SOM portrayal of the peripheral blood transcriptome

Background: The transcriptome of healthy blood offers an option to characterize the physiological state of an individual with diagnostic impact. As a prerequisite such application requires the understanding of the variability of the expression landscape of healthy individual’s as a function of factors such as age, gender, and aspects of the human constitution and lifestyle. Previous studies mostly were limited to only small population sizes and thus lack representativeness in many respects. Methods and data: The present thesis provides an extensive and detailed study on the gene expression landscape of peripheral blood of healthy individuals based on transcriptome data of 3,388 individuals screened in the population study LIFE between 2011 and 2014. For analysis we applied a neural network technique using self-organizing maps (SOM). Our home-made R-program ‘oposSOM’ enables to reduce the dimension of expression data from tens of thousands of genes to a few thousand ‘meta-genes’ and generates portraits of transcriptional activity that allowed us the sample-to-sample comparison of expression patterns. Results: We disentangled the expression patterns of the portraits into 15 well separated modules of co-expressed genes. Their activation patterns allowed us to aggregate the participants into three types (‘1’, ‘M’, ‘2’). Enrichment techniques extracts the functional context of the modules and revealed that corresponding cellular processes are selectively activated and de-activated in a type-specific fashion: For example, ‘Inflammation’ and ‘Blood coagulation’ are activated in type ‘1’ and ‘Metabolic activity’ and ‘Translation’ in Type ‘2’ samples. We map clinical parameters (complete blood count, lifestyle status, medication and the anthropometric body and BMI type) into the expression landscape in order to study the association between them and the transcriptome landscape. We found that red blood cell components, drug consumption of several drug classes affecting the cardio-vascular system or blood forming organs, tobacco and alcohol consumption, and the BMI types of obese participants highly correlate with meta-genes in the type ‘1’ area, whereas type ‘2’ meta-genes maximal correlate with the lymphocyte count and pre-obese characteristics of the participants. For blood count parameters and medication we found a gender-specific bias. Conclusion: Our analysis provides a comprehensive description of the human blood transcriptome in terms of a series of characteristic expression modules and of health-relevant factors. It provides a healthy reference system for blood transcriptome profiling studies in healthcare to extract potential associations between emerging diseases and the gene expression patterns in clinical research.