When Duke's Chris Woods was traveling the world in the 1990s in pursuit of his early interest in infectious diseases, he began to realize something: like politics, all fevers are local. "What causes disease in Sri Lanka during the dry season is different from the wet season and different from either season in Nepal," Woods says. "It's certainly different from diseases in Durham, North Carolina, or Singapore."
During a stint as a member of the Centers for Disease Control's Bacterial Special Pathogens Group, it also became abundantly clear to Woods that expectations about what type of illness a person in a particular place at a particular time had were quite often wrong upon closer inspection. "In Tanzania, we were told that everyone had malaria. More often it was tuberculosis related to HIV infection. We were told everyone in Nepal had malaria, but what most had was typhoid fever but not classic typhoid fever. In Sri Lanka, we were told everyone had dengue. The first time, few had dengue; it was rickettsial disease or leptospirosis. When we went back later, it really was a dengue outbreak. Time, season, geography all affect what someone with [essentially the same set of] disease symptoms actually has."
As a result, the diagnoses being made, preventive measures taken, and treatments being prescribed were often incorrect or at best questionable. For Woods back then, it was crystal clear, "we needed to improve our pathogen-based diagnostics."
Signatures of Infection
Now more than a decade later, Woods is an instrumental member of a multidisciplinary team at the Duke Center for Applied Genomics and Precision Medicine that has been working for several years to devise faster, more reliable ways to diagnose infectious diseases and to make timely decisions, for example, about whether a person would be helped by antibiotic drugs or not. The Duke team hasn't gone about it in the way Woods originally anticipated, by finding better ways to ferret out and identify the pathogen responsible for a person's illness. Pathogens, after all, are small and tough to find, particularly when you don't know what you're looking for.
Instead of looking to pathogens for answers, Woods and his Duke colleagues have looked to their infected hosts, in search of telltale signatures of infection in the bloodstream with the potential to distinguish a viral infection from a bacterial one. In some cases, their tests can even show that a person is becoming ill before any symptoms appear. It's a practical approach to infectious disease diagnosis that's been called a paradigm shift for the field.
It all began years ago, in the relatively early days of the genome era, when Woods and Geoff Ginsburg, an expert in genomic medicine and director of the Duke Center began tossing around ideas for collaboration. Woods had done some preliminary work aimed at using gene expression data to determine which people with the nonspecific symptoms of fever actually had a severe, or even fatal, infection and which had something more akin to a run-of-the-mill cold.
Ginsburg was similarly focused on genomic signatures to classify and make predictions about cardiovascular disease and other common, complex diseases. That's when he learned of an effort at DARPA called the "Predicting Health and Disease" program from Larry Carin, an electrical engineer at Duke's Pratt School of Engineering.
DARPA's interest was in protecting soldiers and avoiding epidemics during military operations. The idea was that "if you have a battalion of 100 guys and one gets sick as you are ready to ship off, should you go forward or should you hold back?" Woods has said. "You want to know if others are incubating a virus."
Carin had done work with DARPA before, developing analytical methods for remote sensing of landmines and other applications. He suspected similar analytical approaches might be applied to data encapsulating the human immune system's genomic response to viral infection.
"This was an opportunity to apply those genomic methods in the completely new area of infectious diseases," Ginsburg said. "There was some evidence it could be done, but no one had taken it to the next level."
Leading the Way
The Duke team put together a seemingly bold proposal, which grew into a $30 million DARPA-funded effort. They conducted a series of "human challenge studies" with live viruses. This meant that they infected healthy volunteers with one of several common viruses—rhinovirus, respiratory syncytial virus (RSV) and influenza—collect blood, urine, breath and nasal wash samples from each exposed individual as they developed the illness or not over the course of two weeks. Those samples were then analyzed for changes in gene activity over time.
Their ultimate goal was to find a signature in that expression data that could predict who would become sick even before the first cough or sneeze. And, it worked. The Duke team found they could make a genome-based prediction about the onset of respiratory symptoms with a very high rate of success. Those signatures developed via the challenge study were later tested in a "real world" setting -- among Duke students during the cold and flu season.
The Duke team's effort to improve infectious disease diagnosis continues today although their overarching aim has shifted. "What's driving us is no longer protection of a war fighter from imminent infection, but the goal of reducing inappropriate use of antimicrobial agents," Woods said. "We've heard a lot about the danger of emerging resistance around the globe. This is a very real threat to all of us."
With support from the NIH's Antimicrobial Resistance Leadership Group, the team, now including Ephraim Tsalik, assistant professor of medicine at Duke and emergency medicine provider at the Durham VA Medical Center, recently introduced their latest genomic test to determine whether a respiratory illness is caused by infection with a virus or bacteria. In an observational study conducted in patients seeking treatment for respiratory symptoms in the emergency room at Duke, they showed their test is accurate 87 percent of the time in distinguishing between flu viruses, rhinovirus, strep bacteria and other common infections. It can also tell when a person's symptoms aren't linked to any infection at all.
With these findings, Duke researchers are a significant step closer to developing a rapid blood test that could be used in clinics to guide appropriate treatment of patients with a fever and general respiratory symptoms. That's key because respiratory infections are among the most common reasons people seek medical attention.
"We use a lot of information to make a diagnosis, but there's not an efficient or highly accurate way to determine whether the infection is bacterial or viral," Tsalik said when news of the new test first broke. "About three-fourths of patients end up on antibiotics to treat a bacterial infection despite the fact that the majority have viral infections."
More precise ways of distinguishing among infections could ultimately lead to more precise treatments of viruses too, as available antiviral treatments expand. With the technology employed in the latest paper, the test takes several hours to complete. But, Woods says, it now looks promising that their signatures could be loaded into a handheld device, with results delivered in less than an hour.
"We thought we could get down to a two to four hour turnaround," Woods said. "Now, we've been catapulted into the potential for a meaningful diagnostic that could deliver a result within an hour. That's incredibly exciting."