Gene Expression Signature Identifies Flu, S aureus

Ricki Lewis, PhD

January 09, 2013

"Signatures" derived from host gene expression patterns can distinguish individuals infected with influenza or Staphylococcus aureus from noninfected individuals, according to 2 reports published online January 9, 2013, in PLOS One.

A rapid diagnostic test that provides results before symptoms peak could limit antibiotic use, identify patients who could benefit from antivirals within the window of efficacy, and limit epidemics by identifying infected individuals earlier in the course of infection.

The 2 studies used a Bayesian sparse factor model to combine differences in the expression of many genes into factors that serve as independent variables in a binary regression model to distinguish infected from noninfected individuals. This approach aggregates genes that participate in the same pathways and accounts for the fact that degree of expression of a gene does not necessarily correlate to the strength of its effect.

In a study on S aureus identification, Sun Hee Ahn, PhD, from the Division of Infectious Diseases and International Health, Department of Medicine, Duke University, Durham, North Carolina, and colleagues established the proof-of-concept for a "molecular classifier" for S aureus in inbred mice and validated it in outbred mice. The researchers used microarrays to analyze 45,101 probes, corresponding to 39,000 mRNAs and 34,000 genes, in 187 inbred mice. The researchers filtered the results for highly and variably expressed genes, eliminating samples with many outliers. Thirty factors distinguished among S aureus infection, Escherichia coli infection, and no infection, and also distinguished methicillin-resistant S aureus from methicillin-sensitive infection.

In the human portion of the study, the investigators enrolled patients hospitalized for community-acquired pneumonia and bloodstream infection (BSI) caused by culture-confirmed S aureus (n = 32) or E coli (n = 19). Blood samples from 43 healthy volunteers provided controls.

A human genome microarray identified 9109 genes after removing uniformly expressed genes and excluding samples with many outliers. Factor analysis represented the human data as 79 factors.

The human S aureus classifier was able to distinguish S aureus from E coli BSI in 82% (42/51) of the cases. When the researchers modified the classifier to discriminate only the 2 types of bacteria, the model had a sensitivity of 62.5% (20/32 S aureus BSI correctly classified) and a specificity of 94.7% (18/19 E coli BSI correctly classified).

The researchers subsequently validated murine and human classifiers amended to remove E coli sequences in an additional human cohort of 46 children with S aureus infection and 10 healthy control patients.

"Terrific Potential"

A diagnostic that works early in pathogenesis could guide clinical decision-making, according to David Liebers, MD, chief medical officer at Ellis Medicine in Schenectady, New York, who was not involved in the studies. "A genomic signature will be useful to separate patients who present with inflammation, fever, and chills. Is it bacterial, or something else? Sometimes we treat for bacterial infection when the patient doesn't have it. If we can distinguish bacterial infection from other conditions, we can focus on work-ups and not overuse antibiotics," Dr. Liebers told Medscape Medical News.

In the study on influenza detection, Christopher W. Woods, MD, MPH, from the Institute for Genome Sciences and Policy and the Division of Infectious Diseases, Duke University Medical Center, and Durham Veteran’s Affairs Medical Center, North Carolina, and colleagues tested a whole-blood mRNA expression classifier they developed for human response to rhinovirus, respiratory syncytial virus, and H3N2 influenza. A diagnostic tool that works before symptom onset would be "an indispensible tool for guiding individual treatment decisions when antiviral supplies may be limited," the researchers write. The transcribed genes included in the classifier are part of the innate and adaptive immune responses to viral infection.

The researchers inoculated 24 healthy volunteers with H1N1 influenza A and 17 with H3N2, then assayed the transcriptome in peripheral blood every 8 hours for a week. Eighteen (44%) of the volunteers became ill. Symptoms for H1N1 began at 24 to 108 hours (median, 72 hours), and for H3N2 at 24 to 84 hours (median, 48 hours). Peak illness was on average 102.7 hours after inoculation for H1N1 and 90.6 hours for H3N2.

The gene expression signature had a detection rate of 94%, appears as early as 29 hours after exposure, and is maximally accurate before the worst of the symptoms are discernible.

To validate the experimentally tested signature on a real-world cohort, the investigators used stored peripheral blood from 36 patients who had visited the emergency department at Duke University Hospital in 2009 who had H1N1. The signature distinguished them from 45 healthy control patients with 92% (33/36) accuracy and correctly identified 93% (42/25) of the control patients.

The influenza gene expression signature illuminates a window during infection that has not been well-studied because symptoms are so mild. The approach can also provide information on the dynamics of an epidemic and reveal emergence of new viral strains.

"The technique has terrific potential to look at a population of patients exposed to certain pathogens and identify those at risk to develop and ultimately spread the associated disease," Dr. Liebers said.

A limitation of the influenza study is that young, healthy patients do not reflect the general population. For both studies, a limitation of factor analysis is that it does not represent responses of the entire genome.

Dr. Ahn has disclosed no relevant financial relationships. Two coauthors have served as consultants for bioMerieux Inc. Three coauthors received research funding from Novartis Vaccines and Diagnostics Inc. One coauthor has served as a consultant to Becton, Dickinson and Company, and Gilead Sciences and has received research or grant support from Cubist Pharmaceuticals, Novartis Pharmaceuticals, and Roche Molecular Diagnostics. One coauthor is supported by a Veterans Affairs Career Development Award. One coauthor has received grant or research support from Astellas Pharma US, Merck, Theravance, Cerexa, Pfizer, Novartis, MedImmune, Advance Liquid Logic, and the National Institutes of Health; has served as a consultant for Achaogen, Merck, Novartis, Pfizer, NovaDigm Therapeutics, The Medicines Company, Durata Therapeutics, Galderma, and Biosynexus; and is on the Advisory Committee for Cubist Pharmaceuticals. One coauthor is a consultant for Covance. One coauthor receives research support from Agennix AG, Alere Corporation, and National Institutes of Health and has previously served as a consultant for Agennix AG, Eisai Pharmaceuticals, Idaho Technology, AstraZeneca, Masimo, and Sangard. Dr. Woods and 5 coauthors have filed a patent on the viral signature for influenza, and Dr. Woods has consulted for Becton, Dickinson and Company and Gilead Sciences and received funding from Novartis Pharmaceuticals, Cubist Pharmaceuticals, and Roche Molecular Diagnostics. Two coauthors work for Retroscreen Virology. One coauthor has received funding from Astellas Pharma US, Merck, Theravance, Cerexa, Pfizer, Novartis, MedImmune, and Advance Liquid Logic and has consulted for Achaogen, Merck, Novartis, Pfizer, NovaDigm Therapeutics, the Medicines Company, Durata Therapeutics, Galderma, and Biosynexus. One coauthor consults for Covance. One coauthor receives support from Agennix AG and Alere Corp. The other authors and Dr. Liebers have disclosed no relevant financial relationships.

PLOS One. Published online January 9, 2013.