NEW YORK (GenomeWeb) – Two new studies provide important data in support of efforts to create RNA-based tests that can distinguish whether febrile infants have bacterial infections.
The papers, published in the Journal of the American Medical Association today,reflect ongoing efforts by two teams — one US-led and one UK-led — to identify and refine an immune-related gene expression signature that can pick out feverish infants with bacterial infections who are at risk of death without antibiotic treatment versus those with viral fevers who are likely to improve on their own.
The two separate efforts were both spurred by growing evidence that specific pathogens or classes of pathogens can be identified by a pattern of human host genes activated during the immune system's response to infection.
For example, in 2013, researchers from Duke developed and validated an RT-PCR assay for a host gene expression signature to determine whether a respiratory infection is caused by a virus.
Eventually, both infant fever teams also hope to be able to translate their discoveries into a PCR- or other nucleic acid amplification-based test — potentially one that could be performed at the point of care in hospitals or other medical centers without the need for complicated instrumentation or even possibly by parents at home to help keep infants out of the emergency room altogether.
In a commentary appearing alongside the studies today, Howard Bauchner, editor in chief of JAMA, wrote that they "represent an important advance" in demonstrating that genomic analysis might be able to successfully differentiate children with bacterial fevers where existing laboratory tests have largely failed.
According to Bauchner, the clinical questions raised by a fever in an infant are different depending on the child's age. "Among very young infants, the goal of a laboratory test is to rule out [bacterial infections] with a very high degree of certainty," he wrote. But among older children, "the goal is to differentiate bacterial from viral infection, so the clinician does not have to prescribe antibiotics, an increasingly important consideration because of concerns about antibiotic resistance."
These clinical questions were central in the design of the two studies. In one, researchers from the National Institutes of Health-funded Pediatric Emergency Care Applied Research Network (PECARN) enrolled infants two months old or younger from 22 different hospital emergency rooms.
Overall, the group has recruited more than 1,800 infants, but for its preliminary study it only tested samples from 279 randomly selected babies — 89 with bacterial infections and 190 without bacterial infections.
"These very young infants are the hardest ones to evaluate," Octavio Ramillo, the study's senior author, told GenomeWeb. "They come to the ER, and even the most skilled physicians can't determine if they have a bacterial infection or not, because they come in so early in the infection."
According to Ramillo, absent a definitive diagnosis, these babies receive a lot of lab work, often including spinal fluid analysis, and are admitted for days-long stays until clinicians can rule out a bacterial cause.
"We wanted to see if this [host immune response gene expression] approach could even work because these babies have a very immature immune system and they are often coming into the ER so early in an infection … so we were very happy to see that the signals were very significant," he said.
Using Illumina microarrays, the investigators identified 66 genes that distinguished infants with bacterial infections from those without, and ten genes that distinguished infants who specifically had bacteremia (the presence of bacteria in the blood) from those without.
According to the authors, the sensitivity and specificity of the overall signature was 87 percent and 89 percent respectively. For the subset of babies with bacteremia, sensitivity was 94 percent and specificity was 95.
Ramillo — along with co-PIs Prashant Mahajan at Wayne State University and Nathan Kuppermann at the University of California Davis — received $5.8 million from the National Institute of Child Health and Human Development last fall to develop a faster and more precise method to rule out bacteria as the cause of fever in young infants.
The first step under this new funding has been to further validate and refine their early findings to reduce the number of genes in the signature, and to try to get it up above 95 percent sensitivity.
"We already have data that we can do better," Ramillo said.
Next, the group plans to work to translate the approach to a platform, like PCR, or potentially even simpler methods, that can be run in a short time in a point-of-care or bedside setting.
In the second study in JAMA today, researchers were able to narrow down to a signature of only two genes, which remained highly predictive of bacterial versus viral infections in infants — albeit in a group with a higher median age.
In that study, a team led by researchers at Imperial College London collected a discovery and a validation group of febrile children presenting to hospitals in the UK, Spain, the Netherlands, and the US.
The discovery group included 52 infants with a definite bacterial infection and 92 with a known viral infection. For 96 others, the diagnosis was undetermined.
Michael Levin, the study's senior author, told GenomeWeb the he and his team had previously identified a gene expression signature that could discern cases of tuberculosis from other infections that might appear clinically similar.
With a mind to the clinical difficulties of assessing infants with high fevers, the team then turned to this area with the same basic approach. Using microarrays to analyze samples from a discovery group of 250 infants, the researchers were able to identify 38 gene transcripts that could discriminate the bacterial group from the viral group
Then, using additional statistical analyses, they managed to narrow this down to just two genes — FAM89A and IFI44L.
"With the first analysis, we're [essentially] picking out the big fish and throwing out the small fish," Levin said. To narrow this down, the team used a novel approach developed by IPL's Lachlan Coin, one of the study's coauthors, which culls genes whose predictive value overlaps with one another out of the signature.
According to the authors, when the resulting two-transcript signature was applied to the independent validation cohort it had a sensitivity of 100 percent and a specificity of 96 percent. Among the children in the indeterminate group, the signature classified 46 percent as having bacterial infection.
In additional validation experiments with patients with meningococcal disease and inflammatory diseases like juvenile arthritis and lupus, the signature remained highly predictive, the investigators wrote.
According to Levin, being able to narrow the field to just two transcripts opens up opportunities for even simpler and more accessible testing platforms as he and his colleagues move forward with a similar ultimate goal as the NIH-funded PECARN team — continuing to validate and translate their findings to the clinic.
"We were very surprised and excited to get it down to just two because that makes it a much simpler problem to turn into a test," he said. Possibilities the group has in mind include more traditional PCR, or newer approaches that forego fluorescent detection for electrical readouts that could mean a more mobile test platform.
Like Ramillo, Levin said that he and his colleague's main goal as clinical researchers is to make sure their signature is as accurate and sensitive as possible, and collect the necessary validation data to make this clear to the physician community.
"There are so many brilliant people in biotech, and now that [this is published] we are hoping the community will pick this up," he explained.
Moving forward, Levin said he and colleagues are working under an EU grant to confirm and validate their findings. As part of this, they have already collected RNA from about 1,000 children from multiple studies in Europe and have plans to recruit an additional 3,500.