between hormones and genes in tomatoes under salt stress based on hormone and transcriptome expression profiles and constructed a gene regulatory network. A transcriptional regulation model of tomato consisted of six types of hormones was also proposed. Our study provided valuable insights into the molecular mechanisms regulating salt tolerance in tomatoes.

Comparative genomics analysis of drought response between obligate CAM and C3 photosynthesis plants

Crassulacean acid metabolism (CAM) plants exhibit elevated drought and heat tolerance compared to C3 and C4 plants through an inverted pattern of day/night stomatal closure and opening for CO2 assimilation. However, the molecular responses to water-deficit conditions remain unclear in obligate CAM species. In this study, we presented genome-wide transcription sequencing analysis using leaf samples of an obligate CAM species Kalanchoë fedtschenkoi under moderate and severe drought treatments at two-time points of dawn (2-h before the start of light period) and dusk (2-h before the dark period). Differentially expressed genes were identified in response to environmental drought stress and a whole genome wide co-expression network was created as well. We found that the expression of CAM-related genes was not regulated by drought stimuli in K. fedtschenkoi. Our comparative analysis revealed that CAM species (K. fedtschenkoi) and C3 species (Arabidopsis thaliana, Populus deltoides ‘WV94’) share some common transcriptional changes in genes involved in multiple biological processes in response to drought stress, including ABA signaling and biosynthesis of secondary metabolites.

Beyond the Biosynthetic Gene Cluster Paradigm: Genome-Wide Coexpression Networks Connect Clustered and Unclustered Transcription Factors to Secondary Metabolic Pathways.

ABSTRACTFungal secondary metabolites are widely used as therapeutics and are vital components of drug discovery programs. A major challenge hindering discovery of novel secondary metabolites is that the underlying pathways involved in their biosynthesis are transcriptionally silent under typical laboratory growth conditions, making it difficult to identify the transcriptional networks that they are embedded in. Furthermore, while the genes participating in secondary metabolic pathways are typically found in contiguous clusters on the genome, known as biosynthetic gene clusters (BGCs), this is not always the case, especially for global and pathway-specific regulators of pathways’ activities. To address these challenges, we used 283 genome-wide gene expression data sets of the ascomycete cell factory Aspergillus niger generated during growth under 155 different conditions to construct two gene coexpression networks based on Spearman’s correlation coefficients (SCCs) and on mutual rank-transformed Pearson’s correlation coefficients (MR-PCCs). By mining these networks, we predicted six transcription factors, named MjkA to MjkF, to regulate secondary metabolism in A. niger. Overexpression of each transcription factor using the Tet-On cassette modulated the production of multiple secondary metabolites. We found that the SCC and MR-PCC approaches complemented each other, enabling the delineation of putative global (SCC) and pathway-specific (MR-PCC) transcription factors. These results highlight the potential of coexpression network approaches to identify and activate fungal secondary metabolic pathways and their products. More broadly, we argue that drug discovery programs in fungi should move beyond the BGC paradigm and focus on understanding the global regulatory networks in which secondary metabolic pathways are embedded.IMPORTANCE There is an urgent need for novel bioactive molecules in both agriculture and medicine. The genomes of fungi are thought to contain vast numbers of metabolic pathways involved in the biosynthesis of secondary metabolites with diverse bioactivities. Because these metabolites are biosynthesized only under specific conditions, the vast majority of the fungal pharmacopeia awaits discovery. To discover the genetic networks that regulate the activity of secondary metabolites, we examined the genome-wide profiles of gene activity of the cell factory Aspergillus niger across hundreds of conditions. By constructing global networks that link genes with similar activities across conditions, we identified six putative global and pathway-specific regulators of secondary metabolite biosynthesis. Our study shows that elucidating the behavior of the genetic networks of fungi under diverse conditions harbors enormous promise for understanding fungal secondary metabolism, which ultimately may lead to novel drug candidates.

A comprehensive study on genome-wide coexpression network of KHDRBS1/Sam68 reveals its cancer and patient-specific association

Human KHDRBS1/Sam68 is an oncogenic splicing factor involved in signal transduction and pre-mRNA splicing. We explored the molecular mechanism of KHDRBS1 to be a prognostic marker in four different cancers. Within specific cancer, including kidney renal papillary cell carcinoma (KIRP), lung adenocarcinoma (LUAD), acute myeloid leukemia (LAML), and ovarian cancer (OV), KHDRBS1 expression is heterogeneous and patient specific. In KIRP and LUAD, higher expression of KHDRBS1 affects the patient survival, but not in LAML and OV. Genome-wide coexpression analysis reveals genes and transcripts which are coexpressed with KHDRBS1 in KIRP and LUAD, form the functional modules which are majorly involved in cancer-specific events. However, in case of LAML and OV, such modules are absent. Irrespective of the higher expression of KHDRBS1, the significant divergence of its biological roles and prognostic value is due to its cancer-specific interaction partners and correlation networks. We conclude that rewiring of KHDRBS1 interactions in cancer is directly associated with patient prognosis.

Genome-scale integrated analysis to identify prospective molecular mechanisms and therapeutic targets in isocitrate dehydrogenase 2 R140Q-mutated acute myeloid leukemia.

The aim of the present study was to identify potential molecular mechanisms and therapeutic targets in regards to isocitrate dehydrogenase 2 (IDH2) R140Q-mutated acute myeloid leukemia (AML). An RNA sequencing dataset of IDH2 wild-type and R140Q-mutated adult de novo AML bone marrow samples was obtained from The Cancer Genome Atlas (TCGA) database. The edgeR package was used to screen for the differentially expressed genes (DEGs), and the potential molecular mechanisms and therapeutic targets were identified using Database for Annotation, Visualization, and Integrated Discovery (DAVID) v6.8, Biological Networks Gene Ontology tool, Connectivity Map (CMap), Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) and GeneMANIA. A total of 230 DEGs were identified between the bone marrow tissues of IDH2 R140Q-mutated and wild-type AML patients, of which 31 were significantly associated with overall survival (OS). Functional assessment of DEGs showed significant enrichment in multiple biological processes, including angiogenesis and cell differentiation. STRING and GeneMANIA were used to identify the hub genes of these DEGs. CMap analysis identified 13 potential small-molecule drugs against IDH2 R140Q-mutated adult de novo AML. Genome-wide co-expression network analysis identified several IDH2 R140Q co-expressed genes, of which 56 were significantly associated with AML OS. The difference in IDH2 mRNA expression levels and OS between the IDH2 R140Q-mutated and wild-type AML were not statistically significant in our cohort. In conclusion, we identified several co-expressing genes and potential molecular mechanisms that are instrumental in IDH2 R140Q-mutated adult de novo AML, along with 13 candidate targeted therapeutic drugs.

PdumBase: a transcriptome database and research tool for Platynereis dumerilii and early development of other metazoans

BackgroundThe marine polychaete annelid Platynereis dumerilii has recently emerged as a prominent organism for the study of development, evolution, stem cells, regeneration, marine ecology, chronobiology and neurobiology within metazoans. Its phylogenetic position within the spiralian/ lophotrochozoan clade, the comparatively high conservation of ancestral features in the Platynereis genome, and experimental access to any stage within its life cycle, make Platynereis an important model for elucidating the complex regulatory and functional molecular mechanisms governing early development, later organogenesis, and various features of its larval and adult life. High resolution RNA-seq gene expression data obtained from specific developmental stages can be used to dissect early developmental mechanisms. However, the potential for discovery of these mechanisms relies on tools to search, retrieve, and compare genome-wide information within Platynereis, and across other metazoan taxa.ResultsTo facilitate exploration and discovery by the broader scientific community, we have developed a web-based, searchable online research tool, PdumBase, featuring the first comprehensive transcriptome database for Platynereis dumerilii during early stages of development (2 h ~ 14 h). Our database also includes additional stages over the P. dumerilii life cycle and provides access to the expression data of 17,213 genes (31,806 transcripts) along with annotation information sourced from Swiss-Prot, Gene Ontology, KEGG pathways, Pfam domains, TmHMM, SingleP, and EggNOG orthology. Expression data for each gene includes the stage, the normalized FPKM, the raw read counts, and information that can be leveraged for statistical analyses of differential gene expression and the construction of genome-wide co-expression networks. In addition, PdumBase offers early stage transcriptome expression data from five further species as a valuable resource for investigators interested in comparing early development in different organisms. To understand conservation of Platynereis gene models and to validate gene annotation, most Platynereis gene models include a comprehensive phylogenetic analysis across 18 species representing diverse metazoan taxa.ConclusionsPdumBase represents the first online resource for the early developmental transcriptome of Platynereis dumerilii. It serves as a research platform for discovery and exploration of gene expression during early stages, throughout the Platynereis life cycle, and enables comparison to other model organisms. PdumBase is freely available at http://pdumbase.gdcb.iastate.edu.

Genome-wide Analyses on Single Disease Samples for Potential Biomarkers and Biological Features of Molecular Subtypes: A Case Study in Gastric Cancer.

Purpose: Based on the previous 3 well-defined subtypes of gastric adenocarcinoma (invasive, proliferative and metabolic), we aimed to find potential biomarkers and biological features of each subtype.Methods: The genome-wide co-expression network of each subtype of gastric cancer was firstly constructed. Then, the functional modules in each genome-wide co-expression network were divided. Next, the key genes were screened from each functional module. Finally, the enrichment analysis was performed on the key genes to mine the biological features of each subtype. Comparative analysis between each pair of subtypes was performed to find the common and unique features among different subtypes.Results: A total of 207 key genes were identified in invasive, 215 key genes in proliferative, and 204 key genes in metabolic subtypes. Most key genes in each subtype were unique and new findings compared with that of the existing related researches. The GO and KEGG enrichment analyses for the key genes of each subtype revealed important biological features of each subtype.Conclusions: For a subtype, most identified key genes and important biological features were unique, which means that the key genes can be used as the potential biomarker of a subtype, and each subtype of gastric cancer might have different occurrence and development mechanisms. Thus, different diagnosis and therapy methods should be applied to the invasive, proliferative and metabolic subtypes of gastric cancer.

Transcriptome-wide co-expression analysis identifies LRRC2 as a novel mediator of mitochondrial and cardiac function.

Mitochondrial dysfunction contributes to myriad monogenic and complex pathologies. To understand the underlying mechanisms, it is essential to define the full complement of proteins that modulate mitochondrial function. To identify such proteins, we performed a meta-analysis of publicly available gene expression data. Gene co-expression analysis of a large and heterogeneous compendium of microarray data nominated a sub-population of transcripts that whilst highly correlated with known mitochondrial protein-encoding transcripts (MPETs), are not themselves recognized as generating proteins either localized to the mitochondrion or pertinent to functions therein. To focus the analysis on a medically-important condition with a strong yet incompletely understood mitochondrial component, candidates were cross-referenced with an MPET-enriched module independently generated via genome-wide co-expression network analysis of a human heart failure gene expression dataset. The strongest uncharacterized candidate in the analysis was Leucine Rich Repeat Containing 2 (LRRC2). LRRC2 was found to be localized to the mitochondria in human cells and transcriptionally-regulated by the mitochondrial master regulator Pgc-1α. We report that Lrrc2 transcript abundance correlates with that of β-MHC, a canonical marker of cardiac hypertrophy in humans and experimentally demonstrated an elevation in Lrrc2 transcript in in vitro and in vivo rodent models of cardiac hypertrophy as well as in patients with dilated cardiomyopathy. RNAi-mediated Lrrc2 knockdown in a rat-derived cardiomyocyte cell line resulted in enhanced expression of canonical hypertrophic biomarkers as well as increased mitochondrial mass in the context of increased Pgc-1α expression. In conclusion, our meta-analysis represents a simple yet powerful springboard for the nomination of putative mitochondrially-pertinent proteins relevant to cardiac function and enabled the identification of LRRC2 as a novel mitochondrially-relevant protein and regulator of the hypertrophic response.

Identification of Candidate Targeted Genes in Molecular Subtypes of Gastric Cancer

Because of high heterogeneity, a further classification should be made for diagnosis and treatment in gastric cancer. Biomarkers selected in subtypes are important for precision medicine. Based on gene expression level, we constructed genome-wide co-expression networks for invasive, proliferative and metabolic subtype in gastric cancer respectively. The hierarchical clustering was used to get sub-networks, and hub gene sets of subtypes were got by analysis in sub-networks. Unique differential expression genes as candidate targeted genes in subtype were gained by a comparative analysis between subtypes. These genes may be helpful for improving diagnosis and therapy methods and developing new drug in gastric cancer.

Use of transcriptome sequencing to understand the pistillate flowering in hickory (Carya cathayensis Sarg.).

BackgroundDifferent from herbaceous plants, the woody plants undergo a long-period vegetative stage to achieve floral transition. They then turn into seasonal plants, flowering annually. In this study, a preliminary model of gene regulations for seasonal pistillate flowering in hickory (Carya cathayensis) was proposed. The genome-wide dynamic transcriptome was characterized via the joint-approach of RNA sequencing and microarray analysis.ResultsDifferential transcript abundance analysis uncovered the dynamic transcript abundance patterns of flowering correlated genes and their major functions based on Gene Ontology (GO) analysis. To explore pistillate flowering mechanism in hickory, a comprehensive flowering gene regulatory network based on Arabidopsis thaliana was constructed by additional literature mining. A total of 114 putative flowering or floral genes including 31 with differential transcript abundance were identified in hickory. The locations, functions and dynamic transcript abundances were analyzed in the gene regulatory networks. A genome-wide co-expression network for the putative flowering or floral genes shows three flowering regulatory modules corresponding to response to light abiotic stimulus, cold stress, and reproductive development process, respectively. Totally 27 potential flowering or floral genes were recruited which are meaningful to understand the hickory specific seasonal flowering mechanism better.ConclusionsFlowering event of pistillate flower bud in hickory is triggered by several pathways synchronously including the photoperiod, autonomous, vernalization, gibberellin, and sucrose pathway. Totally 27 potential flowering or floral genes were recruited from the genome-wide co-expression network function module analysis. Moreover, the analysis provides a potential FLC-like gene based vernalization pathway and an 'AC’ model for pistillate flower development in hickory. This work provides an available framework for pistillate flower development in hickory, which is significant for insight into regulation of flowering and floral development of woody plants.

Endonuclease G: The Link Between Mitochondria and Cardiac Hypertrophy?

### Endonuclease G Is a Novel Determinant of Cardiac Hypertrophy and Mitochondrial Function McDermott-Roe et al Nature . 2011;478:114–118. A recent study proposes novel functions of the long known mitochondrial endonuclease, endonuclease G (ENDOG).1 ENDOG was identified via quantitative trait loci analysis as a mediator of blood-pressure–independent cardiac hypertrophy and identified as a novel component in the mitochondrial biogenesis signaling network. Blood pressure-independent cardiac hypertrophy is commonly observed in subjects with obesity and Type 2 diabetes mellitus.2 It is likely that multiple predisposing factors and genetic modifiers interact in a complex fashion, to induce left ventricular (LV) hypertrophy. In the study by McDermott-Roe et al, the authors analyzed quantitative trait loci for blood pressure-independent cardiac mass in rats using an intercross of spontaneously hypertensive rats (SHR) and Brown Norway rats and identified ENDOG as a novel determinant of LV mass and cardiac function.1 Expression of ENDOG was reduced in SHR rats due to a frame-shift insertion in exon 1 of ENDOG that correlated with LV hypertrophy and impaired cardiac function. In subsequent studies, the authors demonstrated that shRNA-mediated knockdown of ENDOG in isolated neonatal rat cardiomyocytes resulted in increased cardiomyocyte size and that cardiomyocyte area was increased in mice lacking ENDOG. Thus, ENDOG may be a novel determinant of blood pressure-independent LV hypertrophy in rodents. To determine the potential relevance to cardiac hypertrophy in humans, a genome-wide coexpression network analysis in a large human cardiac expression data set was performed. ENDOG was identified as participating in a network that was highly enriched for mitochondrial genes and oxidative metabolism processes. Subsequently, the authors demonstrated that ENDOG is a direct target of ERRα and PGC-1α, master regulators of mitochondrial biogenesis. Overexpression of ENDOG increased mitochondrial mass in HEK293 and H9C2 cardiomyocytes. ENDOG directly binds to mitochondrial DNA suggesting a role in the regulation of mitochondrial gene transcription. Functionally, ENDOG−/− mice showed impaired mitochondrial respiration rates and increased …

The functional network of the Arabidopsis plastoglobule proteome based on quantitative proteomics and genome-wide coexpression analysis.

Plastoglobules (PGs) in chloroplasts are thylakoid-associated monolayer lipoprotein particles containing prenyl and neutral lipids and several dozen proteins mostly with unknown functions. An integrated view of the role of the PG is lacking. Here, we better define the PG proteome and provide a conceptual framework for further studies. The PG proteome from Arabidopsis (Arabidopsis thaliana) leaf chloroplasts was determined by mass spectrometry of isolated PGs and quantitative comparison with the proteomes of unfractionated leaves, thylakoids, and stroma. Scanning electron microscopy showed the purity and size distribution of the isolated PGs. Compared with previous PG proteome analyses, we excluded several proteins and identified six new PG proteins, including an M48 metallopeptidase and two Absence of bc1 complex (ABC1) atypical kinases, confirmed by immunoblotting. This refined PG proteome consisted of 30 proteins, including six ABC1 kinases and seven fibrillins together comprising more than 70% of the PG protein mass. Other fibrillins were located predominantly in the stroma or thylakoid and not in PGs; we discovered that this partitioning can be predicted by their isoelectric point and hydrophobicity. A genome-wide coexpression network for the PG genes was then constructed from mRNA expression data. This revealed a modular network with four distinct modules that each contained at least one ABC1K and/or fibrillin gene. Each module showed clear enrichment in specific functions, including chlorophyll degradation/senescence, isoprenoid biosynthesis, plastid proteolysis, and redox regulators and phosphoregulators of electron flow. We propose a new testable model for the PGs, in which sets of genes are associated with specific PG functions.

Signaling pathway-based identification of extensive prognostic gene signatures for lung adenocarcinoma

Tumor recurrence is the major cause of death in lung cancer treatment. To date, there is no clinically applied gene expression-based model to predict the risk for tumor recurrence in non-small cell lung cancer (NSCLC). We sought to embed crosstalk with major signaling pathways into biomarker identification. Three approaches were used to identify prognostic gene signatures from 442 lung adenocarcinoma samples. Candidate genes co-expressed with 6 or 7 major NSCLC signaling hallmarks were identified from genome-wide coexpression networks specifically associated with different prognostic groups. From these candidate genes, the first approach selected genes significantly associated with disease-specific survival using univariate Cox model. The second approach used random forests to refine the gene signatures; and the third approach used Relief algorithm to form the final gene sets. A total of 21 gene signatures were identified using these three approaches. These gene signatures generated significant prognostic stratifications (log-rank P<0.05 in Kaplan–Meier analyses; hazard ratio >1, P<0.05) in all tumors, stage I only, and in stage I patients not receiving chemotherapy in all training and test sets. In multivariate analyses with age, gender, race, smoking history, cancer stage, and tumor differentiation, a 10-gene signature had a hazard ratio of 3.23 (95% CI: [1.48, 7.06]), which was a more significant prognostic factor than other clinical factors, except cancer stage (III vs. I; with no significant difference). All identified 21 gene signatures outperformed other lung cancer signatures evaluated in the Director's Challenge Study. This study is an important step toward personalized prognosis of tumor recurrence and patient selection for adjuvant chemotherapy, with significant impact on down-stream clinical applications.

A novel network model identified a 13-gene lung cancer prognostic signature

This study presents a novel network methodology to identify prognostic gene signatures. Implication networks based on prediction logic are used to construct genome-wide coexpression networks for different disease states. From the differential components associated with specific disease states, candidate genes that are co-expressed with major disease signal hallmarks are selected. From these candidate genes, top genes that are the most predictive of clinical outcome are identified using univariate Cox model and Relief algorithm. Using this approach, a 13-gene lung cancer prognosis signature was identified, which generated significant prognostic stratifications (log-rank P < 0.05) in Director's Challenge Study (n = 442).

A Novel Network Model for Molecular Prognosis.

Network-based genome-wide association studies (NWAS) utilize the molecular interactions between genes and functional pathways in biomarker identification. This study presents a novel network-based methodology for identifying prognostic gene signatures to predict cancer recurrence. The methodology contains the following steps: 1) Constructing genome-wide coexpression networks for different disease states (metastatic vs. non-metastatic). Prediction logic is used to induct valid implication relations between each pair of gene expression profiles in terms of formal logic rules. 2) Identifying differential components associated with specific disease states from the genome-wide coexpression networks. 3) Dissecting network modules that are tightly connected with major disease signal hallmarks from the disease specific differential components. 4) Identifying most significant genes/probes associated with clinical outcome from the pathway connected network modules. Using this methodology, a 14-gene prognostic signature was identified for accurate patient stratification in early stage lung cancer.

Measuring similarity between gene expression profiles: a Bayesian approach

BackgroundGrouping genes into clusters on the basis of similarity between their expression profiles has been the main approach to predict functional modules, from which important inference or further investigation decision could be made. While the univocal determination of similarity metric is important, current practices are normally involved with Euclidean distance and Pearson correlation, of which assumptions are not likely the case for high-throughput microarray data.ResultsWe advocate the use of a novel metric - BayesGen - to measure similarity between gene expression profiles, and demonstrate its performance on two important applications: constructing genome-wide co-expression network, and clustering cancer human tissues into subtypes. BayesGen is formulated as the evidence ratio between two alternative hypotheses about the generating mechanism of a given pair of genes, and incorporates as prior knowledge the global characteristics of the whole dataset. Through the joint modelling of expected intensity levels and noise variances, it addresses the inherent nonlinearity and the association of noise levels across different microarray value ranges. The full Bayesian formulation also facilitates the possibility of meta-analysis.ConclusionBayesGen allows more effective extraction of similarity information between genes from microarray expression data, which has significant effect on various inference tasks. It also provides a robust choice for other object-feature data, as illustrated through the results of the test on synthetic data.

Modular gene expression in Poplar: a multilayer network approach

By applying a multilayer network approach to an extensive set of Poplar microarray data, a genome-wide coexpression network has been detected and explored. Multilayer networks were generated from minimum spanning trees (MSTs) using Kruskal's algorithm from random jack-knife resamplings of half of the full data set. The final network is obtained from the union of all the generated MSTs. The gene expression correlations display a highly clustered topology, which is more pronounced when introducing links appearing in relatively few of the generated MSTs. The network also reveals a modular architecture, reflecting functional groups with relatively frequent gene-to-gene communication. Furthermore, the observed modular structure overlaps with different gene activities in different tissues, and closely related tissues show similar over- and/or under-expression patterns at the modular scale. It is shown that including links that appear in a few of the generated MSTs increases the information quality of the network. In other words, a link may be 'weak' because it reflects rare signaling events rather than merely a signal weakened by noise. The method allows, from comparisons of random 'null networks', tuning to maximize the information obtainable.