Skip to main content
Premium Trial:

Request an Annual Quote

Study Describes Gene Expression Model to Distinguish Viral, Bacterial Infections in Clinical Setting

NEW YORK (GenomeWeb) – A Duke University-led research team has developed a microarray-based approach to determine whether a patient's respiratory disease is due to a bacterial or viral infection, both, or neither.

By drawing on a cohort of patients who came to the emergency room complaining of a respiratory infection, Duke's Geoffrey Ginsburg and his team developed a gene expression-based model that uses host gene expression profiles to predict the source of patients' infections, as they reported in Science Translational Medicine this week.

Being able to distinguish between viral and bacterial disease may enable antibiotics to be more judiciously used, the researchers noted. Most respiratory tract infections are viral in nature, but nearly three-quarters of patients are prescribed antibiotics anyway. By bringing a test that captures host response to infection into the clinic, Ginsburg said that antibiotics, and eventually antivirals, could be better targeted.

"We know that humans have evolved to have these really robust and high-fidelity mechanisms to recognize certain classes of pathogens," Ginsburg told GenomeWeb. "We've been able to capture that using a genomic platform in a way that allows us to make those distinctions."

Ginsburg's team previously developed other gene-expression based assays to identify whether patients' infections were viral or bacterial in nature.

For instance, as GenomeWeb reported in 2013, the researchers uncovered gene expression shifts that occurred as patients developed flu symptoms after being infected with the H1N1 or H3N2 influenza viruses. In a separate study published at the same time, they found that Staphylococcus aureus infections also generated telltale changes in gene expression in the host.

The team soon followed up with an RT-PCR-based assay that could gauge, again based on a host gene expression signature, whether patients' respiratory infections were viral or not.

This new study, though, tried to focus on what questions a clinician would have when faced with a patient with a respiratory infection, Ginsburg said.

"A patient is sitting in front of you who has a sore throat, congestion, maybe fever … and [may] have a viral infection, a bacterial infection, or maybe they are not infected at all — [maybe they] have an allergy that is masquerading with these symptoms," he said. "So what we wanted to do was develop a way to distinguish between those really important clinical scenarios."

To develop their model, Ginsburg and his colleagues drew upon such a patient population, and used a cohort of 70 people with bacterial infections, 115 with viral infections, and 88 with noninfectious illness who presented to the emergency room, as well as 44 healthy controls. They collected peripheral whole blood from the cohort for microarray-based gene expression analysis.

The resulting model, which uses a sparse logistic regression approach, incorporates three classifiers: a bacterial classifier that examines 71 probe signatures; a viral classifier with 33 probe signatures; and a noninfectious disease classifier of 25 probe signatures.

The bacterial classifier includes genes involved in processes like cell cycle regulation, cell growth, and differentiation, while the viral classifier includes ones involved in interferon response, T cell signaling, and RNA processing. Ginsburg noted that some of the genes in this viral classifier differ from those included in their previous efforts.

The researchers validated their classifiers in silico based on gene expression data from 328 people from five datasets. There, the overall classification accuracy was 96 percent.

They also reported that their classifier fared better than the bacterial infection biomarker procalcitonin and three published gene expression classifiers at distinguishing bacterial and viral infections.

"Even compared with our previous work this is going to me a really highly relevant model and very clinically attractive and applicable," Ginsburg said.

He added that their motivation has been to develop a tool that can be used clinically. Ginsburg noted that they've shown in previous work that models that are array-based or RNA-sequencing-based models like this can be modified for use in clinical labs.

He and his team are in process of migrating this model and have done some work toward using RT-PCR methods, a tool that's more commonly found in clinical labs than are arrays or other expression assays, he pointed out.

However, there's still a long turnaround time to contend with, even with an RT-PCR platform. It can take 10 hours to 12 hours to measure a patient's gene expression profile from a blood sample, he said, and to make clinical decisions based on this data, a shorter timeframe would be needed.

Because of that, Ginsburg said he and his team are working with others at Duke as well as with commercial partners to explore alternative methods such as cartridge-based microfluidic platforms with optical array reader.

"Our goal is — and I think it's doable, not tomorrow, but in the foreseeable future —that we could have something that could have an hour or two turnaround time," he said.

That way, Ginsburg said, patients could walk into a doctor's office and walk out of it knowing whether or not they'd need a prescription.