Novel Algorithms Reveal Streptococcal Transcriptomes and Clues about Undefined Genes

Bacteria–host interactions are dynamic processes, and understanding transcriptional responses that directly or indirectly regulate the expression of genes involved in initial infection stages would illuminate the molecular events that result in host colonization. We used oligonucleotide microarrays to monitor (in vitro) differential gene expression in group A streptococci during pharyngeal cell adherence, the first overt infection stage. We present neighbor clustering, a new computational method for further analyzing bacterial microarray data that combines two informative characteristics of bacterial genes that share common function or regulation: (1) similar gene expression profiles (i.e., co-expression); and (2) physical proximity of genes on the chromosome. This method identifies statistically significant clusters of co-expressed gene neighbors that potentially share common function or regulation by coupling statistically analyzed gene expression profiles with the chromosomal position of genes. We applied this method to our own data and to those of others, and we show that it identified a greater number of differentially expressed genes, facilitating the reconstruction of more multimeric proteins and complete metabolic pathways than would have been possible without its application. We assessed the biological significance of two identified genes by assaying deletion mutants for adherence in vitro and show that neighbor clustering indeed provides biologically relevant data. Neighbor clustering provides a more comprehensive view of the molecular responses of streptococci during pharyngeal cell adherence.: Microarray technology is commonly used to reveal genome-wide transcriptional changes in bacterial pathogens during interactions with the host. Clustering algorithms, which group genes with similar expression patterns, facilitate microarray data organization and are based on assumptions that co-expressed genes share common function or regulation; however, clustering solely by co-expression may not reveal all of the information contained in bacterial array data. We introduce neighbor clustering, a new tool for analyzing bacterial gene expression profiles, which distinguishes itself from other programs by incorporating details unique to the architecture of bacterial chromosomes into the analysis. Neighbor clustering combines two informative characteristics of bacterial genes that share common function or regulation—(1) similar expression profiles and (2) physical proximity on the chromosome—and extracts statistically significant clusters of gene neighbors that are potentially related by function or regulation. We present the analysis of microarray data from group A streptococci during adherence to human pharyngeal cells, the first overt infection step. We show that neighbor clustering identifies more differentially expressed genes than rigorous statistical analyses alone, and can provide functional clues about unknown genes. We extended the analysis to include a previously published streptococcal array study to demonstrate the applicability of the method.