Katherine Hartmann1,2, Michał Seweryn3,4, Samuel K Handelman5,6, Grzegorz A Rempała7,8, Wolfgang Sadee5,6. 1. College of Medicine Center for Pharmacogenomics, The Ohio State University Wexner Medical Center, Biomedical Research Tower, 460 W 12th Avenue, Columbus, OH, USA. katherine.hartmann@osumc.edu. 2. Department of Molecular Virology, Immunology, and Medical Genetics, The Ohio State University, Biomedical Research Tower, 460 W 12th Avenue, Columbus, OH, USA. katherine.hartmann@osumc.edu. 3. Faculty of Mathematics and Computer Science, University of Łodz, Łodz, Poland. mseweryn@math.uni.lodz.pl. 4. Mathematical Biosciences Institute, The Ohio State University, Jennings Hall 3rd Floor, 1735 Neil Avenue, Columbus, OH, USA. mseweryn@math.uni.lodz.pl. 5. College of Medicine Center for Pharmacogenomics, The Ohio State University Wexner Medical Center, Biomedical Research Tower, 460 W 12th Avenue, Columbus, OH, USA. 6. Department of Molecular Virology, Immunology, and Medical Genetics, The Ohio State University, Biomedical Research Tower, 460 W 12th Avenue, Columbus, OH, USA. 7. Division of Biostatistics, College of Public Health, The Ohio State University, 250 Cunz Hall, 1841 Neil Avenue, Columbus, OH, USA. 8. Mathematical Biosciences Institute, The Ohio State University, Jennings Hall 3rd Floor, 1735 Neil Avenue, Columbus, OH, USA.
Abstract
BACKGROUND: Alterations in gene expression are key events in disease etiology and risk. Poor reproducibility in detecting differentially expressed genes across studies suggests individual genes may not be sufficiently informative for complex diseases, such as myocardial infarction (MI). Rather, dysregulation of the 'molecular network' may be critical for pathogenic processes. Such a dynamic network can be built from pairwise non-linear interactions. RESULTS: We investigate non-linear interactions represented in mRNA expression profiles that integrate genetic background and environmental factors. Using logistic regression, we test the association of individual GWAS-based candidate genes and non-linear interaction terms (between these mRNA expression levels) with MI. Based on microarray data in CATHGEN (CATHeterization in GENetics) and FHS (Framingham Heart Study), we find individual genes and pairs of mRNAs, encoded by 41 MI candidate genes, with significant interaction terms in the logistic regression model. Two pairs replicate between CATHGEN and FHS (CNNM2|GUCY1A3 and CNNM2|ZEB2). Analysis of RNAseq data from GTEx (Genotype-Tissue Expression) shows that 20 % of these disease-associated RNA pairs are co-expressed, further prioritizing significant interactions. Because edges in sparse co-expression networks formed solely by the 41 candidate genes are unlikely to represent direct physical interactions, we identify additional RNAs as links between network pairs of candidate genes. This approach reveals additional mRNAs and interaction terms significant in the context of MI, for example, the path CNNM2|ACSL5|SCARF1|GUCY1A3, characterized by the common themes of magnesium and lipid processing. CONCLUSIONS: The results of this study support a role for non-linear interactions between genes in MI and provide a basis for further study of MI systems biology. mRNA expression profiles encoded by a limited number of candidate genes yield sparse networks of MI-relevant interactions that can be expanded to include additional candidates by co-expression analysis. The non-linear interactions observed here inform our understanding of the clinical relevance of gene-gene interactions in the pathophysiology of MI, while providing a new strategy in developing clinical biomarker panels.
BACKGROUND: Alterations in gene expression are key events in disease etiology and risk. Poor reproducibility in detecting differentially expressed genes across studies suggests individual genes may not be sufficiently informative for complex diseases, such as myocardial infarction (MI). Rather, dysregulation of the 'molecular network' may be critical for pathogenic processes. Such a dynamic network can be built from pairwise non-linear interactions. RESULTS: We investigate non-linear interactions represented in mRNA expression profiles that integrate genetic background and environmental factors. Using logistic regression, we test the association of individual GWAS-based candidate genes and non-linear interaction terms (between these mRNA expression levels) with MI. Based on microarray data in CATHGEN (CATHeterization in GENetics) and FHS (Framingham Heart Study), we find individual genes and pairs of mRNAs, encoded by 41 MI candidate genes, with significant interaction terms in the logistic regression model. Two pairs replicate between CATHGEN and FHS (CNNM2|GUCY1A3 and CNNM2|ZEB2). Analysis of RNAseq data from GTEx (Genotype-Tissue Expression) shows that 20 % of these disease-associated RNA pairs are co-expressed, further prioritizing significant interactions. Because edges in sparse co-expression networks formed solely by the 41 candidate genes are unlikely to represent direct physical interactions, we identify additional RNAs as links between network pairs of candidate genes. This approach reveals additional mRNAs and interaction terms significant in the context of MI, for example, the path CNNM2|ACSL5|SCARF1|GUCY1A3, characterized by the common themes of magnesium and lipid processing. CONCLUSIONS: The results of this study support a role for non-linear interactions between genes in MI and provide a basis for further study of MI systems biology. mRNA expression profiles encoded by a limited number of candidate genes yield sparse networks of MI-relevant interactions that can be expanded to include additional candidates by co-expression analysis. The non-linear interactions observed here inform our understanding of the clinical relevance of gene-gene interactions in the pathophysiology of MI, while providing a new strategy in developing clinical biomarker panels.
Authors: Mikko O Hiltunen; Tiina T Tuomisto; Mari Niemi; Jan H Bräsen; Tuomas T Rissanen; Petri Törönen; Ismo Vajanto; Seppo Ylä-Herttuala Journal: Atherosclerosis Date: 2002-11 Impact factor: 5.162
Authors: A M Randi; E Biguzzi; F Falciani; P Merlini; S Blakemore; E Bramucci; S Lucreziotti; M Lennon; E M Faioni; D Ardissino; P M Mannucci Journal: J Thromb Haemost Date: 2003-04 Impact factor: 5.824
Authors: David Seo; Tao Wang; Holly Dressman; Edward E Herderick; Edwin S Iversen; Chunming Dong; Korkut Vata; Carmelo A Milano; Fabio Rigat; Jennifer Pittman; Joseph R Nevins; Mike West; Pascal J Goldschmidt-Clermont Journal: Arterioscler Thromb Vasc Biol Date: 2004-08-05 Impact factor: 8.311
Authors: Adam A Margolin; Ilya Nemenman; Katia Basso; Chris Wiggins; Gustavo Stolovitzky; Riccardo Dalla Favera; Andrea Califano Journal: BMC Bioinformatics Date: 2006-03-20 Impact factor: 3.169
Authors: Katherine Hartmann; Michał Seweryn; Samuel K Handelman; Grzegorz A Rempała; Wolfgang Sadee Journal: BMC Genomics Date: 2016-12-02 Impact factor: 3.969
Authors: Ana Caroline C Sá; Amy Webb; Yan Gong; Caitrin W McDonough; Somnath Datta; Taimour Y Langaee; Stephen T Turner; Amber L Beitelshees; Arlene B Chapman; Eric Boerwinkle; John G Gums; Steven E Scherer; Rhonda M Cooper-DeHoff; Wolfgang Sadee; Julie A Johnson Journal: Sci Rep Date: 2017-11-22 Impact factor: 4.379
Authors: Sebastian Wawrocki; Michal Seweryn; Grzegorz Kielnierowski; Wieslawa Rudnicka; Marcin Wlodarczyk; Magdalena Druszczynska Journal: PLoS One Date: 2019-12-10 Impact factor: 3.240