Abstract
Introduction: Sepsis represents a complex inflammatory response to infection. Gene expression studies based on microarrays have shown that this response can affect more than 80% of cellular functions and pathways, in what has been termed a “genomic storm”. For several years, sepsis was regarded as a pro-inflammatory condition, and this concept resulted in several experimental treatment strategies aimed to block inflammation. However, systematic failure of these therapies and recent evidence demonstrating that anti-inflammatory pathways are also activated during sepsis illustrate the complexity and our incomplete knowledge about the pathogenesis of this condition. In the last decade, microarray-based gene expression studies have been used in attempts to improve our understanding about sepsis. Raw data from most of these studies are now collected in public archives, thus offering a unique opportunity to combine the information from different studies by meta-analysis. It has been shown that by analyzing data from multiple experiments, biases and artifacts between datasets can be cancelled out, potentially allowing true relationships to stand out. Accordingly, an increasing number of bioinformatics protocols and guidelines about meta-analysis of gene expression studies have been published in the last years. In the context of sepsis, several high-quality microarray-based gene expression studies are available. However, no systematic meta-analysis of these studies has been performed. In order to identify genes and pathways robustly associated with the pathogenesis of sepsis, we performed a meta-analysis of gene expression studies in human severe sepsis and septic shock.
Material and methods: Microarray data were identified by searching two public databases (Gene Expression Omnibus and Array-Express) using the following search criteria: (“sepsis or “septic shock”) AND (“peripheral blood” or “leukocytes”) AND (“homo sapiens”). Inclusion criteria were: studies in humans with severe sepsis or septic shock; RNA obtained from peripheral blood leukocytes; availability of raw data; and matched healthy controls from the same study. To improve consistency, only studies using similar platforms were compared. We used the R/BioConductor environment to preprocess the datasets using the Robust Multi-array Average algorithm (RMA) implemented in the ‘oligo’ package and to perform meta-analysis through the ‘RankProd’ package implementation. This is a non-parametric statistical method that utilizes ranks of the log-ratio statistics for all genes across different studies to generate a list of differentially expressed (DE) genes between two conditions, and considered superior to alternative methodologies. For this study, we selected genes with fold-change of expression above 2 and false discovery rate below 0.01, calculated based on 10,000 permutations. Gene set analysis was initially performed using WebGestalt and confirmed in similar tools (KEGG, Pathway Commons, WikiPathways). Only pathways identified by more than one tool were considered.
Results: Forty-five studies were identified, of which five fulfilled inclusion criteria. Our meta-analysis included data from 259 patients and 132 controls. Out of 22,216 probesets, we observed 352 as candidates for DE, 215 of which were up-regulated and 137 down-regulated. Top 5 up-regulated genes were CD177, MMP8, HP, ARG1 and ANXA3. Top 5 down-regulated genes were FCER1A, YME1L1, TRDV3, LRRN3 and MYBL1. The gene ontology term associated with the set of DE genes in both analysis with higher statistical significance was "immune response” (adjP=2.85e-27), and the most significant pathways identified were “Hematopoietic cell lineage” (adjP=8.69e-13), “TCR signaling pathway” (adjP=3.04e-10) and “immune system” (adjP=1.08e-19).
Discussion and conclusion: The combined analysis of data generated by high-throughput experiments is an attractive and validated strategy to improve the sensitivity and specificity of genome-wide expression data. This meta-analysis provides a comprehensive list of genes, pathways and expression signatures associated with severe sepsis and septic shock, confirming several results from individual studies. In addition, our meta-analysis potentially provides new biological insights about sepsis, by listing a comprehensive list of new candidate genes with robust associations with this condition.
Disclosures
No relevant conflicts of interest to declare.
Classical and Bayesian random-effects meta-analysis models with sample quality weights in gene expression studies